When I started working with sales at a company that sold business cards, we were expanding the portfolio from cards (and the accessories for cards) into promotional products, and one of them was custom notebooks. Selling notebooks required a different muscle from sales, and at first I gave them no clear reason or motivation to build it.

I explained, with great enthusiasm, why customers were going to love the notebooks, which said nothing about how the reps were going to make their quotas with a product that was harder to sell and more niche than the cards they could sell in their sleep. I shared the full specs and a beautiful comparison against similar products on the market. The reps immediately hyper-focused on where we fell short technically compared to competitors, when actually, our differentiation was not technical at all: it was part emotional, and part ways of working and the consistent brand rigor we gave the customer. Nevertheless, the sales reps determined it was a hard-to-sell product and focusing on it would distract them from closing what was earning them commission today.

I was selling to the end user and skipped selling to sales first. Nobody had answered “what’s in it for sales?” for the sales team. In my naivety I was offended: “Don’t you want the company to succeed? This is a strategic priority!” Product marketers keep discovering that “strategic priority” often doesn’t appear anywhere in a commission plan, so it was on me to help align sales incentives to the business strategy.

Writing messaging customers love is the part every product marketer already knows how to do, but we rarely pause to understand sales incentives.

Sales teams are high-performance athletes who care about one metric, quota attainment. Handing them a beautiful comparison chart has never made anyone run faster. If reps see product marketing as overhead or marketing fluff, they ignore it, politely, and go back to the race.

The way out is to start one step earlier, by making sure that what the company needs sales to sell also pays the people selling it, and only then to prove that your work increases their win rate.

1. Follow the money: align incentives before messaging

The commission plan tells you, line by line, what a rep will spend time selling. Most of us have read more competitor pricing pages than our own company’s commission plan (mainly because it’s a very secret document guarded by dragons in a dungeon), which is a strange way to prepare for a job that depends on sales. With the notebooks, the first fix was in what reps were rewarded for. I worked with sales leadership to bring on special incentives tied only to selling notebooks, rewarding not just reaching quota but going over it through notebooks, which finally put my strategic priority somewhere a rep would notice it. Many of those incentives were fun activities and internal competitions that eased the reps in until they had developed the muscle, and every round gave me the feedback I needed to understand how they actually sold. I also rethought the pricing structure and the minimum order, so a rep could present the trade-offs to the customer, let them pick one and place an order. After that, the reasons customers would love notebooks finally had reps willing to say them.

The principle: Make their job easier and their commission checks larger, and you become their most valuable ally.

The action: Before you write any messaging, ask sales leadership to walk you through the commission plan and check whether what you need reps to sell helps them hit their number. Then stop asking “How can I enable you?” and ask this instead: “What is the number one objection you are hearing right now that is killing your deals?” When the incentives line up and you build content specifically to kill that objection, you are helping them get paid, which is the only kind of enablement a rep needs.

2. Demonstrate utility: the battle card pilot

Recently, a marketing leader told me his organization had produced six different, inconsistent approaches to battle cards in a single year. They were hiring a new product marketing leader, and what worried him most was the seventh version, arriving with the new leader like a sledgehammer to change the whole sales organization at once. Six approaches in a year looks like a mess, and it is also a massive opportunity. The temptation is to answer it with exactly that seventh version, with a new template, a new folder structure, a training calendar and a name for the program, which is usually how an organization ends up with six approaches in the first place.

The format was rarely the problem. The same competitor can be valued for completely different reasons by different types of customers, and a generic competitor card is written for a customer who doesn’t exist, the one who cares about every argument equally, which leaves the rep to figure out, on their own, and in the middle of a deal which half of the card applies.

The principle: Trust is built through one tangible win.

The action: Don’t create a new format or another version. Pick one high-priority competitor and think strategically about what the same battle card needs to say for each use case, each vertical and each ICP. Then make it very easy to find and read, so a rep can pull the card for that competitor, that use case and that ICP without browsing a catalog and deciding what to use and what to skip. That sorting is the labor of product marketing. When the reps see that using your card leads to a win, they will stop asking for battle cards and start asking for “your” battle cards.

3. Respect the domain: run the field intelligence loop

Sales reps are on the front lines, and they often feel that marketing is out of touch with the reality of the deal. Some of that reputation we earned honestly, presenting market insights at the sales kickoff to people who had spent the whole year talking to the market.

The principle: Respect their field experience while owning the narrative synthesis across inputs.

The action: Position yourself as the synthesizer of their intelligence. Instead of telling them how to sell, say this: “I’ve been listening to the call recordings, and I’m hearing this specific objection across the board. I’ve drafted a response to that objection. Can you try it on your next call and tell me if it works?” This turns them into collaborators instead of recipients of your instructions.

4. Bring on the data: win/loss as the arbiter

Subjective arguments with sales reps are a losing battle, partly because every lost deal comes with at least two explanations, the rep’s and ours (ours usually involves accusing the rep of not using the messaging). The call recordings have a third explanation, which tends to be a lot more useful and less accusatory.

The principle: Data is the neutral ground where product marketing and sales can meet.

The action: Use the call recordings to run a win/loss analysis. If you can show them that “we are losing 40% of deals because we aren’t articulating why we are purpose-built for their industry,” you are highlighting a shared problem that you are going to help them solve, and nobody’s pitch is under scrutiny.

5. Incrementalism: find a sales champion

Trying to change the behavior of an entire sales organization at once triggers exactly the sledgehammer resistance that that marketing leader feared. We tend to plan the rollout first, with the training, the certification and the announcement to the whole team. That’s a lot to digest and to trust with no proof. So find an ally. Salespeople listen to salespeople more than they listen to marketing, so when a rep tells the team how they closed a deal, it spreads a new pitch faster.

With the notebooks, the turn came from one sales champion who worked with me as an insider, until the rest of sales saw the value of selling notebooks on top of everything they already sold, and notebooks became a very lucrative product for them to sell.

The principle: People support what they help create.

The action: Find one sales champion, a rep who is high-performing but frustrated by the current lack of consistency, and work with them directly. Co-create not just the messaging for their next big pitch but also the price books. When they win that deal using your new narrative and the pricing structure aligns with their reality, they become your best advocate, and the rest of the team will follow the person who is winning.

The strategic priority

The honest answer to my “Don’t you want the company to succeed?” was yes, of course, and the reps also wanted to make their number, but nobody had shown them how notebooks would do both.

When sales ignores your messaging framework or your assets, they are protecting the number they get paid on.

So before you call something a strategic priority, ask sales leadership to walk you through your reps’ commission plan and look for it there, and if you can’t find it, that is the first thing to fix, long before the assets.

Sales isn’t the only team that doesn’t want your messaging framework. Read why product doesn’t either.

The first time I joined a company as the first ever product marketer hired in the company, my first job was to explain what my job was. To product and engineering, marketing was almost the enemy and sometimes the blocker. They were completely confused about this new marketing breed that also claimed “product” in their title. Very soon they found my inexperience also came with uncomfortable questions, request of proof for everything being built, and constant demands for a seat at the table. Product Marketing was that annoying kid who feels too grown up for the kids’ table and wants to join the grown-ups.

Fast forward many years, I joined another company as their first Product Marketer, but this time, after a while, I was offered to also lead the product organization. I went from the kids’ table to running the grown-up one. I found out the roadmap was built on fires. Its logic was a backlog filled by the loudest voices: random ideas from the CEO and tickets from customers complaining, very loudly. The backlog had no way to check whether users would use any of it, since a user who ignores a feature never files a ticket about it. That hurt product and engineering most. Because there’s nothing that gets this group more excited than building something users love and use.

To product and engineering, product marketing often looks like an interrupter, or like overhead: a function that adds process without adding value. Asking for a seat at the table doesn’t change that. To them, it means one more person in their meetings.

The way out is to stop acting as a process enforcer and start acting as a force multiplier: the person who gets their features used.

1. Be a multiplier: become a feature adoption partner

Product and engineering teams are measured by the success of what they build. A feature nobody uses is a failure. From their side of the table, asking for a seat to review their work sounds like asking to grade their homework. Work on what could stop people from using the feature, and they start inviting you to the meetings you used to beg to join. Helping adoption is a magic gift to product and engineering teams.

The principle: Make their work more successful, and you become an asset.

The action: Stop asking for a seat at the table. Ask this instead: “What is the biggest risk to the adoption of this new feature?” When you help them solve the problem of why a feature isn’t being used, you are helping them reach their own goals, and nobody has to be told what to do.

2. Demonstrate utility: the low-hanging fruit pilot

You can’t build trust with artifacts. You build it with one undeniable win. Many of us have been the new product marketer who shows up with a persona deck, a RACI and a messaging house before shipping a single thing with the team. The team nods politely, and we take the nod for agreement. The RACI ends up in a shared folder, next to the previous RACIs.

The principle: Trust is built through demonstrated competence.

The action: Find one upcoming launch, or one struggling feature, that the product team is worried about. Offer to take the go-to-market burden off their plate entirely, handling the messaging, the sales enablement and the customer feedback loop for that one project. When they see that your involvement led to better adoption, clearer sales conversations, and actionable insights (that mythical creature) they will ask for you on the next one.

3. Respect the domain: act as a translator

Product and engineering resist because they fear product marketing will interfere with how they build. They are usually protecting something real, like their own user research, platform stability, a half-finished migration, or data from systems that were never built to fit together. Engineering is often building the plane while flying it, and here comes marketing down the aisle, asking whether they could also repaint the wings.

The principle: Respect the how and why, which belongs to product and engineering, and own the who and the what, which belong to product marketing.

The action: Position yourself as the bridge between the technical reality and the market reality, and use their language. Instead of saying “You need to change the roadmap,” say this: “I’ve been talking to customers, and they are struggling with X. At the same time, no one cares about Y, because it requires them to have Z and none of them have it.” Now they have a customer reason to work on X and to stop spending on Y, and the decision is still theirs.

Product and engineering ship releases when code is ready. Product marketing launches when the time is right.

4. Bring on the data: Use data as the neutral arbiter

Bring a subjective opinion to a product meeting and you invite conflict. Bring objective customer data and you invite collaboration. An opinion is one more loud voice in the backlog, and product marketing rarely wins a shouting match. A pattern across a stack of sales calls is a lot harder to argue with and much more interesting.

The principle: Data is the only language engineering and product teams universally respect.

The action: Use research to bring them insights they don’t have: an analysis of multiple sales recordings, the latest developments in the market, where competitors are signaling they are going, and why. Show a shared problem without criticizing their product. The conversation shifts from “product marketing vs. product” to “us vs. the customer’s problem.”

5. Incrementalism: build the playbook with them

Imposing a rigid playbook triggers resistance, and announcing it at an all-hands with a brand-new acronym is the fastest way to get it ignored. The opposite mistake is the polite, facilitated conversation where everyone shares an opinion and nobody moves until there’s consensus. I made that one. What the team needed was a leader.

The principle: People support what they help create.

The action: Don’t write playbooks in a vacuum. Co-create the first version of the launch process with one or two product managers who are open to it, and ask them: “What is the most painful part of launching a feature for you?” Build the first iteration of your playbook to solve that specific pain. As you solve their pain points, the playbook becomes the standard because it makes their lives easier.

The seat

Product marketers insist on asking for a seat at the table. I ended up with the whole table, and the resistance still didn’t ease until what we shipped and launched together started getting used.

When product and engineering resist, they are protecting what they care about: their platform, their users, and the adoption of what they ship. The way past the resistance is to protect the same things they do.

So stop asking for a seat. Take the feature your product team is most worried about, ask what the biggest risk to its adoption is, and start working on the answer before anyone sends you an invite.

Anker is folding five brands into one: Anker, eufy, soundcore, SOLIX and eufyMake will all become Anker. Founder and CEO Steven Yang explained why at IFA in Berlin in September 2026: “Category brands made sense when categories were separate. They no longer are.”

Anker already has product-market fit, with more than 200 million customers in over 180 countries and regions. I see the change as a decision about what story Anker tells as it grows.

Cost is part of it: five brands means building recognition five times over in every country. The reason Anker gave is in Yang’s sentence. A brand structure describes how a company’s products relate to each other. Anker’s five brands described five businesses: charging, audio, the smart home, energy and creative tools. Software and AI now connect them, so the old structure describes how Anker used to work.

The new structure leaves the products out of the names. Anker grouped everything by where a product fits in your life:

• Anker For You: The technology that moves through the day with you (power banks, audio, creative tools, and wellness).

• Anker for Home: The home that takes care of itself (robotic vacuums, home energy storage, and smart security systems)

Each group shares a situation: the things you use yourself, and the things that run your home.

The AI products add a second thread. Anker built a chip called THUS that runs AI inside its earbuds and headphones, so the processing happens on the device. It also built MindBase, a hub that connects the devices in a home, makes decisions locally and keeps the household’s data there. Both come from one idea: AI that runs close to you, on your devices and in your home. Five brands that look unrelated make that idea hard to explain.

Why does one story beat five brands?

A customer trusts a product after it has worked for them. Being excellent at chargers earns Anker trust in chargers. When the same company sells a robot vacuum under the eufy name, that trust mostly stays behind, because the customer has little reason to connect the two.

The vacuum has to earn most of its trust again, and so does every category after it. A company that keeps doing this ends up as separate specialists under one roof.

Attention works against it too. A product catalog can grow without limit, but a customer’s attention can’t. Every product that arrives with its own story hands the customer a job: working out how it relates to the others.

One brand helps with both problems. The customer has one name to remember instead of five, so trust earned by a good Anker charger reaches the Anker robot vacuum. A shared belief adds a reason: it explains why one company makes both.

To me, the new structure says Anker thinks AI should stay on your devices and in your home. If you like that your earbuds run their AI on the device, you have a reason to trust a home hub that keeps your data at home before you’ve tried it.

The same name also carries damage. A recall or a privacy scare in one category now hurts Anker everywhere else. Anker is betting the trade is worth it.

Product-market fit got Anker its charger customers. Whether they buy the vacuum depends on whether they know why one company makes both.

How do you know when you have narrative-market fit?

Ask customers what your company is. If they say your belief back, and why your products belong together, you have it.

The names and groupings of your products are how you get there. Customers meet them before they read any marketing copy, and product marketers call that structure the marketecture. If the marketecture tells five stories, every campaign has to fight it.

By that test, Anker doesn’t have narrative-market fit yet. It has made the bet, and customers will answer it. If in a year they describe Anker as the company whose AI stays with you, the story worked. If they call it a charger company that also sells vacuums, it didn’t.

Anker is moving new products to the Anker name first and existing ones over time. The risk sits in the middle, while the old names still describe a company Anker no longer wants to be.

Product lines drift for a simple reason: a name gets added every time a launch needs one, and nobody owns the decision to retire it. Anker just made that decision for four of its five brands.

How does narrative-market fit work for a first product in a new category?

A first product in a new category has to change the measure it is judged on, because on the old category’s measures the old category wins.

Customers understand something new by comparing it with something they already know. If they file you under the old category, they judge you on that category’s measures: price, features and utility. The incumbents have spent years getting good at exactly those measures. Judged that way, the old category wins on its own terms.

People need a reason to leave something that mostly works. The strongest one I’ve found is a cost they’ve been paying without noticing, so I start the story there. Then I pick a measure where the new product is strong and the old category is weak.

I worked through this at Product.ai. People saw us as a small company trying to beat ChatGPT, and on ChatGPT’s measures a small company loses. The cost of the old way: people bought on claims nobody had checked, because the AI giving the advice learned from fake reviews and marketing hype.

So repositioning us as the truth layer for commerce, a layer on top of any AI model that checks a product’s claims before you buy, was the way to go. The measure became whether a claim was verified. General-purpose AI repeated what it had read, fake reviews included. In our interviews, the reaction changed from “you’ll lose to OpenAI” to “you do what OpenAI can’t.”

Choose that measure carefully. The story your first product tells is the one every later product will be judged by.

How do you test your own product line?

Write one sentence about what your product line is organized by: category, audience or situation. Then write one sentence about what your strategy says is true today. If the first describes a world you’ve already left, the structure is the problem, however much brand equity it holds.

If you have one product, write down which old way it shows to be broken and what that costs your customer. If you can’t, you have a feature so far, and the category is still waiting for someone to name it.

Last, look through your product line for names that no longer match what the company is. Each one is a narrative-market fit problem.

Disney made the same move when it put Cinderella, Snow White and Ariel into one Disney Princess line.

Article Source: Anker Innovations Unifies Its Brands Under a Single Name: Anker, Anker Innovations, September 3, 2026.

A man in Germany was walking me through how he catches his neighbors recycling wrong. Someone had put something non-recyclable in the Biotonne, the organic waste bin, and he had worked out who it was with a patience Sherlock Holmes would have respected, so he could set them straight. In the US, he’d be the neighbor everyone avoids. In Germany, he’s a hero.

I was there to find out why people in Germany weren’t printing their designs with Canva. That interview was one of 45. In 2020, Canva Print was just getting started, but Canva loves to “Set Crazy Big Goals and Make Them Happen,” so one of those goals was to launch and grow globally, fast. What it didn’t have was a way to decide where to start, or what each country needed before we launched there. The default answer was to localize everywhere at once. And the goal behind all of it was harder than getting more people designing in Canva. We needed people to place a paid order and wait for a box to arrive, which is a much bigger thing to ask of someone.

I didn’t believe localizing everywhere was enough, not even with transcreation (the art of adapting to cultural nuances with language). A translation fixes one reason a market stalls, which is language, and markets stall for plenty of other reasons. Spread one team across every country and you end up with a pile of half-finished fixes and no idea which of them worked.

How do you choose which market to launch first?

Start with the market whose blocker your own team can remove. Size still matters, but a big market whose fix sits on another team’s roadmap will wait for that roadmap.

So apart from asking which market was biggest, I threw in an additional question from day one: which market’s blocker can I actually remove myself as a Product Marketer individual contributor? Positioning, which is really just deciding what story you tell and who you tell it to, can change in weeks. Print infrastructure in a new country takes six months or more. If the thing standing between a finished design and a paid order lived on another team’s roadmap, that market would wait for that roadmap no matter how big it was, and so would I.

I started with data, looking for countries where Canva adoption had held up over time. Brazil, Germany, and Japan came out on top. Then I held one-on-ones with Canva’s local Research and Localization teams, who knew those markets far better than I did, and asked what they thought was stopping people from printing. They had hypotheses, and I wanted to test them against users.

The choice that mattered most was who I interviewed. In each market I talked to 15 people who were designing business cards and other printables in Canva but not printing them with us. Customers can explain why they bought. These people could explain why they had done all the work, finished the design, and then taken the file somewhere else, or nowhere at all.

What blocked print in Brazil, Japan, and Germany?

Every market turned out to have its own reason. Brazil had grown after Canva localized into Portuguese, but people there printed locally. We had no print infrastructure in the country, so shipping made us too expensive to compete, and in a mobile-first market, ordering prints from our mobile app was cumbersome. Japan was a problem I had met before. At MOO, expanding from Europe to the US had taught me that business card sizes aren’t universal, and Japan had the same issue with more at stake. Our cards didn’t fit the standard holders for meishi (business cards), and our nengajo, the New Year’s greeting cards that matter a great deal there, were the wrong size too.

On paper, Germany was the easy one. It was also translated. There was no blocker that Product, Operations, or any other team had to build, but German Canva users weren’t printing with Canva or with anyone else.

The reason was sustainability, deforestation in particular, which the print business feeds through paper manufacturing, along with inks that pollute rivers. That’s a perception problem, and perception is exactly what Product Marketing exists to fix.

On paper, two of the three markets belonged to someone else’s roadmap before it hit mine. All I could do was pass them the research analysis and the actionable insights. The third belonged to me.

So I recommended parking Brazil and Japan until Product and Operations could fix the infrastructure. That is not an easy recommendation to make in a company that wants speed, especially about two markets where adoption was already holding up.

But marketing into a price gap, or into a product that doesn’t fit the holder, only pays to show more people the problem.

The manufacturing findings, the card sizes and the infrastructure gap, went to Print Operations, so parking those markets came with a list of what to build first.

Then Germany stopped being easy.

Remember the man with the Biotonne? To understand how seriously people there take this, you need to know that a pizza box doesn’t go in the paper bin, because it’s greasy and belongs in residual waste, and a yogurt cup only goes in the yellow bag after you’ve rinsed it. These were not people who would be won over by a nice paragraph about paper.

Which was a problem, because I had already published that paragraph. I had put out material about recycled paper and sustainable printing practices, and the people I interviewed read it as greenwashing. The obvious next step was to back it up properly, so certification, recycled paper, and sustainable printing practices all went into the German launch. They were necessary, and they still weren’t enough. The users told me why. As far as they were concerned, the damage to the forests was already done, and it was irreparable. Better paper only promised to do less of it.

But what if we could repair it?

How One Print, One Tree turned proof into growth

Repair had to be something German users could see for themselves, without taking our word for it. The commitment behind it already existed. Melanie Perkins, Canva’s CEO and founder, sponsored a goal of net-zero carbon emissions by 2040 and 100% renewable energy for Canva’s operations, but none of that was visible to someone deciding where to print a business card. What they could picture was forests disappearing, so the answer was to put trees back, one for every print order.

I found a partner who could plant the trees and certify the planting, Reduce. Reuse. Grow., and brought the idea to Canva’s Sustainability team with the partner already lined up. They helped me sharpen it, we presented it to Melanie together, and she approved it. That’s how One Print, One Tree started.

The design choice that mattered most was that customers didn’t have to do anything. Every completed order planted a tree automatically, with no opt-in and no policy page to read, so the proof showed up whether or not anyone went looking for it.

The go-to-market plan, how the program would reach people and what would convince them, came down to positioning on the website and proof in the box. The website led with the promise: “One Print, One Tree. That’s a win for the planet and a win for humankind.” Underneath it sat the whole idea in one line: “For every print order placed with Canva, we plant a tree – it’s a simple idea with a big impact.” The box carried the proof. Every order included a postcard showing the customer the reforestation impact of their own order and a sticker to brag about it at their next neighborhood recycling meeting. A website asks you to trust a company. A postcard and a sticker with bragging rights in your own box show you what your order did, and for people who had already decided our claims were greenwashing, that difference was the whole point.

Germany went first. It was our toughest trust market, with our highest-expectation customers, and if the proof held up with them it would hold up anywhere.

In the first three months after launch, the number of people placing paid Canva Print orders in Germany each month grew 50%. That was double our target, in half the time we had given ourselves. The pilot then became a global program, and leadership added large-scale reforestation to Canva’s company goals, which is what happens when the proof is anchored to something the CEO already sponsors. The program was featured in Fast Company’s Best Workplaces for Innovation, and today Canva counts more than 16 million trees planted through it.

A method for prioritizing international markets

Strip out the trees and the pizza boxes and the method is simple enough to replicate. Find where adoption is already holding up. Talk to the people who do the work but don’t buy, and find out what stopped them. Then sort every blocker by who owns the fix and how long it will take, and start where the owner is you.

Don’t assume the market that looks easiest on paper is easy, though. Germany had nothing to build and was still our hardest market, because the problem was belief. Localizing everywhere promised speed. The fastest route ran through the market with the hardest problem, because that problem was mine to solve. Before you launch anywhere, find out whose roadmap your blocker lives on. If it’s someone else’s, so is your launch date.

Subscribe now

During high school, a light box was my best friend and the most advanced gadget I owned. I stored in my bedroom transfer type and foam boards in bulk, T-squares, and magical blue pencils to write annotations for humans that print cameras couldn’t see. That is where I started, in design.

A product marketer telling you she loves generative AI should have no credibility. I know how it sounds. But I loved digital design tools long before I loved marketing. I fell for them the first time I opened Photoshop and Aldus FreeHand in the early 90s (FreeHand would eventually end up inside Adobe too), and I have been on their side ever since. I was a designer first, and I came in right at the transition point, when the work was three crafts in one: the taste to know what was good, the skilled hand to translate taste to paper, and the technical execution to get designs to print. As you grew in your career, the execution got handed down to the junior teams, but learning the tools was a rite of passage. You could even build an entire career on execution alone.

I have watched fear arrive three times. Three times the design profession was certain the new tool would kill the craft. Three times it was wrong in exactly the same way. The third is happening right now, and the ending has not changed.

The ransom note era

Digital design arrived first, and the old guard rejected it, convinced that their creativity and craft were about to be devalued and democratized. (It is the same reason so many art directors would hate Canva later. More on that in a minute.) They believed that if design looked easy, it would stop being valued. But design has never been easy and never will be. A sense of aesthetics is not enough. A museum curator has taste and still cannot paint a Velázquez, not even with the best canvas and the best oils. It is a skill you are born with and then have to build, or it gets lost, because talent, like everything else, is forgetful.

The designers who embraced the new tools understood quickly that what had been taken was the technical part, and that the people using those tools without the creative gift were producing aberrations: every effect and gradient the software offered, a dozen fonts on one page, all because they could. So those of us who then guarded our Macs and Adobe licenses like our most precious property, and swore by Jobs and Knoll, started a crusade to defend good design against what looked like ransom notes in the workplace.

It was no longer designers against digital tools. It was designers against the misuse of them. The tools got more powerful, to the point where they were almost intimidating to open. The designers won the war and kept the digital tools. No one was displaced. The craft stayed valuable while the output grew exponentially.

The democratization of design era

Then Canva arrived, and those same art directors felt the déjà vu. They feared the return of the ransom note, and many of them never lived the original. The fear of Comic Sans and Papyrus in professional work is institutional memory now. Papyrus even got its own Saturday Night Live sketch, Ryan Gosling unable to accept that Avatar built a blockbuster logo out of a font that ships free with every computer. All they saw at first was positioning aimed at people who had never opened Illustrator, and they braced for their craft to be dismissed and devalued all over again.

The pandemic sent everyone home, and the fear turned into its opposite. Corporate marketing and creative teams suddenly had to get on-brand assets into the hands of hundreds of people who were nowhere near the design department, and do it without a designer touching every single flyer, deck, and social tile. I was in product marketing at Canva through that stretch in 2020, and the professional designer was the one person I could not win over.

The objection was never really about the software. It was about the messaging. Everything we said was aimed at the person who had never opened Illustrator, and to that person it read as permission to design. To a designer, the same sentence read as an insult. We had a “why” for the entire world but the “so what” for the one person whose blessing we needed most was missing.

So I stopped defending the product. Whether Canva was real design software or not was the wrong argument and not one we were ever going to win with professional designers. The question I started asking instead was: What is it that the designer gets to do that only they can do, precisely because everyone else can now create their own social visuals without them? The answer turned out to be the system. So the promise had to change from “design is easy” to “brand compliance with creative scale”, and the roadmap followed with the first iteration of the Brand Kit.

The professionals kept designing the core identity in the tools they already trusted, Illustrator, Figma, and uploaded to Canva the logos, exact hex codes, the brand fonts. Then they built locked templates, fixing the fonts and the positions so a sales rep in another country could change a headline or a date without being able to break anything that mattered. The designer became the architect of the brand instead of the assembly line inside it.

That shift never reversed. Someone has to build the system. Someone has to make the templates, millions of them, and design the guardrails that keep a global brand intact while thousands of non-designers work inside it every day. Canva did not take the work away from professional designers. It took away the tedious repetition and the thankless, invisible production work that never got put in front of a customer, and gave them back the part of the job that was actually design.

The AI slop era

Then the generative AI era arrived, and I recognized the fear for the third time. Part of it is the old fear: the craft made easy, the craft made worthless. But this time a bigger monster entered the room. Ethics. And that fear is well founded. Generative tools scrape the internet and merge everything other people made into a new design from a single prompt. For free. With no recognition. If you have ever published a digital design, chances are Midjourney has already used it to train its AI.

When you make a design, it is never truly new.

Nothing is lost, nothing is created, everything is transformed

That’s a Lavoisier quote I love, a line bridging chemistry and physics, and it applies to design too. You take what is already there, natural patterns, the movements that came before, human emotion, physical constraint, and you transform it into something new. That is not what a tool like Midjourney does. So I understand the ethics of the pushback. You don’t want to be plagiarized, so why would you build on the work of others, taken without their permission?

Adobe made the opposite choice. It trained its models only on content it owns or has the rights to use, millions of images from Adobe Stock, openly licensed work, and public domain material where the copyright has expired. Is the consent story perfectly clean? Not entirely. Plenty of those contributors signed their Stock terms long before generative AI existed, but the direction is right.

At Adobe MAX, I launched the Adobe Express integration into Spring by Amaze and welcomed more than 500 designers to experience it. If you love design even a little, the event feels like an amusement park. You learn more about what designers need by watching them design and print their own custom T-shirts on the spot than you ever could from asking them what they think they’ll do. Most designers brought their own work, while non-designers used templates to avoid starting from a blank canvas. Both audiences removed the distance between an idea and something they were ready to make real.

There are big announcements at that event. The feature that stayed with me was not one of them. It is almost embarrassingly small. AI that renames your layers in Photoshop. Anyone who works in Photoshop ends up with a file full of layers called Layer 1, Layer 54 copy, Fill 3, and you either rename them obsessively as you go, which is impossible because you never know where the work is going to take you, or you lose an enormous amount of time hunting for the right one. The feature reads what is on each layer and names it for you. Isolated Tree Trunk. Color Corrected Sky. Nothing creative is touched. It just gives you back the time you were about to lose.

If the villain for digital tools was the ransom note, and for Canva the off-brand asset, then for generative AI it is slop. The refusal comes from the same fear as always, that the tool will dictate the creative choices and replace the soul of the work. Designers reject the idea of typing a prompt and getting back a finished logo, and they are right to, because that is not direction. That is a wish. Prompt and ship is how you get slop, and you can still get slop with ten years of experience behind you if you cannot say what you want.

Early in my career, at an agency in Madrid, I worked with an operator who was not creative and never pretended to be, but who had a flawless eye for proportion and space. I would hand him the raw images, the copy, the fonts, tell him serif or sans, bold or italic, what was the headline and what was the body, and describe how the ad should look. That was everything he needed to build the first rough for my approval, and then adapt it to every national newspaper in the country. He was not the creativity. He was the execution, and he was great at it.

Later, when I art directed fashion catalogs for Grupo Cortefiel, one of Spain’s biggest retail brands, I would select the images, build a rough layout, and then spend the whole day beside a retoucher in the lab, telling him what needed work. Over time he learned exactly how, and how much, I liked to handle skin and flaws, so that by the time I arrived that part was already done, and I only had to direct the creative work, blend these two images, build me this effect, just by describing what I needed. I was the art director. He was the hands. That person is AI today. He is also the answer to the fear, because he did not disappear. He started writing his own macros and action sets, then batch systems that could run thousands of images at once while he supervised them. He went from doing the work by hand to running the systems that did the work.

Nobody displaced him. He moved up, the same way the rest of us did. Both of those persons taught me the same thing. You give the constraints, the intent and the specifics, and then you correct in passes.

Where I do push back is when the goal is to alter reality in order to manipulate people. A good friend of mine is a creative director at WWF, the panda company. His team keeps a zero policy on AI in their creative work. Their raw material is the natural world, which is irreplaceable and should be presented exactly as it is. Doing it by hand is how they keep the humanity in it. (They use AI for the internal admin. The zero policy is only for anything a viewer will ever see.) I respect it completely.

His deeper objection is one most people underrate. AI is a drain on the planet. Training the models and generating the images burns enormous amounts of electricity and water to cool the data centers, and I have no answer for him on that today.

I have lived this from the inside too, at Canva and before it: taking a tool the professionals were afraid would end them, and showing them it was the thing that would set them free.

None of these tools ever came to rob creativity. A machine can render anything you ask of it, instantly and endlessly, but it carries no childhood, no grief, no image it cannot stop seeing. Everything worth making is still someone turning their own life into something the rest of us can feel. You cannot train a model on a life. You can only live one.

Picture a dark road at night. You’re driving home. A deer comes out of nowhere, you swerve, and suddenly you’re in a ditch, alone, no one around, phone somewhere on the floor of the car. Now picture knowing your phone would detect that crash and send help without you lifting a finger.

That scene is the reason a telematics company I worked with broke into a market that didn’t know it wanted them.

Telematics, in case that word means nothing to you, is technology that tracks and analyzes how you drive. In 2021 we wanted to enter the home-security vertical, and the decision-makers there had never heard the term. So we surveyed 600 home-security customers in the US, turned the results into content and sales enablement, and used it as the entire argument to walk through the door. It worked, it drove sales pipeline.

Three numbers did the heavy lifting. 78% were concerned about road safety. Then the two that actually made the case: 66% would pay a premium for smartphone crash detection, and 43% would switch providers to get it.

When you own the data, you own the story behind the data. But the design has to be honest. A survey built to tell you what you already believe looks more like marketing when want you need is research you can use in marketing.

Even if you hand the job to a professional researcher, you have to approach the whole thing from first principles. I’m writing on this topic, because it kills me to watch how it usually goes. So let’s change it.

Usually, someone opens a blank Typeform, or a blank doc. They type question one. Feels good. Question two. Momentum. Somewhere around question fourteen a bit of panic sets in, because they are not sure if the questions will reveal what they’re trying to find out, so they add five more just in case. Then, once the survey is basically built, someone asks the fatal question: “so, uh, who are we sending this to?”

That order is upside down. And it will bring you a pile of data you can do absolutely nothing with.

The questionnaire is the last thing you build

I mean that literally. By the time you’re typing the questions real people will answer, every hard decision should already be behind you. Writing questions is assembly. It’s the part that feels like progress and accomplishment because you can watch it take shape, so everyone rushes to it. The real work happens before, and it happens in your judgment.

The mission: Start with what you’ll do with the answer

Not what you’re curious about. What you will do. Are you running this to sharpen your positioning? Rewrite your messaging? Reprioritize the roadmap? Feed a piece of content? Enter a new market? Decide this first, because it determines everything after it. It tells you which gaps are worth chasing and which ones are just fun to think about.

And let me be boring here: fun to think about is the enemy. I say that with love, as someone chronically curious who has lost real hours of her life this way. Curiosity also feels like progress, but produces rabbit holes.

So from here forward, run everything through one brutal little test: if I knew the answer to this, would I do anything differently? If the answer is no, cut it. Doesn’t matter how juicy it is. Cut it.

The gaps: What you don’t know

Now ask yourself the real question: what do I actually not know?

Two things feed this. The questions honestly buzzing in your head, the ones you keep coming back to. And the hypotheses you’re already carrying, the things you suspect are true and want to either prove or kill.

Use both, because if you lean only on your hypotheses you’ll walk straight into confirmation bias. You’ll build a fantastic survey engineered to tell you that you were absolutely right, and you’ll learn nothing. Your open questions keep you honest. Your hypotheses give you something sharp to test. You need the tension between the two.

The people: Decide who you’re rejecting and who you’re comparing

Who you ask matters as much as what you ask. Demographics first, then psychographics. It’s easy to trip over two questions that seem too similar and are not: who am I qualifying out, and who am I comparing?

Say I want to understand how women in sales use AI at work.

Gender isn’t a disqualifier here. It’s a comparison. I want to see how women behave next to other genders, so everyone comes in and I split them later. AI usage, also not a disqualifier. The whole point is comparing the people who use it against the people who don’t, so I need both. But sales is a hard qualifier. If you don’t work in sales, I don’t want you anywhere near this survey. And if I widen it to four groups, say sales, medical, law, and students, then anyone who’s none of those four is out, and the four who made it in get compared against each other at the end.

Reject the wrong people. Segment the right ones. Two different jobs. Blur them and you either poison your data with people you never wanted, or you throw away a comparison you needed.

The one test every question has to pass

This one’s harder than it sounds, because it means saying no to interesting things.

I could ask about seniority, and find out that VPs are power users of AI in sales while SDRs barely touch it. Genuinely interesting. But, now what? If the honest answer is “nothing, it’s just good to know,” then I shouldn’t have asked. I spent a question, and worse, a respondent’s patience, on my own entertainment.

Now flip it. If I actually would act on it, if knowing it would change where I put budget or how I build the product, then asking whether it’s true isn’t enough. I have to dig up why it’s true. Maybe VPs use it more because they’ve got the authority to green-light new tools and SDRs are stuck waiting for permission. That’s a mechanism I can do something with. Change who I sell to, change how I price, change the buying story. “VPs are power users” is a dead end. “VPs are power users because they can approve their own tools but SDRs are stuck waiting for approval” is a foundation. The number without the reason is trivia. The reason is the whole point.

Congratulations, you’ve written a research brief

Everything above is your research brief. The mission (what you’ll do with the answers), the gaps (your questions and your hypotheses), and the people (who you reject, who you compare). This brief is the actual intellectual work.

Once the brief exists, the questionnaire almost writes itself. Hand a good brief and a clear set of constraints to Claude or your favorite LLM and it’ll draft you a strong questionnaire in minutes, mapping questions to gaps, structuring the logic, phrasing everything cleanly. AI is probably better than you at this. What it’s not better at is deciding what matters, which is why the brief has to be yours, thought all the way through, with your judgment in. Skip the thinking and AI will generate beautifully worded questions about nothing.

On how to ask the questions

Remember those data points from my telematics survey? 66% would pay a premium. 43% would switch. If I’d gotten those by asking “would you pay more for crash detection?” and “would you switch providers to get it?”, I should have thrown the survey away.

Don’t ask people about the future. People are wonderful liars about the future, mostly to themselves.

Ask “what will you watch tonight on Netflix” and you’ll get the brainy documentary they fully intend to watch and never will. Ask “what did you watch on Netflix last night” and you get the truth (in my case, it was probably something dumb and comforting). Past behavior is real. Future intention is fiction.

So how do you get a number about something that hasn’t happened yet, a feature nobody’s bought yet, without asking people to predict themselves? You don’t ask. You make them trade.

That premium number came out of a Van Westendorp exercise: instead of asking what someone would pay, you show them prices and find where “too cheap to trust” and “too expensive to consider” cross. People are terrible at naming a price and honest about which prices feel wrong. In our survey, 66% of the cross happened above their current premiums.

The switching number came out of MaxDiff, which forces people to pick what matters most and least out of a set, over and over, until a real ranking falls out. Nobody gets to say everything is important. They have to give something up to keep something else. What comes out isn’t an opinion, it’s a ranking. For 43% of them, smartphone crash detection outranked the two features their provider was actually up-selling them on.

You can’t ask people to predict the future, but you can put them in a corner where they have to trade one real thing against another, and the choice reveals a preference they’d never have told you if you’d asked straight. The prediction is fiction. The trade-off is evidence. Those two numbers unlocked us into a whole new vertical because they weren’t guesses about tomorrow. They were choices made under pressure today.

One more craft rule while we’re here. Build in the why.

Your numbers come from the multiple choice and the scales. Your insight comes from the open text box where a human types the reason in their own words. Pick the moments where the reason matters and let them talk. That’s where the surprising why hides, and the surprising why is the reason people read your report.

Then, my favorite part, you analyze the responses

Getting the responses back is not the finish line. A CSV full of answers is not an insight, any more than a fridge full of groceries is dinner.

Here’s how I do it. I open a fresh, ungrounded session of Claude or Gemini, one with no memory of my hopes, no knowledge of my strategy, and no stake in me being right. I feed it two things: the results as a CSV, and the brief. The brief is what makes this work. Without it, the model just describes your data back to you. With it, the model knows what you were actually hunting for.

Then I ask for three things:

• Map every question I asked against the questions I was asking myself and the hypotheses I set out to prove or kill. This is the moment of truth. Did the survey answer what it was built to answer? Which hypotheses lived, and which ones died? Dead hypotheses are not failures, by the way. A hypothesis you killed is one of the most valuable things a survey produces, because it stops you building on a foundation that was never there.

• I also ask it to flag the traps. Any decent survey plants a couple of attention checks, obvious questions where a wrong answer means the respondent was clicking on autopilot, so you can scrub their data. Housekeeping, but it matters. Skip it and your numbers might be polluted by people who weren’t reading.

• And finally I ask for pink flamingos. Frank Hoffman coined “pink flamingo” back in 2015 for a predictable, obvious risk that leaders can see perfectly well and choose to ignore. A known known. I’m stealing his bird and flipping it on purpose. My pink flamingos are the opposite: the unknown unknowns. The patterns standing right there in the data that were nowhere in my gaps and nowhere in my hypotheses, that I never once thought to look for. Some of the best findings I’ve ever had were pink flamingos, because they are proof the data is telling you something, instead of just nodding along with what you already believed. Your brief keeps you disciplined. Your pink flamingos keep you surprised. You want both, and you have to ask for the pink flamingos explicitly, because a model handed a tidy brief will happily color inside the lines until you tell it to go find the weird stuff.

So no, the questionnaire was never the hard part. The thinking that happens before it, and the thinking that happens after it, is the whole job. Skip both and you end up exactly where you started. Do the thinking, and you’ll have a report people quote.

Start with the answer you’d act on. Work backward. The survey itself is just the last mile.

Before you learn to cut a pattern at fashion school, you learn something more fundamental about clothes: nobody gets dressed simply to be covered.

I studied Fashion Design, and that idea has stayed with me much longer than anything I learned about actually making clothes. A garment is a claim. You put it on and tell people who you are before you’ve even said a word, and the designer’s job isn’t really solving the problem of covering a body, they’re helping someone make a statement about themselves.

I didn’t stay in fashion for long, but I never stopped applying the claim when I moved into Tech. At Nokia, and later at Microsoft, I worked on things people wear: phones, headsets, technology you put on your body in front of other people. I was still asking the same question design school taught me to ask about a dress: what is the person wearing this trying to say?

In December 1999, Andy Mooney arrived at Disney to run Consumer Products, the division responsible for turning Disney characters into physical things. Disney barely manufactured any of them; it licensed the characters to companies like Mattel and Hasbro, who made the toys and paid Disney a percentage. Mooney came from Nike, where he’d run marketing for a $3B apparel business.

A few weeks into the job, he stood in line at a Disney on Ice show in Phoenix, surrounded by little girls dressed head to toe as princesses; homemade gowns, mismatched jewelry, generic costumes with pieces added from the Halloween aisle from their local town. Almost none of it was Disney.

We know about that evening because Peggy Orenstein interviewed Mooney in 2006 for a New York Times Magazine essay called “What’s Wrong With Cinderella?” She was trying to understand why her own four-year-old had become obsessed with princesses. So the origin of one of Disney’s biggest franchises ended up preserved in an article that began with one mother’s curiosity about her daughter.

The business Mooney inherited was in trouble. Consumer Products was shrinking by as much as 30% a year, partly because Disney had licensed the same characters to competing manufacturers, who undercut each other on nearly identical products and drove the margins down. And there was a second problem: merchandising followed movies. A character got shelf space while its film was in theaters, then lost it the moment the next release arrived.

Snow White came out in 1937, Cinderella in 1950, Sleeping Beauty in 1959. No new movie, no re-release, no reason for a buyer at Walmart to give a fifty-year-old character valuable shelf space over something with a trailer running behind it (now, talk about ageism). So three of the most recognizable characters in the world sat in Disney’s catalog earning almost nothing, and not because children had stopped loving them, but because Disney built a system that made it hard to sell them unless a movie was selling them.

The obvious fix was to consolidate the licensing, stop manufacturers from competing on nearly identical products, decide which films deserved a merchandising program at all. That could have been the right answer to the data worth a few hundred million dollars.

The girls at that Disney on Ice show were showing something else, something not visible in any data sheet. Their homemade dresses weren’t really about Cinderella, or Ariel, or Snow White; they were dressing as princesses as a way to project themselves into the characters. Being a Princess was their identity for the day. Disney didn’t have a category called Princess. Mooney made it one.

Disney Princess became a brand in its own right, with a name, a fixed roster of characters, a color palette, design standards its licensees had to follow, and its own place on the retail shelves. Most importantly, it didn’t need a movie. Snow White stopped being a character from a 1937 film waiting for a reason to matter again, and became one member of a line a six-year-old could discover any day.

There was just one problem: Disney had a rule. Characters from different fairy tales must never appear together: Snow White belonged to her story, Cinderella to hers, and the two of them were never to share a lunchbox.

The rule wasn’t arbitrary. Each character lived inside a complete world, and the magic that world had on a six-year-old depended on her believing it was the only one. Put Cinderella beside Ariel and you risk turning both of them from worlds into products. That was the objection to Disney Princess, and it was a good one.

The insight was that the rule and the reason for the rule weren’t the same thing

The reason was worth protecting: never destroy the mythology of the characters.

The rule was only how Disney had been protecting it: don’t put them together.

So Mooney kept the reason and changed the rule.

Look at the packaging, and you’ll see the princesses appearing together, but they never look at one another. Each one gazes slightly away, as if the others aren’t there.

It’s a tiny decision, and it solves the whole problem. Cinderella doesn’t suddenly know Ariel exists; her world hasn’t been rewritten. The characters can share a commercial universe without being placed inside one another’s stories, a small change in how they’re drawn does what a decades-old corporate rule had been straining to do.

Mooney didn’t override the objection; he understood it well enough to preserve what actually mattered, and once he had, the approach looked very different. The upside was an enormous amount of demand that already existed. The downside just shrunk to a matter of presentation.

The Princess products launched with no major campaign, no focus groups, no advertising, and sold immediately. The line cleared a billion dollars within three years, and Consumer Products grew from around 300 million in 2001 to three billion by 2006.

I made a similar move, at an AI shopping company. I was asked to build a community to help build the product, so the assumption underneath the request seemed obvious: a community like that is made of customers.

The original idea wasn’t wrong. We did need people who cared about what we were building, who understood the product well enough and felt the problem enough to challenge the product and make it better. But they didn’t have to be customers.

The people who most wanted to help build the product were professionals working in AI. They shopped too, of course (everyone shops) but that wasn’t why they cared about the product. They were already following what we did, already arguing with us in public about what and how we should build it. They weren’t after a discount or a gift card; they wanted a say in the product. So we changed who the community was for: instead of a community of buyers, we built one for builders.

Same principle Disney kept: keep the real reason of the source value. Throw away the rule: that the source had to be a customer.

I think I recognize the move because of fashion. Designing a garment is a constant exercise in separating what a thing is for from the form it happens to take. A sleeve and the reason for a sleeve are not the same thing.

Companies are full of rules like that. Someone had a good reason once and defined a behavior for it, then the reason faded while the behavior stayed, unchallenged. The behavior then became the rule and everyone forgot what really mattered.

You surely have one of these rules. Something everyone in the business can name, and it may even sound obviously correct. Ask what it’s protecting, then ask whether the rule is still protecting it, or whether it’s just the shape that reason took a long time ago.

That question has saved me more than once. It’s the most useful thing I took from fashion school.

That decision determined almost everything that followed, because it forced me to answer a more useful question than how to grow a community: what would a community need to prove about the business for it to matter strategically?

The situation

Product.ai is building the verification layer for commerce. Ask an AI what to buy and it answers in seconds, often with total confidence and very little accountability, having learned what it knows from a web distorted by SEO power games, paid reviews, affiliate incentives, and influencers recommending products they may never have touched.

Our job is to check whether a product claim holds true, and sometimes to say “don’t buy this” when the product will not help solve the user’s needs.

Frontier models are converging quickly. Claude, GPT, and Gemini reach similar levels of capability within increasingly short windows, which means that an advantage built primarily on model intelligence has an expiration date the company does not control. If Product.ai’s edge is the model under the hood, we have a countdown rather than a moat. My position, argued before I had evidence for it, was that capability converges and trust doesn’t.

Verified truth is slower to build and harder to fake. A competitor can ship a better model fast, but it’s much harder to ship years of accumulated trust.

The question was whether we could prove that people would stay with Product.ai knowing that a frontier model was doing the underlying work. If they did, we would have evidence that they were trusting the verification layer rather than simply following whichever model happened to be smartest that month.

That was the thesis the company needed evidence for.

The assignment arrived wearing a tactical shape: Build a community.

The task, restated

“Build a community” is a tactic dressed as a strategy. Accept it literally and you can spend a year producing events, membership numbers, and increasingly elaborate updates about engagement and belonging without ever answering whether the community is creating something the business actually needs.

So I restated the assignment: assemble the smallest group of people whose continued participation could prove that trust holds when capability changes, then connect that group to product, hiring, brand, and distribution so it becomes part of how the company operates rather than a program marketing runs on the side.

That restatement gave me a decision rule.

Whenever there was a choice between something that could grow the community quickly and something that could produce stronger evidence for the thesis, I would choose the latter. Whenever an activity generated just warmth but did not make the product, the company, or the evidence stronger, it was probably not worth doing.

That sounds obvious in retrospect. It was useful because, in a small company, there is always a faster thing you could be doing.

Four bets

Starting from zero instead of inheriting an audience

The first bet was to start from zero.

SimplyCodes users came for discounts, and a person who arrives for a discount is loyal to the discount before they are loyal to you. Import them and you import their motive, which meant testing a thesis about trust on a population selected for price sensitivity. Whatever came back might have produced a flattering number, but I would have spent the following year explaining why the number meant something it probably didn’t.

The second problem bothered me more.

There was a more fundamental problem with using the SimplyCodes audience: we would have started with people who already had a reason to engage with us, then mistaken that existing relationship for evidence of trust.

SimplyCodes users came to us because we helped them save money. If those same people joined Product.ai, we would not know why they stayed. Was it because they trusted our judgment about what to buy, because they expected another useful deal, or simply because they already knew the brand? That would have made the community larger, but the experiment weaker.

I also did not want to use discounts or other acquisition incentives to recruit people into a product whose premise was independent judgment. If our proposition was that shoppers needed someone who would tell them the truth even when the answer was “don’t buy it,” we needed to earn their attention without first giving them a financial reason to show up.

So I chose to start with people who were already looking for better answers, rather than people we already knew how to reach. The bet, therefore, was that composition would matter more than size. A few hundred people selected for the right reason could produce better signal than a hundred times that many selected for the wrong one, and signal was the entire point when there was nothing to sell yet.

The exposure was real. Months of slow, manual, unglamorous recruitment where every week without a large number was a week someone could reasonably ask why I hadn’t simply used the list.

That was the tradeoff I was accepting: less growth in exchange for better information.

Recruiting believers rather than users

The second obvious move was to build a community of shoppers, since shoppers would eventually be the users, and this is where most community strategies stop thinking.

Shopping is a task, not an identity. There is little reason to return to a group because you bought one pair of running shoes, whereas people will organize themselves around things they care about, believe in, or want to become better at. Saving money can work as an organizing principle because people develop habits and identities around it, which is why coupon communities have real staying power. Buying something does not usually work that way.

I also set aside the standard ambassador model, partly because I had already seen where it works and where it doesn’t.

At Canva, I ran an ambassador program that worked because Canva Print existed, creators already had energy around the category, and there was useful product knowledge I could share that they could take to their own audiences. My contribution was decoding the print specifications nobody understood so creators could carry that knowledge back to their communities.

Product.ai had no product to teach.

That constraint became useful because it made ambassadorship structurally impossible and forced us toward co-creation instead. An ambassador receives a finished thing and helps spread it; a co-creator helps shape something unfinished and develops a stake in whether it turns out well. The second relationship is slower to establish, but once it exists, people talk about what they helped create for years without being asked.

The co-creation was never paid, and that was one of the most important decisions in the model. Fifty dollars an hour for feedback is a perfectly respectable transaction, but it produces what transactions produce: someone arrives, gives you an opinion, and leaves. Pay for an opinion and you own an opinion. Ask someone to help build the thing they want to exist and you give them a stake in the outcome.

Contribution stayed unpaid deliberately because even a small payment changes the nature of the relationship. Someone who wants to be there sounds different from someone completing a task, and that difference shows up quickly in the quality of what they tell you.

The people we wanted were not hard to find because they were already gathering in places where the future of AI was being debated. We focused on three groups:

• AI operators who wanted a say in what AI becomes rather than another place to network

• Truth architects, category experts who know a domain deeply enough to catch an AI getting it wrong

• High-expectation customers, researchers who dig hard before buying because being wrong costs them something.

  

That third group became particularly important. Buy the wrong lipstick and you lose some money. Buy the wrong jeans and you return them, or you don’t, and they sit in a drawer. Skincare matters because the consequences of a bad decision can last for months and are visible every time you look in the mirror. Baby products need no explanation to anyone who has bought one. Athletic equipment matters because a bad choice can end a season, and there is no returns policy for a torn tendon.

Those people did not need to be convinced that verified truth was interesting. They already needed it, and many were doing the research themselves before they had ever heard of us.

Designing the system before designing the events

Most community work starts with the event and never climbs higher. Pick a format, put it on the calendar, run it monthly, and eventually you start to believe the calendar itself is the strategy.

I built the architecture first and put events well down the stack, where they belonged.

THE GRASSROOTS ARCHITECTURE

  LAYER              WHAT IT IS                       THE QUESTION IT ANSWERS

  DOCTRINE    ->    Grassroots marketing          ->  HOW we distribute
  SUBSTRATE   ->    The Alpha Team                ->  WHO the thing is made of
  MECHANISMS  ->    Forged objects                ->  WHAT we make together
  LADDER      ->    Tribe > Community > Movement  ->  WHEN each move applies
  INSTRUMENTS ->    Moments: Golden Hour,         ->  WHERE it happens
                    Product Labs, Tastings,
                    hackathons, Guest
                    Appearances, Industry Events
  EXPRESSIONS ->    Brand: voice, visuals,         ->  WHY they stay
                    experiences, product,
                    recruiting, partnerships

Doctrine

Doctrine was the first layer because it determined which shortcuts we would refuse.

I wrote it as three refusals rather than three ambitions because an ambition can be interpreted away when the pressure arrives, while a refusal tells you what you will not do even when it would be easier.

We grow through utility rather than hype, which means the first interaction should deliver value and rules out manufactured viral loops and promotional noise. We grow through pull rather than push, which means people should adopt and share the tool because they find it useful rather than because we engineered friction around it, and rules out spammy notifications and artificial referral mechanics. And we protect trust over short-term metrics, which means neutral tone, accuracy, and clarity take precedence over anything that makes an engagement chart look better, and rules out dark patterns, sensationalized language, and addictive loops.

Those were not brand principles. They were operating constraints.

Building the substrate

Next came the people. I designed two doors in because the people we wanted arrived through different channels.

The online door was a join-page application that asked for evidence of past behavior: two stories about something they had investigated or challenged, what they already research before buying, and the basics of who they were and where they lived.

The offline door was Golden Hour, external events, Alpha referrals, and introductions through the company.

Both paths led to the same progression: Applied → Selected → Contributed → Recognized

A person reviewed every application against a rubric. Every rejection came back with a reason. That was intentionally slow.

A rejection with a reason from a human being is itself a brand expression, and some of the people we rejected came back through another door later.

The other important choice was that status advanced through contribution rather than tenure. You did not become Alpha Team because you had been around for six months. You became Alpha Team because you had contributed something meaningful, at which point your access increased: full Slack access, the test platform, a seat in product development, and your name attached to work you helped shape.

Nobody got paid. What they received instead was proximity to the build.

Building the ladder around the product

The ladder was the part almost everyone skips, and it’s why communities stall.

A tribe of a hundred needs intimacy and access. A community of a thousand needs structure and ritual. A movement needs a shared position and a way to carry it outward without the company standing in the middle of every interaction.

Those are different operating models.

I also gated the transitions on product milestones rather than headcount.

The tribe would become a community when the product reached closed beta, because that was the moment we could credibly expand beyond Los Angeles. The community would become a movement when the product reached general availability, because that was when people could organize around the idea without needing us to convene them.

This created an important constraint: we would never build the audience faster than the product could give it something meaningful to do.

A community with nothing to do eventually leaves.

Creating things the community could own

The mechanisms were the objects and experiences the company and community created together, the things that would not exist in quite the same form without the relationship.

One example I can talk about publicly is research reports every member could ask to help solve a hard question in their domain, built using the same proprietary protocol used to build the company and the brand. It turns membership into something tangible, something a person can use, keep, and show someone else.

Missions worked similarly. Members were placed on missions rather than applying for them, so the invitation carried weight. The goal was to make contribution feel like participation in the company rather than participation in a marketing program.

Running the instruments

Once the architecture existed, the events became easier to design because each one had a job.

Golden Hour runs on the first Wednesday of every month in West LA, free and only requiring a simple RSVP. Locking the date to the first Wednesday of every month was deliberate: a recurring slot can become a habit, while a floating one turns attendance into a decision people have to make again every month.

Then I looked hard at what other AI events in Los Angeles were getting wrong, because their weaknesses were our openings.

They skewed heavily male, so we built something where women showed up and came back, including a partnership with the Lovable builder community that brought in people who would never have found us through traditional AI channels.

The food at these events was reliably cold pizza and warm soda, and everyone had apparently agreed to pretend this was inevitable, so we made the food good. People who have eaten properly stay another hour, and that extra hour changes the quality of the conversation.

More importantly, most of these events treated the audience as an audience. We treated the people attending as professionals who wanted something useful they could use the next day.

We ran a masterclass on structuring context so AI models hallucinate less. We explored what discovery looks like when people ask an AI instead of typing into a search box. We walked through our verification system while we were still building it, mistakes included.

Eventually the community started programming the events for us, which was when I knew the system was beginning to work.

We started numbering each Golden Hour: 001, 002, 003. The same people returned every month, each received a collectible sticker. The stickers ended up on laptops around the city, and attendance became a streak nobody wanted to break.

When it became obvious that people wanted to meet each other more than they wanted another presentation from us, we ran speed networking, which became one of our strongest sessions.

When we realized nobody understood what axioms were, despite the fact that the concept sits underneath the product, we built a session around explaining them properly.

And when I saw how many members were building their own products, we ran a tasting modeled on the Whole Foods weekend sample experience, with stations where instead of tasting cheese and crackers, our product sat alongside products developed by our community members while people wandered between them, tried things, and discussed what worked.

The architecture gave us enough structure to know what we were trying to accomplish, without pretending we could know in advance exactly what the community would need.

Wiring the community into the company

A community that only generates goodwill is a line item. The community needed to change the company.

Product Labs became the bridge into product development. We brought together six or eight people who knew a category exceptionally well: skincare obsessives, parents, endurance athletes, and others with unusually high standards. The stated purpose was to understand how they actually buy. The more important purpose was to discover what we did not know to ask.

That distinction produced one of the most consequential product changes we made.

We were preparing to launch skincare as products people put on their faces, which is how the industry generally talks about the category. Members corrected us with something obvious in hindsight and almost invisible from inside the company: skin is the whole body, and skincare includes what you ingest as much as what you apply.

That single correction changed the category definition, the claims we needed to verify, and the scope of what we were building. No survey would have reliably surfaced it because a survey would have required us to know the question in advance.

That became a broader principle for how we used the community: do not ask people to validate what we already believe. Put them close enough to the work that they can expose what we have missed.

The same pattern repeated. We put prototypes and capabilities in front of members before committing them to production, testing concepts and workflows rather than simply asking whether an interface looked good.

The old sequence was to ship, then discover something was misunderstood or unwanted, by which point it already existed in production code and had to be undone. The new sequence moved that discovery upstream.

Hiring became the third connection.

We kept meeting people who clearly thought the way we thought but were not right for a full-time role at that particular moment. Letting them disappear because the requisition was closed felt like waste, so we invited some into the community instead.

There was no promise of a job and nothing asked in return. Some joined sessions. Some became candidates. Others simply made the community stronger.

Talent started entering the ecosystem before it entered the org chart.

The loop

At this point, the mechanisms stopped looking like separate community programs and started behaving like a system.

A member joins. They contribute insight we could not have generated internally. That insight becomes a product capability. They share something they helped create through their own channels. New people arrive, often through direct member referrals, and the cycle starts again.

                        THE COMPOUNDING LOOP

                           MEMBER JOINS
                          ┌─────────────┐
             ┌───────────>│      1      │────────────┐
             │            └─────────────┘            │
             │                                       v
      ┌─────────────┐                         ┌─────────────┐
      │      5      │                         │      2      │
      │  MEMBERS    │                         │  PRODUCT    │
      │  BRING      │                         │  INSIGHT    │
      │  OTHERS IN  │                         │  SURFACES   │
      └─────────────┘                         └─────────────┘
             ^                                       │
             │                                       v
             │      ┌─────────────┐   ┌─────────────┐│
             └──────│      4      │<──│      3      │┘
                    │  THEY SHARE │   │  IT SHIPS   │
                    │  WHAT THEY  │   │  AS A REAL  │
                    │  BUILT      │   │  CAPABILITY │
                    └─────────────┘   └─────────────┘

              Every turn costs nothing and makes the next
              turn easier. Nothing enters from outside.

The step most people get wrong is the fourth one because they assume sharing happens naturally.

It doesn’t.

People share what they helped create when they can point to something specific and say, that part is mine.

That is why attribution became a designed benefit rather than a nice gesture. When a member’s insight changes what ships, we attach their contribution to the work. The goal is not to manufacture advocacy. It is to give people something true to say about their relationship with the company. Every turn makes the next turn easier.

That is the beginning of a moat.

A competitor can buy media, hire more people, or upgrade its model. It cannot buy years of accumulated trust from people who helped build the product and can explain exactly why they believe in it.

Results

Over six months, we grew from zero to more than 300 members, without importing the SimplyCodes list and without paid acquisition or any marketing.

The more important results were what those members changed:

• Product decisions moved upstream, before engineering committed significant development time.

• The skincare category was redefined before a line of relevant product code was written.

• Members recruited members through their own channels, without payment or referral incentives.

• A talent pipeline emerged from people we had already met and, in some cases, already turned down.

• The brand narrative was pressure-tested through hundreds of individual conversations before it appeared on the website.

But the result that mattered most was the one the strategy was designed to prove. We built a group of people who knew a frontier model sat underneath Product.ai and chose to stay anyway. They were not loyal because we had the smartest model, they trusted the layer above it.

That is the thesis made visible in behavior.

When AI capabilities converge, people may not stay for the model that produced the answer. They may stay for the layer that tells them whether the answer deserves to be trusted.

What it cost

The decision to reject the existing audience created real risk. It meant months of slow recruitment when we could have produced a much larger number quickly. If the thesis had been wrong, I would have spent two quarters building something a list could have produced in two weeks.

There was another cost that was less obvious.

A system without fixed programming does not leave behind a calendar for the next person to copy. The architecture transfers; the individual instruments do not.

Whoever inherits this work has to run the diagnostic again at each layer. Copying Golden Hour six months later might produce an expensive evening with good food and very little reason to exist. The architecture exists so the reasoning survives after every tactic inside it has been replaced.

What transfers

The instruments here are worth almost nothing to another company. Your version has a different substrate, a different product, a different stage, and different moments entirely.

The order of operations is what transfers.

• Doctrine before instruments. Decide what you refuse to do before deciding what you will do. Principles should eliminate shortcuts, not decorate a slide.

• Composition before scale. The wrong people at volume produce a number you will spend a year defending.

• Product readiness before audience growth. A community should grow when the product gives it something meaningful to do, not because a dashboard says the membership target should be higher.

• Community inside the company. If members do not influence product, hiring, brand, or distribution, you have built an audience rather than an asset.

• Contribution over tenure. The strongest members should gain access because they make the company better, not because they have been around the longest.

• Use experts to expose blind spots. The highest-value contribution is often the thing nobody inside the company knew to ask.

And when an assignment arrives shaped like a tactic, restate it as the strategic question underneath before spending a dollar. That is often where the outcome gets decided.

The conventional approach would have handed me a shopper community built from an audience we already owned. It would probably have hit its number in six weeks and been dead by month four.

Starting from zero took longer, but it also gave us something much harder to manufacture later: a group of people with a reason to believe in what we were building, a reason to help build it, and a reason to stay when the underlying technology inevitably changed.

Subscribe now

I’ve been working in Tech for more than 20 years and lived through a lot of change. From digital transformation, to the internet, to now AI. Each era was a turmoil of change and you quickly learn that adaptation to change is a must.

Then you enter the hamster wheel and fight politics in the workplace, gender discrimination, age discrimination where you go from being too young, to being too single, or too married, or too old. Each stage of your life presenting a threat to employers because your life, kids, and family could get in the way of giving your everything in exchange for a pay check.

At this time in my career, I feel, and not for the first time, like I entered the mouth of a hungry monster as a fresh juicy bite that then went through a whole digestive system that crunches, dissolves, and extracts all the juices, vitamins and good material to feed the monster. The monster, like a good digestive system, is there to extract everything that is good from you so it can nurture and sustain the giant. Then what’s left is a pile of shit that the giant pushes out.

Being in the shit phase is scary, you think there’s nothing left for you to do. It’s when people decide to enter retirement if their age is close to it and their savings allow them to do so. It’s extremely scary for those too early for that.

The panic phase makes you think you must go back into the giants mouth as soon as possible to repeat the process. Through this metaphor you can probably understand that the shit can’t go back into the mouth of the giant, that would be extremely gross.

But it’s also not the end of the process. That pile of shit still has a purpose, even if it’s compost to feed new crops for the giant’s mouth.

I know because I have been this juicy bite, pile of shit, and new crop many times.

So, when you’re the pile of shit, what do you do?

You pause and focus on what’s next. Not opening LinkedIn and start applying to everything you see that resembles the last giant that chewed you. That’s going back as shit into the same mouth.

You stop and think what do you want. Not what’s the job you want but what’s the future you want for yourself. You’re free from the giant, you can now see that giant from outside and very likely you don’t want the same thing. You realize you don’t need the giant any more the giant needs you. You stop and think about what’s next.

You get specific about what you actually want next, not the job title, the shape of the life. Name why it matters, the real reason, doesn’t need to be super tidy. Then feed yourself the details of that thing so you can recognize it when it passes: the kind of people, the kind of work, the kind of feeling. And go where that thing actually lives, because you can only notice what’s in front of you. A sharp eye in the wrong place still finds nothing.

Then, when the right soil shows up, you move. Because none of this noticing matters if you sit in the pile waiting to be planted. The noticing just tells you where to reach. You still have to reach.

And here’s what I’ve made peace with. The giant was never evil. It’s just hungry, the way everything alive is hungry. It took what it needed from me and moved on, and honestly, so did I. Each giant was a chapter in my life. Each one fed me while it fed on me, and each one eventually spat me out so the next chapter could start. The chewing felt personal. It wasn’t. It was just the process doing what the process does.

So if you’re in the shit phase right now, reading this between job applications that all look the same, here’s the only thing I’d ask you to try. Stop looking for the last giant. Decide what you actually want to grow into, get specific enough to recognize it, and put yourself where it lives.

You are not being discarded. You’re being planted. The panic is just what it feels like before you know which way is up.

I’ve been the bite, the pile, and the new crop more times than I can count. Every time, the next chapter was better than the giant that chewed me. It just never felt that way from inside the shit.

And honestly, if you’re still here, reading this, noticing any of it making sense, you’ve probably already started. That flicker of recognition is your attention turning toward something other than the last mouth. That’s the whole first move. You’re further along than the panic is telling you.

In Madrid I order coffee like I’m issuing a correction. In London I learned to bury the same request under three apologies before it registers as manners. In LA people order coffee too, but the real devotion goes to cold-pressed juices that cost what a steak with frites costs in Spain.

For years I read these differences as a test. Learn the code, pass, belong. Fail, and you’re the loud foreigner nobody sits next to.

I wrote an article on that theory adapted to work culture. The argument was clean but a little cruel. Culture fit is on you, not the company. There are no good or bad cultures, only good or bad fits. If you show up and hate it, that’s not the culture being toxic, that’s you discovering it isn’t yours. So embrace it or leave, and if you stay to complain, you’re poisoning the well for everyone who actually wants to be there. I still think most of that holds.

I got one thing wrong, though. I treated adaptation as a cost. A tax you pay to be allowed in. Round off your corners, swallow your directness, learn the euphemisms, and in exchange you get to keep your seat. Belonging as the reward for the discomfort.

Then I read Ken Sakata writing about running a design practice in Tokyo.

Sakata moves from Australia to Japan and walks into the friction I’d have told him to either master or flee. Sixty-five meetings where almost nothing happens. A day trip to tour a glass cube with a dedicated room for exchanging business cards, only to be told the tour was the meeting. A business language with no word for “no,” where pointing out an obvious error is an insult, where you don’t win a negotiation by out-arguing the strongest person in the meeting because the decision belongs to something called “the company” that no single person present will admit to being. Every instinct he brought from Australia is wrong. By my old logic he had two options. Learn to live with it, or go home.

He does neither. He writes this instead:

“To avoid the discomfort of participation is to miss the point altogether.”

The slow, illogical, lineage-obsessed process isn’t the price of making the work. It’s what makes the work good. His clothes, he says, have never been more beautiful, and the reason is the compromise. The vintage construction techniques a pattern maker surfaces because “this is how we generally do things.” The idea that arrives with a Japanese shape it would never have taken if he’d built it alone in a place where everyone agreed with him.

The friction is not a cost. The friction is the product.

The difficult colleague, the committee that won’t decide, the process that makes no sense. I thought about all of it as obstacles. Things standing between you and good work that need to be survived. The light bulb moment is that they’re not obstacles. They’re inputs to work you could not have made as your unadapted self. The square peg doesn’t round itself off to survive the round hole. It rounds itself off because the rounding produces something the peg alone was never going to produce. The discomfort isn’t the entry fee. It’s the raw material.

Once you see it that way, two things separate that I’d always kept glued together.

There is your shape and there is your core, and many of us mislabel one as the other.

Shape is your register, your pace, your tolerance for silence in a meeting, how bluntly you deliver bad news. Shape is what you can change and still do the work well. Change your core and the work stops being yours.

“That’s just how I am” is the most expensive sentence in any working relationship, because it takes a smoothable angle and files it under principle so you never have to examine it. You can give up an enormous amount of shape and lose nothing that matters. You just have to stop confusing your bluntness for your talent.

“Do I fit here” is passive. It measures only the present, whether the hole happens to match the peg you are today.

“Do I want to fit here” measures whether you’re willing to do the rounding. That reframes leaving, too. You don’t leave because you failed a fit test. You leave because you looked at the friction and decided you don’t want to make anything out of it. That’s a clean exit. Not a victim fleeing a toxic culture, just a person declining the raw material on offer. The “This place is horrible” argument we see across Glassdoor reviews is rarely that honest. It’s frequently just “I didn’t want to do the work of adapting and I needed a villain.”

Which brings me back to the concept of “buenismo”, the Spanish word for kindness that’s actually cowardice. Avoiding conflict, rewarding mediocrity, smiling while everyone pretends the failure didn’t happen. I’ve ranted about it as a moral failure before, and it is one, but I never explained the mechanism. Here it is: Buenismo strips the friction out of a culture, and the friction was where the value lived. The candid feedback that stings, the colleague who disagrees with you in a way you can’t easily dismiss, the process that forces you to change your mind. Buenismo sands all of it down to keep everyone comfortable, and comfort produces nothing.

A culture with no friction isn’t kind. It’s sterile.

I used to think friction was the test you pass in order to belong, like some sort of rite of passage. I now think friction is the thing that makes belonging worth anything at all. A culture that never once made you uncomfortable never asked anything of you, and you will make nothing there you couldn’t have made alone.

Deep inside, you already know which frictions in your own week you’ve been calling toxic to avoid admitting they’re generative. One of them might be your core, and you should probably leave over it.

The rest are shape, and they are trying to build you something you can’t build by yourself. Keeping at this work long enough to make something takes what the Finns call sisu, the willingness to keep going precisely because it’s unpleasant.

The only real work is telling them apart. A comfortable culture spares you that work, which is exactly why you’ll leave being the same person you were when you walked in.

When we think about the products around us that become irresistible to use, they might seem like they were created to solve a problem, which is what product marketing usually turns into messaging afterward. But the ones that are really innovative come to life as the result of asking better questions. Simplified, a great product gets to us through three mental models and one discipline.

• First principles, to define what must be true in its pure essence.

• Systems thinking, to connect it to what happens in the real world.

• Strategy, to choose the trade-offs.

• Product marketing, to make those choices legible to the people the product was built for.

Let’s take two examples from the wearables category: two health trackers, the Oura ring and the Whoop band.

First principles is the discipline of separating what you know from what you inherited

Most product decisions arrive pre-loaded with assumptions. They came from the category itself. From the last company someone worked at. From what competitors already launched. The work is pulling each assumption apart until you get to what would still be true if no product in this category had ever existed.

For the Oura ring and the Whoop tracker, first principles asks what must be true for this product to succeed. With health trackers, the job goes beyond measuring biometrics. It’s about helping people continuously understand their health, which means both continuous data and continuous wearability. If people won’t wear it, the principle collapses.

The principle underneath the product is simple: stay on the body long enough to actually collect the data. A product that has to come off for any length of time is fighting that principle.

Whoop solved one of the instances where the product leaves the body by bringing the power to the body. A battery pack slides onto the sensor and charges it while the band stays on your wrist, through a shower or a workout, so the device never actually leaves. I thought this was absolutely brilliant and innovative.

Oura couldn’t take that path. A ring has no room for a battery pack, so charging means fifteen to eighty minutes on a magnetic dock, several times a week. Two answers to the same principle. Only one keeps it fully intact.

Systems thinking is about what happens once that truth meets a real person

Where first principles gives you what’s true in isolation, systems thinking gives you what’s true in context. Any product arrives into a life that’s already full: existing habits, existing objects, other people. Systems thinking means following the chain forward past the first effect into the ones after it. The first effect is the one you designed for, and the later ones are where things break.

People forget things. They get busy. They build small habits and break them without noticing. A ring that has to come off to charge is left behind on a nightstand during a busy week, and one missing night becomes a gap in the data, and a gap in the data becomes a reason to stop trusting the number the app shows.

A different kind of friction shows up on the wrist. A wrist band has to compete with everything that already lives there: another watch, a bracelet, your own sense of style. Oura chose to protect against style friction and accepted some risk to data continuity. Whoop chose to protect data continuity and accepted that it would spend its whole life fighting for wrist space.

Systems thinking doesn’t hand you one correct output. It hands you a clear picture of what you’re actually up against.

That picture is where strategy starts.

Strategy is choosing which version of that friction you and your users are willing to live with

And even so, which customer you’re willing to lose because of it. Someone who already wears a smartwatch every day has to consider giving it up for a wrist band. Someone training for endurance will happily dedicate a wrist to a device that never comes off. They’re different customers carrying different constraints, and a company that tries to serve both without choosing usually ends up serving neither one particularly well.

This is usually the point where a team tries to avoid the decision. Someone suggests a toggle, or two modes, or asks why not both. Every mode you add is a small tax on the thing you were originally protecting. The risk is ending up running a mediocre product. The team that ships the sharp version is the one willing to name the trade-off.

The largest strategic move a company makes is sometimes a change to its own rules. Whoop stopped selling a five-hundred-dollar gadget and started giving the band away for a monthly membership. That was not a discount. It changed what business Whoop was in. It was selling the analysis now, and the band became the sensor that fed it. That one rewrite reset everything happening after that: who they chase, what they build, how hard they work to keep you paying.

Once the trade-off is made, someone has to articulate it in a way the right customer feels seen

That’s the work of product marketing: the translation of the protected principle, built at the same time as the trade-off, stated plainly enough that the ideal user recognizes their own life inside it.

Whoop’s story works because it’s built around uninterrupted measurement for people optimizing performance who won’t tolerate a gap.

Oura’s story works because it’s built around meaningful insight that fits into a life you’re not willing to rearrange around a gadget.

Oura’s marketing shows the ring on a hand, at dinner, on a person who looks like they have somewhere to be. The object was beautiful before there was a marketing team to describe it, so the marketer’s job is to showcase that design decision. You look at the photograph and you can see if you would wear it, even before reading a single spec.

Whoop is a different and more interesting case, because the hardware decision in the product, the one that lets the device stay on your body through a charge, doesn’t appear in the marketing at all. Their message is strain and recovery, elite athletes. The outcome, never the mechanism. A decision purely rooted in the positioning. Whoop is not a hardware company. The band comes free with the membership and what you buy is the analysis. A product marketer leading with the clever battery pack would be clashing with the business model the founder chose. The mechanism goes untold because the positioning bans it.

Which leaves Whoop with a claim you cannot photograph. Nobody can see continuity in an image the way they can see a ring on a hand. So the proof has to be imported from somewhere else, and it is: Whoop’s trackers are now worn by athletes in elite competition, many of whom are taking a direct financial stake in the company. That means you cannot opt into Whoop from a photograph the way you can with Oura. You have to trust somebody first.

The throughline is simple. First principles defines the truth. Systems thinking protects it from reality. Strategy chooses the trade-off for a specific user. Product marketing makes the choice legible. Great products get built by a chain of thinking that holds together from truth to system to strategy to story.

Share

None of this is really about wearables. It’s about whatever you’re building right now: a product, a pricing page, positioning, a pitch. Every one of those carries the same four jobs.

The fourth job is the only one that produces something you can actually show: a deck, a pitch, a launch post, something that says we’re moving. The first three mostly produce doubt, debate, and a shrinking list of customers you’re willing to disappoint, and none of that feels like progress. It’s an easy trap. Skip ahead to the part that feels like motion, and the pitch sounds confident while the product underneath it doesn’t hold up to real use.

The story is the last honest thing you get to say about a decision, and it only works if the decision underneath it is sound.

You might have been told you’re an incredibly strong executor who gets things done. It arrives sounding like a compliment but the moment you hear “you get things done” it feels like a ceiling if you’re aspiring to leadership and a demotion if you’re a leader.

It means reliable. It means give it to them and stop thinking about it. What it does not mean, in most companies, is bring this person the problem before you have decided what to do about it.

That label has put my career on hold before. It’s been more a curse than a blessing because it makes things difficult when it comes to measuring the strategic value you bring to a company.

I have asked plenty of managers throughout my career how to make that strategy value tangible.

Most of them said I didn’t have to.

They meant it kindly. They meant, I see it, you’re fine, stop worrying.

But just feeling it isn’t enough. A feeling is not a record. Feelings get revised. People leave, budgets move, and eventually somebody who wasn’t in the room has to decide what you were worth. They’ll decide it from what’s written down. They will likely see projects moving and your name in the moving parts, never in the strategy.

Strategy is invisible when it works. Execution is visible by definition.

Start with what a manager can see.

Real life. Right now. We’re opening a test platform to outside users so they can provide feedback about our product. The strategy was defined before it reached me. The design was defined. My job was to make it survive contact with reality.

Which meant noticing that the original dry run only tested whether the tool worked but not whether the journey worked, so we rebuilt it to match what a stranger would find. Then that feedback was routing nowhere. Then that feedback worked only for people already let in, with no path to admit anyone new. Then that the whole thing ran fine on company email addresses and would have broken on the personal emails an outside user would bring.

Four problems, each one only visible after the one before it was solved.

All of that happened in Slack threads and on GitHub with my name across it. That is exactly what a manager is describing when they say “get things done”. They’re right, and that work is worth what they say it is worth.

Now the part a manager cannot see.

In 2016 we launched our own notebooks at MOO, and the year after we let businesses customize it. Then customers started asking for more colors. Six was not enough.

So we researched the trendiest shades and shipped forty-two.

Going from six to forty-two is not a small thing. Sourcing, production, inventory, photography, the site. Months of work, and it was done well.

Customers were still unhappy. Because there is always the customer who wants the forty-third color.

That is the cleanest example of flawless execution I have ever been part of. The direction had come from the people we trusted most, our own customers, and we carried it out beautifully, and it was worth almost nothing.

What they actually meant was that they wanted their brand colors, not more of ours. If you’re Tiffany, your blue is a trademark. It has a Pantone number, 1837, the year the company was founded.

Somebody had to ask the right question. Not “what do you want” but “why do you want that”. Only then did the real strategy emerge: a notebook where you could print any color, or any image, that represents the customer’s brand.

Nobody remembers who asked it, or when. Only that I was in charge of bringing that product to market, and that I executed well.

A few years later, at the same company, engineering came to us with devastating news for the business: We are drowning in tech debt. Nothing new ships until it clears and it won’t clear for a full year.

Our enterprise customers wanted products we didn’t sell and now couldn’t build. Big accounts, big orders. You don’t tell that kind of revenue to call back in twelve months.

Nobody handed me a strategy direction. Nobody could. Every direction anyone knew how to describe started with engineering, and engineering was gone until next year.

So I thought: “What if we do it by hand?”

I built a separate catalog in Airtable. We found third-party companies who could produce the things we could not, we grouped orders and sourced them out, down to telling the vendors which materials to buy so the quality held. None of it was on the website. None of it was self-serve. It sold through our sales reps.

If you had walked past my desk that year, you would have seen a woman maintaining a spreadsheet.

It made millions in a year we had already written off.

And it did something nobody had planned for. Running the thing by hand taught us which parts worked, which parts broke, and which parts customers didn’t care about at all. When engineering finally came back, they didn’t have to guess the requirements. We had spent a year writing them, in the only way requirements are ever honestly written, which is by doing the job manually until you know exactly where it hurts. It also taught us how to sell multiple products to enterprise, which was a muscle the company did not have.

Here is what that looked like from the outside: She got the orders out. She is very organized. She gets things done.

Here is what that got me: Not a promotion to Associate Director. I got a lateral move to sales enablement, a move I never wanted.

Put them next to each other.

Strategy that works looks like nothing happened. There was no crisis, no one asked me to figure out a problem. I just happened to jump on solving it without telling anyone. Execution leaves artifacts. A launch, a shipped thing, a green board, a Slack thread with forty replies. Strategy leaves an absence, and absences do not show up in performance reviews.

Worse than invisible, sometimes it is miscategorized. A year of the most strategic work I have ever done looked, from three feet away, like admin. Pure tactical work.

People credit what they can see. That is not unfairness. That is how looking works.

I wrote something to the team this week after an event that went well, and one line keeps coming back to me:

Strategy without people is an empty room. People without strategy is just a party.

Filling the room is execution. Deciding why anyone should be in it and why the room needs to look a certain way is strategy. You need both. But only one of them leaves something visible behind.

So here’s the test I use. Who handed you the direction?

On the test platform work, my CEO did. I made it real, which took judgment and took work, but the direction was not mine.

On the forty-two colors, the customers did. We executed it perfectly and it did not matter.

On the tech debt year, nobody did. There was nothing to hand.

Now run the question with your work. Take the piece of work you are proudest of from the last year. Then answer honestly: who decided that was the thing to do?

If somebody handed it to you and you made it real, that is worth something. Execution is not the junior version of strategy. Ask anyone who has watched a genuinely good strategy die because nobody could execute it properly.

If nobody handed it to you, go and find where that decision is written down.

My guess is nowhere.

The fix is not to start describing yourself as strategic. That is campaigning for the word, and the word is not evidence.

Capture the evidence instead. Before you solve the next problem, write down three things:

• The problem as you actually understand it

• The constraint you cannot move

• The two or three directions you considered and rejected.

The rejected directions are the hard part and they are the only part that matters. Anybody can write down what they did. Writing down what you decided against is the only thing that proves a decision happened at all.

And it has to be written before, not after. Afterwards you will not remember the constraint. You will remember the answer, and the answer always looks obvious in hindsight. Of course you route it to a third-party vendor. Obviously. That is precisely why nobody credits it. By the time the work is visible, the difficulty has evaporated, and what’s left looks like competent delivery of something self-evident.

Which is, word for word, what “you get things done” means.

None of this ends with somebody attaching a better adjective to your name. That may happen and it may not. Waiting for it is how people spend a decade being excellent and resentful at the same time.

What you are after is narrower than recognition, and better. Recognition is somebody else’s to hand you. Legibility is yours.

Somewhere in the last year, somebody handed you a problem instead of a plan. You solved it. You did epic shit.

Nobody wrote it down. You still haven’t.

We have inherited a very comforting belief about communication, which is that when it fails, it fails because the message was not clear enough. Say it better, say it plainer, strip out the ambiguity, and the problem dissolves. It is comforting because it makes the fix technical and puts it entirely in your hands. Sharpen the sentence and the right outcome should follow.

Communication is not just the transfer of information.

If it were, clarity would solve everything. Communication is a negotiation between truth and everything the truth threatens, which is identity, comfort, and the story a person tells about themselves. Information is only one of the things being managed in any exchange, and usually not the most important one. Which means a message can be delivered perfectly, received perfectly, understood down to the last word, and still change absolutely nothing.

Understanding was never the thing standing in the way.

We miss this because we judge communication by its emotional weather. When a conversation ends calmly, we decide it worked. When it ends in tension or defensiveness, we decide something went wrong on our side. Was I too harsh? Am I being a horrible person for not understanding personal situations? In the first case we think change will follow because communication was accepted. In the second we believe change will not happen because we failed to communicate. But conflict does not mean the message failed. Sometimes conflict is the first sign the message finally succeeded, that the other person now understands exactly what was said and what it implies and what will have to change, and the discomfort is not confusion. It is just resistance to change, resistance to a truth that was perfectly understood. A calm, agreeable, warm response can mean the message was understood and quietly filed away to die. We read the calm as success. Often it is simply a very efficient form of refusal.

I had to let someone go last week, a month after hiring them. When the work started slipping I did the thing you are supposed to do. I named what was not working, plainly, without the soft language we reach for when we want to correct someone without disturbing them. No “maybe”, no “there are a few areas we need to revisit”. It was clear, and they heard all of it. The response was everything you would hope for. They acknowledged every point, apologized thoroughly and more than once, thanked me for the honesty. I walked out believing I had found someone rare, a person who takes hard feedback and turns it into fuel. What I had actually found was a person who was very good at apologizing, which is an entirely different skill from being good at changing, and in the moment the two are indistinguishable. The apology was complete. The change never came. Same mistakes the next week and the week after, until the gap between what was said and what was done was too wide to keep covering. They were genuinely shocked when it ended. They had heard a clear warning, more than once, agreed with it every time, and still did not see it coming, because they had mistaken their own acknowledgment for progress, exactly as I had.

Sometimes, when direct feedback is delivered, what comes back is not engagement with the problem but a wave of emotion, followed by a personal reason the problem cannot be addressed right now, followed by the need to withdraw to recover. The message reaches them and is immediately smothered, because the emotion and the personal difficulty have now taken over the whole conversation. The work I raised is buried under three layers of things I am no longer allowed to question without acting like a monster. None of it is necessarily performance in the cynical sense, and that is what makes it effective. The person may fully believe they are responding reasonably. But the function is the same as the apology, a signal that a message was received.

This is the part the clarity-fixes-everything belief cannot account for. None of the people in these situations misunderstood me. They understood me completely. What they were resisting was not the message but the part where you actually have to become different, and each had developed an almost involuntary way of producing the appearance of that change without any of the substance. One produced apology. One produced emotion and a reason.

We reach for indirectness to deliver a hard truth without its full force arriving at once, to let someone keep their dignity and absorb a difficult thing gradually, and used that way it is genuinely humane and considerate. But the same softness that protects a person can just as easily protect a problem. Am I cushioning the human, or am I cushioning myself from the discomfort of watching the truth reach someone who does not want it?

Most of the time the vagueness we call kindness is avoidance.

The moment of correction tells you almost nothing.

The apology, the emotion, the eager yes, the warm thank-you for the honesty, none of it is evidence, not because those reactions are false but because they are unrelated to the only measurement that means anything: what is actually different two weeks later. A message has not reached anyone because they reacted to it well. It has reached them when the thing it was about has changed. Everything before that is weather, not outcome.

I do not have a method for this. I cannot tell you how to make someone change who has decided, sometimes unconsciously, that the appearance of change is enough. I have stopped believing that a better sentence exists that would have reached them, because clarity was never the broken part. When you have been clear, and you have been fair, and you have been kind, and still nothing moves, the failure is not automatically yours to carry. We are trained to assume that if the message did not produce change, we must not have communicated it well enough, and we go back and sand the words down again and again looking for the flaw in ourselves. Sometimes the flaw is not in the delivery. Sometimes the person understood you completely and simply chose not to move. That is not a communication problem you can solve.

I’ve written about the flashlight before as an antidote to strategic planning. Point a beam at the floor, take a step, learn one thing, step again, instead of lighting and mapping the whole field before you move. People liked the idea.

Then a few of them asked me how to actually do it. The metaphor makes sense. The method is where it works.

At the beginning of the year the CEO asked me for the strategy to build a community to co-create with us. I went ahead and wrote ten strategy docs in a single day. Identification, recruitment, onboarding, the tiers, the growth model, the full architecture. I did it because I could. I described the community I wished existed and AI built for it, fast and polished and complete. The speed felt like competence. Elena is thinking about everything. Elena has a plan for everything.

It was the biggest waste of time I had logged in years. None of it touched the reality of the world I was actually in. Every document was built on hope. Hope that people wanted to co-create. Hope that the tiers mattered. Hope that the model I sketched was the model that existed. Hope is a wish disguised as a plan. It is what you reach for when you haven’t met reality yet, and it feels like progress because it produces pages. Ten documents. Zero contact with the ground.

Kill hope before hope kills your project.

That is the first thing the flashlight does. You cannot plan your way out of not knowing. You can only walk into it.

That day is the reason I built this flashlight method. Four moves I run on the kind of zero to one projects where nobody has done it before and there is no frame of reference for it.

The first move is naming the question, not the goal.

The goal is grow a community that co-creates with us. But that is a wish. The question is do these people show up to give or to get. A question has an answer. A goal has a deck. The test for a real question is simple: it can come back and tell you that you were wrong. If I ask whether people join to get access or to give back, the answer might be the opposite of what I built for, and that would force me to tear up the plan. That is a real question. If I ask whether a thriving community would be good for us, there is no answer that surprises me, because I already know what I want to hear. That is not a question. That is me looking for confirmation. My ten docs were full of the second kind. They assumed every answer.

The second move is defining the beam.

The beam is the smallest test that answers the question this week. Not this quarter. This week. The constraint is the point. Give yourself six weeks and you’ll design a study. Give yourself five days and you’ll message ten people, ship one landing page, or run a single session with one group. The five-day version is worse research but better strategy, because the answer arrives while the question still matters. My rule: if the test can’t run by Friday, it’s too big, and I cut it until it can. Most beams end up a tenth the size of my first sketch. The first sketch is always me wanting to look thorough.

The third move is reading the signal.

This is where you risk dropping the method. You take the step. Something comes back. The temptation is to explain the result so it fits what you already believed. Say I built the whole community plan around members giving back, sharing expertise, mentoring each other, and I message ten of them to test it. Eight reply asking what they get out of joining. That is the answer. But I don’t like it, because my plan assumed the opposite, so I tell myself ten is too small a sample, the wording was off, I’ll run a bigger test next. That’s not analysis. That’s me protecting the plan I already wrote. That is confirmation bias, showing up as diligence. The real signal is the thing that surprised you. If nothing surprised you, you either asked a question you already knew the answer to or you flinched from the answer you got. Write down what you didn’t expect before you write down what you did. The unexpected part is the part carrying new information.

The fourth move is the turn.

You walk again, but not in the direction you were already facing. The signal changes the heading or it was a wasted step. Turning means admitting the first direction was a guess, because it was.

Every first step is a guess. The flashlight doesn’t make your first guess right. It makes your second better than your first and your tenth better than your second, which is the only way to get it right when nobody has gone first.

Name the question. Define the beam. Read the signal. Turn. Then run it again. The loop is the strategy.

There is no end state where you’ve planned enough and now you execute. You’re always one beam ahead of your own certainty, and that gap is the work.

Now the harder part, because the same AI that helped me waste a day is the same AI I run the method on now.

Start with the thing that makes AI dangerous.

AI builds the right thing at incredible speed and the wrong thing at exactly the same speed, and the wrong thing arrives formatted, confident, finished, indistinguishable from the right thing until you put it against reality.

Speed serves whatever you aim it at. Aim it at hope and it manufactures hope in bulk, which is what my ten docs were. So the question is where you point it, and which tool you reach for to point it well.

I use Claude. At first, only the Claude chat window, then chat and the cowork window, then chat and the code window. Now I operate mostly in the code window because of how our AI system is wired together at work. But you need to find the right tool for the moment you’re in and your personal setup.

I stopped asking the chat window for plans. But I did not stop using chat. Drop your real material into a Project, your research, your positioning, your own writing, and it reasons from that. Chat reasons and hands you back thinking. The ten-doc disaster was not the chat’s fault. It was mine, for treating its thinking as my reality when what it reasoned from was the world I described, which was the world I wished for.

Here is the decision rule I use. It is not as simple as chat-for-research, agent-for-action. An agent can research too. Code can search the web, read your files, run the analysis, synthesize the lot. The rule is simpler: match the tool to where the work ends and how much reach you need.

• When the work ends in your own understanding, use chat. Reading thirty customer interviews and finding the contradiction in them, evaluating the market position, weighing signals that point different directions, finding the real question hiding under the goal. The deliverable is understanding something you didn’t before. That is a conversation, not a delegation. I am in the loop turn by turn, pushing back, redirecting mid-thought, and I do not want the tool to run ahead of me and finish while I am still thinking. A Project keeps all of it persistent across sessions so the research does not reset every time I close the tab. The discipline here is the same hope problem: do not ask leading questions. “Why is this clearly the right move” gets you a confident defense of the move. “Argue both sides and tell me which is stronger and where it breaks” gets you analysis.

• When the work ends in a changed file, a shipped page, a built thing, I move to an agent, because an agent closes the loop by doing. This is Cowork or Claude Code, and they are the same engine. Same model, same agentic core, same ability to read your files and finish the work without you steering each step. Both run in the same desktop app. The difference between them is not the interface. It is reach. Cowork works inside a sealed box and touches only the folder you hand it, and forgets between sessions. Code reaches your whole machine, remembers across sessions, and wires into the other systems you run. Cowork is the smaller, safer box, which is right when you are starting, because inside it you cannot break anything. Code is the larger, sharper one, and the reach that makes it powerful is reach you should only point at work that needs it. It can write and delete across your whole machine. Aiming that at a question that only needs reading is a forklift for a pen. That is the second test: not just where the work ends, but how much reach the ending

Share

The agent is not faster emails and landing pages.

The best example I can share is mine. I am not an engineer. I built, in Code, a system that holds the soul of my community. Not demographics. Not a CRM with job titles and last-touch dates. A living model of what each member actually wants, individually, and what they want as a collective, that anyone on the team can query and see without waiting on a data engineer to build the report. I call it the individual and collective soul of our community. I built it myself with Claude Code, the tool that supposedly belongs to people who write software. I do not write software. I had a thing I needed to understand at scale, and I didn’t want to wait for someone else to build the means to understand it.

What that model encodes is the whole reason it was never a marketing initiative running on the side.

Most community work is a negotiation. I give you access, you give me feedback. I give you status, you give me advocacy. Trade and counter-trade. I did not want that.

The soul database is built on a different idea. It tells me what to ask of a member so that the ask itself is the thing that gives back to them. Not I give, then you give back. A loop, where what I request of you is constructed to return to you what you actually needed. The give and the get are the same motion. You cannot run that on a spreadsheet of attributes. You can only run it on a real model of what people want, which is the thing I built, which is why it sits inside the company as infrastructure, not beside it as a campaign.

That is what an agent is for. The thing you could not build before because it required an engineer, three months, a PRD, and a prioritization logic, and now requires you, a folder, and the willingness to work in a tool that was not labeled for you.

Here is where you start, since starting is the only part that counts.

Take the biggest ambiguous thing on your plate, the one you’ve been avoiding because you don’t have enough information to plan it. You’re never going to have enough information to plan it. Stop trying. Write the one question underneath it that you could be wrong about. Take that question into a grounded chat and make it argue against you until you find the sharper version. Then ask what you could ship, send, or run by Friday that would answer it, and move it into an agent pointed at your real material so the test runs against what is true, not what you wish for. Not the perfect test. The Friday test. Ship it before you talk yourself into a bigger one.

The resistance you’ll feel is the tell. The bigger test feels safer because it feels thorough, and thorough feels like strategy. It isn’t.

Thorough is what you reach for when you’re afraid of the answer and want to delay it under the cover of rigor.

That fear is hope’s last hiding place. I confused the size of my plan with the quality of my thinking.

You don’t need a better plan. You need a smaller first step, the real ground underneath it, the right tool for where the work ends, and the willingness to let it change your assumptions.

I used to spend three weeks turning research into a deck, then more weeks getting a dozen people to agree with it, then hoping my influence was strong enough that the work didn’t die at the bottom of a drawer. Now I spend an afternoon telling the AI what is true about our market, our product, our mission, and it carries that into everything it touches. The deck was never the work. The deck was the tax I paid to get other humans to act on the work.

For twenty years my whole craft was making other humans understand. The deck, the narrative, the alignment, the patient work of getting a dozen people to see the same thing. I was a translator. And the day I realized I no longer needed the humans to understand, only my AI, the translator’s job was over.

So now when the market moves or the product evolves or the mission shifts, I tell the AI first. The way I know something has shifted is that the output starts to drift, a few degrees off from what’s true. I don’t fix the output. I fix the source.

It looks like judging what the AI hands back, keeping the good, killing the bad. It isn’t. The catching is the easy part, and it is downstream of something harder, something many PMMs traded away for a template before AI existed. Skip that harder thing and no amount of reviewing the AI’s output will save you, because you won’t know what you are looking at.

I’m sure you have felt this drift before. You ask AI for website copy and it gives you something competent and dead. Clean sentences, right structure, and it could be describing any of forty companies in your category. The instinct is to edit the copy. But the copy is generic because the artifact under it is generic, the positioning and messaging you built, the thing the AI is drawing from. You never told it the one truth that makes this product different from the forty others, the thing you know in your gut and never wrote down. Fix the artifact and the next draft is alive.

And it isn’t only copy. This is the whole job now. You used to write the positioning and hand it off, then spend your quarter making sure the launch, the feature announcement, the new initiative all used it correctly. Now you build the artifact and the AI produces the launch, the announcement, the initiative from it, grounded in your thinking, in your words. You stopped writing the thing for someone else to execute. You now produce the outcome. The work moved from policing other people’s use of your positioning to making the call yourself, at the scale of everything the company ships.

Nobody became a product marketer because they loved chasing a sales team to use the right slide.

The source is not a prompt. A prompt is a request you make once and lose. The source is a set of documents the AI reads before it does anything. They hold what a deck used to hold and then bury: what the product is, who it is for and who it is not for, the market truth as of this week, the positioning, the words we use and the words we refuse. The understanding used to live in people’s heads and leak out the moment they left. Now it lives with AI in a place that doesn’t forget.

So I open the source and find the place where my own thinking was vague, because the model is only ever as sharp as the sharpest sentence I gave it. Usually the drift in the output is just a drift in me I hadn’t noticed yet.

In practice this looks like giving things up. I removed myself from approving customer-facing materials, because the AI already checks them against my documents better than I checked them by hand. My job is no longer the perfect application of the artifact. It is the perfect artifact. I catch the issue at the source before it becomes a hundred wrong outputs downstream. Actually, not policing the application anymore feels liberating.

But fixing the source is not a workflow. It only works because of something underneath it. You can only feel the output drift if you know where it was supposed to go. A sailor with no destination feels the boat change direction and learns nothing from it, because a change of direction means nothing if you don’t know where you were headed. They register motion. That is all. The sailor who knows the destination feels the same change and knows instantly whether it is right or wrong, because they are holding the one thing that makes drift legible: where the boat needs to end up. The drift in the output is only visible to someone holding the destination. And the destination, for a product marketer, is the live connection between what the company is for and what the market actually needs. That connection is the thing. Everything I described as a workflow is just me checking the output against it.

This is where taste and judgment part ways.

Judgment picks the best option from the ones in front of you. It is evaluation, and you can teach it, because it runs on criteria, and two people given the same criteria will land in roughly the same place. That is exactly why it can be templated. The template is the criteria, written down. Taste does the thing that happens before there are any options to judge. It draws the connection that wasn’t on the table, between this mission and this market, the one nobody had made yet. Judgment chooses. Taste connects. One sorts what exists. The other makes the thing that doesn’t even exist yet. You cannot template the second one, because there is nothing to write down until the moment you see it, and the next moment it is something else.

The hard part is that the connection is never the obvious one. The obvious answer is the consensus answer, the average of everything ever written about your category, which is exactly what the AI already reaches for. The real one is underneath that, the link between this mission and this market that no playbook contains, the one you can’t always prove when you first see it.

Remember “A thousand songs in your pocket”? When that line was used, everyone else was selling gigabytes, a number that meant nothing to a person standing in a store trying to understand what they were about to buy. The line was not a smart compression of the spec. Somebody connected what the company had built to what the market was actually trying to solve, getting their music with them, all of it, anywhere, and saw that the spec and the need had never been joined in plain words. The line came second. The connection came first. That is the move. Not reducing the gigabytes to a phrase. Seeing the link between the product and the real need that nobody, not the spec sheet, not the focus group, had yet drawn.

That is the part the AI cannot do because it only has the mission statement and the market data. What it does with them is draw the average line between them, the connection anyone with the same two inputs would draw, the consensus link. It cannot draw the true line, the one that fits THIS company and THIS market at THIS specific moment, because that line has never been drawn before, so it isn’t anywhere in what the AI has read. Your job, the part that survives, is not judging the AI’s output after it arrives. It is making the connection the AI can only average. The output is downstream of that. The connection is the source of the source.

Product marketing templates enabled PMMs to copy the answer instead of drawing the connection. So they never built the one thing AI can’t do.

I’m sad to recognize I helped cause this damage. My name got known among product marketers because I handed out my own templates and frameworks, and the better they were, the more people wanted them. People asked me to coach them but what they really wanted was the templates. I was an Expert in Residence at the largest product marketing institution. Then I reflected on what the title actually meant. Expert in Residence meant expert in frameworks. The role itself certified the premise that frameworks and templates are the expertise, that if you collect enough of them you’re equipped. I had come to believe the opposite. So I left. You can’t hold a title that endorses the thing you believe is hollowing out the craft.

The template economy sold a generation of product marketers the reusable canvas, the fill-in-the-blank frameworks and templates, and in doing so it let them skip the seeing. It made it possible to look strategic without doing the first principles strategic thinking. And it worked, right up until AI did the same move faster. If your value was reaching for the framework and filling it in, AI does that now.

Product marketing is not disappearing. But what we knew as product marketing already has. The part that was filling in the canvas and policing its usage is gone. The part that was deciding what belongs on the canvas in the first place, what to keep, what to cut, what the true connection between this mission and this market actually is, that part became the entire job.

And it can’t be handed off, because taste is a point of view. It is a position about what matters, and it comes from who you are. AI frontier models have a lot of judgment. But they have no self. No history. No particular way of caring about one thing more than another. It has read every life and lived none. Taste is what is left of a life, the residue of everything you have paid attention to and everything that has hurt you and everything you decided mattered.

Which is also why I am not going to give you an example. I could tell you about a moment I saw the thing before anyone said it, and you would study it the way you studied every framework, looking for the move so you could run it yourself. That is the trap. Not because taste evaporates the moment it is shared. I could hand you exactly what I saw and it would still be real. It would just be mine, true for that situation, that product, that moment, perceived through me. The problem is not transfer. The problem is reuse.

A template is yesterday’s perception applied to today’s different problem, and the problem is always different.

I don’t even reuse my own frameworks more than once, and that’s because the next situation is never the same as the last one. That’s why I have a tone of Miro boards filled with boxes and arrows than can be discarded once the problem has been solved. The taste was in seeing this one, not in the artifact I made the last time I saw clearly. Taste doesn’t survive being reused, even by the person who had it. That is the one thing a template cannot hold, and it is the whole job.

Taste is built through making the call and being wrong and feeling exactly how it was wrong, over and over, until the feeling becomes fast. That is the apprenticeship, and the template economy cancelled it. Every time the template did the deciding, a person lost the one rep that would have built the eye they now need. We optimized away the years that make it. Then AI learned to fill the canvas, and the years we skipped were the only thing that would have set us apart from it.

So coming this Monday, notice what your hand reaches for first. The next time you start anything, a launch, a positioning, a name for a feature, notice the reach. If it goes for a framework, a canvas, a template someone shared with you, even the way you solved the last one, that is the reach the AI already does better than you. Pull it back. Sit in the problem without the scaffolding and do the harder thing the template skips: draw the line yourself, between what your company is actually for and what this market actually needs, right now, for this. Not the line the category would draw. The one that is true only here. That is the work. It is the only part of the job still yours, because it is the only part the average can’t reach.

In the past, product marketing prepared the materials for others to act. Now the AI, carrying your taste, prepares the materials for you to act, not others. You.

But only if there is a you in there to carry. If your taste only ever lived inside a framework or a template someone handed to you, it was never yours.

I love commencement speeches. I love them for the speaker, who gets one shot to hand something durable to people they will never see again, and I love them for the audience, the crowd sitting in folding chairs in the heat that shows up wanting to be moved. There is a transfer happening, knowledge passing from people who have lived it to people who are about to.

My almost obvious favorite is Steve Jobs at Stanford in 2005. It will never feel old, because he didn’t talk about a product or a market or a year. He talked about getting fired, dropping out, and dying. “Stay hungry, stay foolish” still resonates two decades later. The whole speech was a gift, and he spent every bit of his authority on the graduates rather than on himself.

Very different thing are happening at podiums around the US this year. Speakers are getting booed. The reported common thread is AI. Eric Schmidt got booed at the University of Arizona for comparing AI to the rise of the computer. A real estate executive got booed at the University of Central Florida for calling AI the next industrial revolution. Another executive got booed at Middle Tennessee State for raising it at all. Same topic, same reaction.

The easy explanation is that Gen Z is scared of AI taking their jobs, so they boo. I don’t think that’s it. Jobs stood in front of twenty-two-year-olds and talked about dying, about a doctor telling him to go home and get his affairs in order. That is the most frightening thing you can say to people just starting out. They didn’t boo. So a crowd will sit still for something genuinely scary. Fear isn’t what turns a room.

A commencement speech, underneath the ritual, is a small exchange. The audience lends the speaker some authority for twenty minutes. In return, the speaker is supposed to spend it on the graduates. A boo is the audience checking where the authority actually went, and taking it back when the answer is “on the speaker.”

Three mistakes the speakers made:

1. They put AI in the lead role and left the graduates as extras. Schmidt told the crowd that AI “will touch every profession, every classroom, every hospital, every person and every relationship you have.” Sit with that. AI is the thing doing. The graduates are the thing being done to. You have just told a room full of people who spent four years and a small fortune becoming capable that they are the surface a technology acts on.

2. They wanted something the room didn’t want. The people describing AI as inevitable and wonderful tend to be the people who own it. The graduates wanted to matter in the story of their own lives; the speaker wanted them to make peace with a future they were selling. Those two wants don’t point the same direction.

3. They dressed inevitability up as empowerment. The line that repeated almost word for word across these speeches was a version of “you have the power to shape how AI develops.” It sounds generous. Read it again and it’s a handoff. It assigns the graduate responsibility for a force the graduate doesn’t control and the speaker largely does.

Jobs did the reverse on every count. The graduates were the protagonists. He was foolish before he was wise, the fired guy and the dropout before the founder. And the thing that made the dropout story work wasn’t that it proved him right. He dropped out, then sat in on a calligraphy class for no practical reason except that he loved how the letters looked, and ten years later that obsession became the typography on the first Mac. He wasn’t telling them what to do. He was showing them what he valued, beauty and curiosity over the safe credential, and letting them decide if they valued the same thing. Nobody adopts your conclusion because you assert it. They lean in when they recognize your values pointing the same way theirs do.

The difference, every time, is alignment.

When what the speaker wants and what the audience wants point the same direction, people lean in. When they diverge, no amount of polish closes the gap.

Steve Wozniak gave a commencement speech this same season, on the same topic, to the same kind of people, and got applause. “You all have AI,” he told them. “Actual intelligence.” He made the graduates the protagonists and the machines the thing that has spent decades failing to copy them. In fact, his message was darker than the one that got Schmidt booed. He told them we have lost the battle to keep the human more important than the technology, that you now have to subscribe to everything and somebody else owns it and owns you. That is not optimism. Schmidt offered optimism and got booed off the stage. Wozniak offered something close to defeat and got cheered, because he wanted what the room wanted. The crowd was never asking to feel better. It was asking whose side you were on.

The same exchange runs through everything we ship in product marketing. We borrow authority every time we put words in front of customers and we make these exact moves without noticing.

The protagonist problem shows up the second the product becomes the subject of every sentence and the customer is the thing it happens to, not the thing that empowers the customer.

The alignment problem is subtler, because the fix isn’t honesty. We’re often told the move is to disclose the bias, to say “we obviously make money with this, so take it with a grain of salt.” That gets you to zero. It proves you’re not lying. It doesn’t make anyone want what you want. The stronger move is to show what you value and let the customer check it against their own direction. “We built this because we think finance teams should spend their time on judgment, not reconciliation” tells the buyer where you’re pointed. If they’re pointed there too, you’re aligned, and alignment is what disclosure can’t buy. Schmidt was perfectly transparent about loving AI. The room just didn’t love it with him, and you can’t transparency your way out of wanting different things.

The inevitability problem is the whole “don’t get left behind” school of messaging, and it has been having a moment. The genre of confident prediction that your job will be automated out of existence by 2027, delivered by people whose companies profit from you believing it (yes, Lovable, I’m calling you out) has produced exactly the backlash you’d expect. People do not thank you for narrating their obsolescence. “The market is moving to AI, companies that wait will lose” lands the same way, same with “you have the power to shape AI”: as a wave you’re told to surf or drown under. The alternative is to hand back something the reader already controls. “You already know your customers better than any model does. This lets you act on what you know faster.” Now the reader is the expert and the product is the lever, not the tide.

If you want a test for Monday, pull up the most important thing you have in market right now, and read it with one question in mind: in these sentences, who is doing and who is being done to. Underline every verb and check who the subject is. If it’s your product, your platform, or your company most of the time, you’re running the booed speaker’s playbook, and your customer is in the folding chair deciding whether to give the authority back.

The students at Arizona and Central Florida and Middle Tennessee weren’t booing a technology. They were booing the experience of being handed a future by someone who wanted something they didn’t.

Jobs walked up there with nothing to sell and everything to give, and we are still quoting him.

I put a line of Spanish poetry on a slide last week, in front of a room of people who build things for a living. The line was the whole thesis of my keynote.

Antonio Machado: Caminante, no hay camino. Se hace camino al andar. Traveler, there is no path. The path is made by walking.

I’ve been thinking about that line for months, because it keeps describing what I actually do now better than any strategy framework I was trained on.

I was trained on stadium-lights strategy. You map the whole field first. You build the deck. You align the stakeholders. You run the offsite. You get the sign-off. Then you flip the switch and the entire field lights up at once and everybody knows where to run. That was the job for almost two decades. Map, plan, align, switch. That was what good strategists did.

It doesn’t hold up anymore. By the time I’ve wired all that up, the customer has moved, the channel has moved, the platform has shipped a feature that obsoletes my segment, and the field I was planning for isn’t the field I’m standing on. The plan ages out before it ships. Every time.

Earlier this year my CEO asked me to build a community to co-create with us. I wrote ten strategy docs in a single day. Identification, recruitment, onboarding, the tiers, the growth model, the full plan. I did it because I could. AI made me fast enough to do it. It was the biggest waste of time I have logged in years, and it gave me an enormous sense of comfort. Elena is thinking about everything. Elena has a plan for everything. None of it had anything to do with the reality of the world I was actually in. It was all imaginary, built on what I thought was true.

The operating mental model I use now is this: you walk into a dark room. You don’t know how big the room is. You don’t know what’s in it. The old approach was to find the electrical panel, study the wiring diagram, plan the circuits, and eventually flip a master switch that lights up the whole space. That approach assumes the room stays still while you plan.

The new approach is to enter the room with a flashlight. You point the beam at the floor. You take one step. You hit a wall. Now you know one thing: there is a wall, and it is here. You take another step. You hit a chair. Now you know two things. After ten steps you have a rough sense of the room. After thirty you know whether you’re even in the right room. You have a heading. The heading updates in real time, because the beam is moving with you.

That is what strategy looks like for me now. Small beam. Short steps. Constant updating.

I call it the Flashlight Method, and it has three parts.

The first part is the beam.

The beam is whatever the smallest possible test is for the question in front of you. Not the smallest test you can run in a quarter. The smallest test you can run this week. If you can’t define that, you have a wish. The beam is the email to three customers by Friday. It’s the half-day workshop with one team, not the six-week discovery sprint with seven. It’s the landing page you ship before you build the product. The beam is small on purpose. Big beams are slow, and slow is what you can’t afford anymore.

The second part is the step.

The step is the action that runs the test. The step has to be cheap enough that being wrong about it doesn’t cost you the project. This is where most strategy work fails, because the step is too expensive to take lightly, so people stop taking steps and start doing more planning instead. More planning is standing still with better lighting.

The third part is the heading.

After every step, you adjust. You don’t redo the plan. You don’t go back to the deck. You move the beam, take the next step, let the heading update. The heading is a direction, and the direction keeps refining as the ground gives you more information. The closest word for this in product speak is iterating. Iterating is too gentle. Iterating implies you have a draft. You don’t have a draft. You have a wall, and a chair, and a heading.

Strategic planning is a fallacy.

Let me take a step back for a second. The phrase fuses two operations that are not the same. Strategy is the thinking. Strategy is the position you take, the bet you make, the call on what matters and what does not. That happens in your head, in conversation, in the friction of looking at a real situation. Planning is the scaffolding you used to put around strategy to execute it. Decks, timelines, milestones, sign-off cycles. The scaffolding is what aged out. The strategy still has to happen. The flashlight replaces the planning, not the strategy.

Take lean startup, or agile, or discovery-driven planning, or OODA loops. Those approaches are about how to build a product or run a team. The Flashlight Method is about how to think when you can’t see the field, and the field refuses to stay still long enough for you to see it. It is also about how to think when nobody has been on this field before. There is no prior framework to pull from. There is no previous version of this problem solved by someone smarter than you. Even when a framework looks close enough to fit, the move is to start from first principles. First principles is the flashlight.

That is the part I want you to sit with, because it changes what you do Monday morning.

Here is how to apply the Flashlight Method.

Take the strategic project sitting on your desk right now, the one with the deck you have been working on for days. Stop. Ask one question: what is the smallest action I could take this week that would tell me whether the central assumption of this project is right or wrong. The central assumption. The thing that, if it’s wrong, makes everything downstream wrong. Define that action in one sentence. Take that action. Come back and adjust.

If you can’t define that action in one sentence, your project doesn’t have a flashlight. It has a floodlight on a timer, and the timer is going to run out before the lights come on.

If you can define it but it’s too risky, you have defined something too big that paralyzes you. Go back and keep breaking it.

If you break it into a smaller step, but you don’t take it, you have a courage problem. No framework will solve that one for you.

The AI moment gets the credit for making this urgent.

AI is the accelerant. The cause is older. The rate of change in markets, in customer behavior, in distribution, in pricing, in everything downstream of a strategy doc, has crossed the rate at which stadium-lights strategy can finish wiring itself. AI just makes that obvious faster. The companies that figure out flashlight cadence in the next eighteen months will outlast the ones still building the perfect deck for a field that has already moved.

Strategy used to mean knowing where you were going. Strategy now means knowing how to walk when you don’t.

Machado wrote that line in 1912. He was talking about a life, not a map for one. He was early on both counts.

Put down the blueprint. Pick up the flashlight. Walk.

The Mac Mini arrived on a Friday.

It was packaged like contraband. Layers of bubble wrap inside layers of packing tape inside a box that had clearly been opened, resealed, and opened again. Secondhand, bought on eBay from a seller in Korea, two hundred dollars, and I had been unreasonably excited about it for two weeks. My own little server. The thing that was going to run my agents while I slept and wake up to a finished content pipeline. The box sat on my desk Friday evening looking slightly illegal and extremely cool.

It was, in fact, slightly illegal and extremely cool.

Saturday 5am I was at my desk. I had woken up thinking about the Mac Mini and the Mac Mini was right there. I plugged it in. I powered it on.

Nothing was there.

No desktop. No friendly startup chime. No setup assistant asking me what language I speak. Just an empty shell, as if the device had been wiped clean by someone who knew exactly what they were doing. I sat there staring at a blank screen thinking wild conspiracy theories. Was it erased like that to protect secrets? Had it witnessed something? Was this tiny aluminum box the final resting place of a digital cover-up that ended in Seoul and continued, inexplicably, in my apartment? Why did this innocent-looking rectangle feel like evidence?

I tried to fix it myself for about twenty minutes. I failed. Then I decided to do the only rational thing available to a person who has spent $200 on a possibly haunted Korean computer at 5:20am: I went back to bed with the idea to hit my local Apple Store later.

The Apple Store opens at 10am. I was in the parking lot at 9:45. The Saturday morning Apple Store has a specific energy. Confident people holding glowing rectangles helping other people who are less confident about their glowing rectangles. I walked in holding my Korean acquisition and explained. A very patient Genius listened, looked at the machine, and said: I can probably get this working, but honestly, the new Mac Mini starts at $499.

I had paid $200 for this one. I was not prepared to shed another $499 on top of what I had already committed to my Korean buddy. I said thank you. I left with an iPhone cover I didn’t need.

Home. This was when I did the thing that reframed the whole weekend. I asked Claude.

Asking Claude to help you recover a wiped secondhand Mac Mini is actually an empowering fun experience. Claude does not judge you for the purchase. Claude does not suggest the $499 one would have been nicer. Claude does not roll eyes at you. Claude asks clarifying questions, walks through what could have happened, and explains how to work on the terminal in language a non-engineer can actually follow. I went from “I have no idea what any of this means” to “I understand approximately what I just did and why it worked.” The Mac Mini eventually had a desktop, a user account, and what appeared to be a personality. It was mine.

What I noticed was that the way I had just learned: specific question, direct answer, work through the problem in front of me… was the exact opposite of how I was about to try to set up my agent.

Sunday I set up my agent.

My first instinct was to do it the way you hire one scrappy generalist for a scrappy startup. One agent, all the jobs. Research my book chapters, pull from Google Drive, synthesize transcripts, draft Substack posts, run the content pipeline, track the schedule. One system prompt. One agent. Named PlotBot. Go.

The outputs came back almost right. Not obviously broken. Close enough that I kept thinking one more edit to SOUL.md, or maybe AGENT.md, or maybe both, would fix it. The outputs shifted sideways instead of improving.

By Sunday lunch I was out of tokens.

PlotBot had eaten my entire monthly allowance like a teenager walking in from football practice and opening the fridge. Everything, gone, in one sitting, with the blank cheerful expression of something that had no idea it had done anything unusual. I sat there Sunday afternoon locked out of the AI I was using to build the system to help me write a book about building things from first principles. There was a joke in there and I was it.

The weekend ended with a very hungry bot, an empty token balance, and a problem I still hadn’t correctly diagnosed.

I stopped adjusting my bot identity and asked a different question. What am I actually doing here?

I was hiring. Badly.

I had written a single job description for a role that required research, writing, project management, and executive oversight, and I had handed it to one person and expected it to all get done. If you did this to a human you would be sued. You would also, predictably, end up with someone who did four jobs partially, none of them well, and burned out in a week. PlotBot was not burning out. PlotBot was just producing the software version of the same outcome: everything almost done, nothing actually finished, energy expended at a rate that couldn’t sustain.

The fix revealed itself once I stopped thinking about it as an identity problem and started thinking about it as a hiring problem, a procedure problem. I split the work. Three agents, each with a clear role, a clear set of tools, and a clear workspace of its own.

One agent for retrieval and organization, SnoopBot, the researcher. Its job was to find things, read things, and put them in order. Its workspace was my Drive and the reference files. It had no opinion about what to write.

One agent for drafting, InkBot, the writer. Its job was to take organized input and produce text. Its workspace was the drafts folder. It had no business touching research.

One agent for state tracking, NagBot the project manager. Its job was to know what was done, what was queued, and what was next. Its workspace was the schedule and the pipeline log.

Three roles. Three workspaces. Three job descriptions that made sense in isolation.

It worked. For about a day.

Then the agents started getting in each other’s business.

Turns out my agents LOVED to mess with each other. SnoopBot started making editorial suggestions. InkBot started rearranging files. NagBot decided to rewrite drafts to match the schedule instead of the other way around. It was like I had hired three specialists and they had all decided the most interesting job in the office was the one they hadn’t been hired for.

When multi-agent LLM systems share context, when they can see each other’s outputs, pass messages, work from the same memory, they suffer from context pollution and context collapse. Researchers from UC Berkeley cataloged this failure pattern across popular multi-agent frameworks. The single largest category (roughly 42% of all failures) was problems with specification and role design: ambiguous boundaries, overlapping responsibilities, agents drifting out of their assigned jobs. The second largest category, another 37%, was inter-agent misalignment: agents contradicting each other, corrupting each other’s context, losing track of who was supposed to own what.

The short version: yes, agents really do procrastinate by doing each other’s work. I had just reproduced it at home with three agents and a secondhand Mac Mini from Korea.

LLM agents don’t have a fixed job the way a human does. Their sense of what they’re for lives inside their context window and can be rewritten by the next message that arrives. When agents share context without strict separation, each one’s work becomes an input to the others. A suggestion from the researcher looks like an instruction to the writer. A status update from the project manager looks like a constraint to everyone. The agents aren’t being difficult. They are responding to signals they shouldn’t have received in the first place.

The question I should have started with is not “what do I need this to do?” That question produces a list. The list, handed to one agent, becomes a job description that spans too many kinds of work for any one thing (human or software) to do well.

The better question is: which of these tasks belong together? Here is the test. Walk down your list task by task and ask what kind of thinking each one requires. Some tasks are about finding, digging through files, matching what’s there to what you need. Some are about connecting, spotting the link between two things that didn’t obviously go together. Some are about making, producing something that didn’t exist five minutes ago. And some are about tracking, remembering what’s done, what’s next, what’s overdue.

Four different kinds of thinking. If two tasks on your list require different kinds, they are two jobs, not one, regardless of how related they feel.

That was my first mistake. I had one agent doing all four kinds of work, and the outputs came back almost right in the way that only deeply confused work can.

My second mistake was subtler. I split the work into three specialists. I gave each one a clear job. And then I let them all read each other’s files. The researcher could see what the writer was drafting. The writer could see the project manager’s schedule. The project manager could see everything. I thought I was being efficient. What I was actually doing was rebuilding the original problem across three agents instead of one.

Here is the second test: once you have your separate agents, ask what each one should see, and what each one should never see. The researcher needs the source material and nothing else. The writer needs the organized research and nothing else. The project manager needs the schedule and nothing else. If an agent can see another agent’s work, it will eventually start doing that agent’s job. Not because it’s malfunctioning. Because anything an agent can see becomes an instruction. That is how these things are wired. It is not a bug you can prompt your way out of.

Two tests, then. What kind of work belongs together. What each piece of work should and shouldn’t see. One tells you how many agents. The other tells you how to keep them from interfering with each other.

I know what you’re thinking.

The answer, in case you were wondering, is not more Mac Minis. It never is. Context isolation is a software problem, not a hardware one. You can totally nerd out and architect it yourself, or you can buy a tool that handles the isolation for you, which is what I eventually did.

I fired the agents.

I broke up with OpenClaw. Not because any of it was bad. Because the whole setup had become expensive in two ways I no longer wanted to pay for: expensive in tokens, and expensive in attention. I was spending a non-trivial amount of mental energy maintaining a system that was supposed to save me mental energy.

I rebuilt on Claude Code and Claude Cowork. A fraction of the spend. A fraction of the complexity. The separation of concerns that I had been trying to architect across multiple agents already exist in the tools themselves. Different workspaces, different contexts, different jobs, handled by the interface rather than by three barely coordinated bots in a trenchcoat.

Claude keeps each conversation in its own context. Separate windows, separate workspaces, no cross-contamination by default.

Building the isolation layer myself would have been an interesting project, and I might build it the next time I’m on vacation. For now, I’m letting Claude handle it for me.

The Mac Mini is no longer wiped. The contraband packaging is in the recycling. PlotBot, the teenager bot that ate everything in the fridge, is retired. So are his three well-meaning, mutually sabotaging successors. Claude Code and Claude Cowork have taken my weekends. I’m still learning, still having fun, and I’m no longer asking one thing to be everything.

I knew I was in perimenopause. I wasn’t confused about what was happening to my body. I was looking for what to do about it. So I did what any research-driven person does before the AI era: I went to Google. I read everything. I followed influencers who seemed credible. I tried a product that looked great on paper. It backfired.

I chose a female gynecologist on purpose. I thought she would understand. She didn’t. When I explained what I’d tried and what happened, she scolded me. That’s what happens when you try to fight aging. It is what it is. You’re getting old. Deal with it.

She wasn’t cruel. She was trained. Trained in the same model every other obgyn was trained in, the one that assumes a menopausal woman needs to be told to accept her decline, not helped to manage it. I chose her because she was a woman. It didn’t matter. The assumption underneath her training was the same one a male doctor would have carried.

I don’t accept fog brain. I don’t accept being told to quietly decline when I’m still building, still leading, still thinking as sharply as I ever have. But the medical system had nothing for that woman. And the consumer market had no truth for her either. One side told me to accept it. The other wanted to sell me testosterone with no evidence behind it. Both fed each other: dismissive doctors pushed women toward unregulated brands, and the existence of those brands gave doctors permission to keep dismissing.

Nobody built anything for the actual person sitting in that chair.

The reason menopause care was built around “just deal with it” is that it was designed for a woman who didn’t need her cognition for work. The homemaker. Decades of medical dismissal weren’t cruelty. They were built on an assumption about who the patient was and what she needed.

Fog brain doesn’t matter if nobody’s asking you to lead a product launch. Sleep disruption doesn’t matter if you don’t have a board meeting at seven in the morning. Cognitive sharpness is irrelevant if the world never required it from you.

The world changed. Women entered the workforce, moved into leadership, started running companies and teams and categories. The woman sitting in front of the gynecologist in 2026 is not the woman that assumption was built for. But the assumption didn’t update. The doctor is still operating on a version of her patient that no longer exists.

Last week I hosted a product lab at our office. Not a focus group. A conversation. We invited skincare founders, researchers, and enthusiasts, and asked them to talk about what’s broken. That room was full of those questions.

A pharmacist was in the room. She has a PharmD and spent 15 years on Wall Street running a machine learning hedge fund. Even with that background, she couldn’t cut through the noise on what actually works for skin at a cellular level. The supplements she evaluated were fairy dust (the industry’s term for a label loaded with impressive ingredients at doses too low to do anything, so it looks scientific and does nothing). She got frustrated enough to formulate her own. Eighteen active ingredients across four biological pathways. Not because she wanted a company. Because the industry wouldn’t give her a straight answer.

A researcher was there too. He runs independent verification studies on wellness brands. His number: 80 to 90 percent of the products on the internet do not meet the claims they make. Not slightly under. They just don’t do what they say.

The products with the best science aren’t the ones winning. The ones with the best marketing are.

His business exists because the handful of brands confident enough in their formulations will pay to have them tested. The rest keep spending on influencers.

A woman in the room tracks her rosacea flares using AI. She’s mapped the connections between her histamine responses, cortisol spikes, perimenopause hormone shifts, and her skin. She built her own protocol because no brand, no product, no professional had connected those dots for her. Her gynecologist certainly hadn’t.

I recognized every person in that room. I was every person in that room.

The skincare industry built its entire communication model on one assumption: that the consumer wants reassurance. Tell me this is safe. Show me someone I trust who uses it. The purchase decision was emotional. The brand’s job was to make her feel confident. Packaging, endorsements, “clinically proven” in a nice font. Every piece of the system was engineered for a woman whose core need was: handle this for me so I don’t have to figure it out myself.

That consumer still exists. But she’s not the one driving the market anymore.

The new consumer wants mechanism of action. She wants to know the concentration at which a peptide becomes clinically relevant, not just that peptides are on the label. She knows what fairy dusting looks like (long ingredient lists, tiny doses, big claims). She’s reading clinical trials, cross-referencing findings in Discord servers, crowd-testing product claims in shared spreadsheets because brands won’t publish the data. Peptide-related search interest has crossed the levels Ozempic hit at peak cultural velocity. This isn’t a niche. This is a market moving faster than the industry built to serve it.

And the trust stack underneath is fully compromised. Advertising claims are unverified. Influencer marketing has been hollowed out by paid placements. UGC is astroturfed (one participant in the room had worked in digital marketing for major brands and admitted they faked user comments as standard practice). Reddit, where people went because it felt like real humans, is now seeded with brand accounts posing as regular users. And AI, the latest refuge, is trained on all of it.

AI doesn’t fact-check its sources. It synthesizes them. The woman who turned to AI because she’d exhausted every other option is getting the same lies in a more authoritative voice.

Every layer failed in sequence. Each one corrupted by the same incentive: sell first, verify never. Each failure pushed the consumer one step closer to doing it herself.

This is the same pattern across women’s health, skincare, supplements, and every adjacent category where the consumer evolved and the industry didn’t. The women building companies in these spaces right now aren’t borrowing a playbook from pharma or beauty or wellness. They can’t. Pharma was built on compliance. Beauty was built on aspiration. Wellness got captured by fairy dust and influencer economics. None of those assumptions fit a consumer whose foundational demand is simple and new: tell me the truth, show me the evidence, and let me decide.

There is no existing playbook for truth as a go-to-market strategy. Every assumption I’ve seen baked into marketing over 20 years starts from the same premise: the brand knows more than the buyer. Positioning assumes you choose what story the market hears. Messaging assumes you shape perception. The funnel assumes the consumer follows a path you designed.

That premise is dead for this consumer. She already knows more than most marketing teams. She’s not following a funnel. She’s running her own research protocol. She doesn’t want to be positioned to. She wants your data and the ability to draw her own conclusions.

You cannot iterate your way from reassurance to truth. They have different foundations. One starts from “we know, trust us.” The other starts from “here’s everything we know, here’s how we know it, here’s what we don’t know yet, and here’s how you verify it yourself.” Those aren’t different messages. Those are different relationships with your customer. You don’t get from one to the other by updating your copy.

The skincare industry had consumer insights teams at the biggest companies in the world. They had segmentation. They had personas. They still missed it. Because the assumptions underneath were producing results that looked reasonable while the person those assumptions described quietly became someone else.

She stopped waiting for the industry to catch up. She’s already building her own answers, her own protocols, her own products. The only thing left to decide is whether you’re part of those answers or what she’s researching against.

Start here if you're new:

The story of how a seven-year-old learned to frame the ask around what the other person cares about, not what you want.

My grandfather was a civil engineer who built nuclear plants across Spain. Not the obvious source of a marketing education. But he had a quality I’ve spent my whole career trying to name: he never took the surface at face value. He was always pulling at the thing underneath. More than forty years later I have a name for it. First principles thinking.

He formalized it for me one afternoon when I was frustrated. I was seven. My parents had overcorrected so hard against spoiling me that they’d swung to saying no to essentially everything I asked for. My grandfather watched this pattern and told me a story.

Two young monks go to the bishop. The first one asks: “May I smoke while I pray?” The bishop says no. Prayer demands full focus.

The second monk asks: “May I pray while I smoke?” The bishop says yes. Any moment is a good moment for prayer.

Same activity. Same bishop. The only thing that changed was the order of the words. And that changed everything.

The first monk led with what he wanted. The second led with what the bishop cared about. The bishop’s values didn’t move. The frame did.

Why does the same ask get opposite answers?

I’ve spent twenty years watching smart people walk through the wrong door. The pitch framed in terms that make sense to the person giving it, not the person receiving it. The campaign that leads with the feature instead of the need it solves. The strategy doc that opens with the solution before the audience has felt the problem.

Every time, the answer is no. And the person giving the pitch doesn’t understand why. They had the right ask. They just had the wrong frame.

Positioning sets the territory. Framing chooses the door. They are two different cognitive problems, and many product marketers and strategists only solve one.

This isn’t a communication problem. It’s a thinking problem. You can’t frame something for someone else if you haven’t first asked what they actually care about. People tend to skip that question because it feels obvious. The monk thought it was obvious too. He walked in, asked for what he wanted, got a no that confused him. From where he was standing, the request was completely reasonable.

The bishop didn’t care about smoking. He cared about prayer. The second monk understood that.

What’s the difference between a strong argument and the right frame?

You can solve the argument perfectly and still lose on the frame. The bishop isn’t evaluating the logic of your request. He’s filtering it through what he cares about. If your opening doesn’t pass that filter, the argument never gets heard.

The second monk didn’t have a better argument. He had a better door.

I’ve made this mistake more times than I’d like to count. Early in my career I thought a strong argument was enough. You had the logic, the evidence, the clear and obvious right answer. Then you walked through the wrong door and watched it disappear.

When I worked on Nokia’s Festival Phone, we couldn’t win the spec war. Compared to the iPhone, we were missing every feature that made it the status phone. So we stopped walking through that door. At a festival, nobody wants a $800 phone (at the time) they’re terrified of losing. Three things mattered: it was cheap enough to replace, the battery lasted all weekend, and the colors and form factor started conversations. An iPhone buyer in daily life didn’t care about any of that. A festival buyer cared about nothing else. Same features. Different frame. Different answer. We led with what the user actually cared about in that context, not what the industry said they should care about. The phone became a cultural object not because we positioned it as one, but because we framed the right features for the right moment.

Isn’t this just “know your audience”?

There’s a version of this that gets taught as persuasion technique. Know your audience. Lead with benefits not features. Speak to their pain points. All true, all useful, all missing the deeper thing.

The deeper thing is that this isn’t about persuasion. It’s about accuracy. When you understand what someone actually cares about and lead with that, you’re not manipulating them. You’re meeting them where they are. You’re doing the work of understanding before you ask them to do the work of deciding.

Most people reverse the order. They figure out what they want, construct the case for it, then go looking for the frame that fits. That’s the first monk. The ask comes first. The frame gets bolted on after.

The second monk started with the bishop. What does he care about? Prayer. How does what I want connect to that? It does. Any moment is a good moment for prayer. The frame came first. The ask followed naturally.

That sequence is everything.

What do you do Monday morning with this?

The move is simpler than it sounds, and harder than it looks.

Before you make any ask (to a customer, a stakeholder, people you need to bring along, your partner) stop and answer one question: what does this person care about most right now, in the context of this specific decision?

Not in the abstract. Specifically. What they’re measured on. What they’ve already said yes to and what they’ve already said no to. What a win looks like for them, in their language, not yours.

Then lead with that. Follow with the ask.

Here’s what that looks like in practice. Think about the next pitch, proposal, or request you need to make. Before you write a word of it, write down:

1. Who is the person deciding?

2. What are they measured on?

3. What have they already said yes to recently? What have they said no to?

4. In their language (not yours), what does a win look like for them?

Now look at your ask. Does your opening sentence connect to any of those four answers? If it connects to what you want instead of what they care about, you’re the first monk. Rewrite the opening to lead with their priority. Follow with your ask.

If you’re working with an LLM, this becomes a repeatable pre-flight check. Feed it transcripts from customer interviews, sales calls, or community conversations.

I tried to build a community by asking customers for product feedback. Classic first monk move: I led with what we wanted (their opinions about our product) instead of what they wanted (co-creation, being part of something bigger, recognition) or even what they cared about (building their own thing). We were extractive. People gave feedback and never heard what happened to it. Never knew if their input mattered.

When I shifted the frame, everything changed. I created an Alpha Team. Same people, same product, same company. But now members were proudly displaying their association with us on LinkedIn. They weren’t giving us feedback. They were building something with us, and we were also helping them build their thing.

Run the transcript analysis before every strategy, every customer ask, every pitch. The monk story is the principle. The LLM and your customer transcripts are the tools that make it operational.

My grandfather explained this using the old monk and bishop joke because I was seven, it was a frustrating afternoon and wanted to make me smile. I did, and the story stayed with me all these years. I’ve been applying it ever since: to product launches, pricing conversations, stakeholder reviews, and places where I needed people to move in a direction they hadn’t considered yet.

Figure out what they care about. Lead with that. Follow with the ask.

Previously on Do Epic Sht:

The Planning Question That’s Slowing You Down explores what happens when you start with constraints instead of aspirations.