Free live session: Why your offer needs MCP

Your clients already talk to Claude, ChatGPT or Codex every day. On Tuesday 1 December at 17:30 CET I’ll show you how I’m building my own Unstuck MCP, and how you can turn your own framework or know-how into an MCP people use right inside their AI chat. It’s free, about 60 minutes on Google Meet, with live Q&A at the end.

Save Your Spot Here

Speaking of, a great offer needs great positioning, which is the actual topic of today’s edition.

If you sell what you know, your profile starts selling before you’re in the room. 1,371 people viewed my LinkedIn profile in the past 90 days. That page has to explain me when I’m not there.

So this weekend I rebranded all of it with Claude Code: LinkedIn, X, Instagram, YouTube and Substack. New line, new bios, new banners. It took about 30 messages, and a lot of them were me saying no.

By the end of this edition you’ll have the 3 parts of a profile people can repeat, a way to write your banner line, the 5 tests that killed my first drafts, and the free skill that runs the same process on your own profile.

My first drafts were boring

I asked Claude Code for one statement I could use everywhere.

My reply to the first version was: not clear, not catchy. So I pulled up the profile of Chris Do, the brand strategist, and asked how I could be him, but for AI.

His line is “Success requires all of you. I’ll make the introductions.” A truth, then a turn. We copied the shape of his whole profile, and soon my banner sounded like Chris Do with my name on it. One title Claude gave me was “Reformed tool hoarder”, a spin on his “Reformed introvert”. I didn’t know what it meant, so a client wouldn’t either.

What fixed it was my own story. In 2023 I got rejected from 50+ jobs.

My English would collapse mid-sentence in interviews, and I could see it in their eyes that I wasn’t the one.

So I started posting on LinkedIn to practise writing. Claude found my old post about it and connected 2 things: I got rejected from jobs, and today I fire business owners from the wrong jobs.

So the whole brand could live on one word: jobs. Overqualified is what a company says when it turns you down. A handover is what you do when you leave a job. Fired is how a job ends.

Put those words on the reader’s day and you get the line:

You’re overqualified for most of your day. AI isn’t. I’ll do the handover.

Same shape as Chris: a truth, then a turn. The words come from my story. Copy the structure, never the words.

Because you need something that is not sounding like 10000 other people, such as “I help [X] to do [Y]”

Think about how many people sound the same online, and think about how many people now sound like me.

Read more

At Kiss The Future in Sarajevo, one demo answered a question I’d been chewing on for weeks.

Pedro Rodrigues opened a chat and asked about Sarajevo. A tour guide showed up inside the conversation. Maps, photos, places to pick from. He never opened an…

Read more

Before we start with the main topic, here are a few updates from last week:

1. I launched my Unstuck network (free & paid tiers) and the recent system I built was a video editing system. You can find more details below:

The Unstuck Community is free, where you can find other systems as well, but in the paid Unstuck Academy you have this full video editing system.

Join the Unstuck Academy here

2. New podcast episode drop:

   
   I sat down with Reyhan Merekar, who puts AI into big companies for a living (founding forward deployed engineer at a venture-backed AI firm in New York, ex-Accenture), to ask how an enterprise AI pilot goes from first call to production.

   He says a high-return pilot can get a yes in 1 to 2 weeks, not the 3 to 6 months most of us assume, and he walks through data privacy, guardrails and what to check before you build your own MCP server.

   In your case, if you sell AI to small and medium businesses and want to move up to enterprise, you get a simple path: start in the mid market, have more problem-first conversations, and win one problem at a time. Here is the full edition here:

Subscribe now

Now to the main topic…

I opened Google Sheets with 7,819 LinkedIn connections and told Claude: “Don’t do anything; just tell me the possibilities.”

I wanted a list of people worth a closer look. First, I asked what the data could tell us.

Ever scroll through your connections and still not know where to start? That was my problem.

The first batch now has 69 Qualified, 51 Maybe and 84 Not fit. Here is how I got there, and the skill you can use too.

1. I asked what the sheet could tell us

My opening question was:

“Tell me with the current data we have on this sheet what type of data manipulation can we do when it comes to creating the organized list of potential prospects that fits my offer?”

“Don’t do anything; just tell me the possibilities.”

The export had names, profile links, emails, companies, job titles and connection dates.

Of 7,819 people, 7,613 had a job title. Only 303 had an email. Another 206 had no title or company.

So I started with job titles. The email list had too many gaps.

There was no location, team size or income data. “Founder” gave me a reason to check, but did not prove they needed help.

My first question in the Claude sidebar. I asked for possibilities before any edits.

2. I separated filtering from paid research

I asked if we could use Trigify to check the people we kept. Once connected, I gave Claude this order:

“We need to use first cheap filtering with the data we have. Once we have winners we will do enrichment to do final qualification with more data by scraping their profiles, company pages and recent activity.”

Enrichment means finding missing data. It costs credits, so I wanted a smaller list first.

I called them “winners,” but they still needed research. The plan was simple: filter, test a small group, then decide what was worth paying for.

A note on the tool. I used Trigify to spend my leftover credits. Its team is joining HubSpot, and the tool is closing. Most accounts and API access end at 11:59pm BST on 22 October 2026. Some contracts differ. Trigify’s notice.

You can use Apify or Bright Data to pull LinkedIn data. My friend Walid also just launched looot. It links your AI to data tools and compares their prices. Check its API if you want to cut research costs.

Test the data and price first. Profiles and posts are separate jobs. The skill now checks which tool can do each job through an API or MCP connection. It does not need Trigify.

3. I supplied the business decision

Then I explained who fits my work.

Some owners want to learn to build with AI. Others want someone to find repeat tasks and build for them.

I named two groups: solo coaches, consultants and advisors; and owners of agencies, service firms and local shops. This helped Claude sort them.

Tell Claude what “good prospects” means. If you leave that out, it will guess.

A title like “CEO & Coach” still cannot tell you their income, or if they want to build or buy help.

4. I asked to see the result before the build

This was my next instruction:

“I want to see what prompt you will create, and what outcome that prompt will give us after we run it. Don’t run the prompt.”

I asked for each group in its own sheet tab. Then I could see who was there, why they fit, and who to check next.

Connections kept the raw data. Rules held the words and scores. Scored showed the result for each row.

Self-build, Partner, Dual-track, Review queue and Peers each got a tab. Dashboard held the counts. Batch 1 kept a fixed list for research.

A fixed batch keeps paid research tied to the right person, even if rules change. Each result keeps its profile link.

5. I challenged the first answer

Before saying yes, I asked:

“Do you think we forget something? If so, make sure to double check.”

This caught titles in other languages, vague words, missing data and people who fit two groups.

“Product owner” has the word “owner,” but is often a staff role. The Rules tab now checks for false matches like this.

The Rules tab during the build: owner words, employee titles and false matches are separate.

Claude claimed it found 146 repeat profile links. My check found no exact repeats. I am not counting those as removed rows.

The rules also included healthcare firms, which I do not target. I still need to check the rules against my offer.

A recent connection date does not mean we have talked or they want to buy.

LinkedIn connection filtering: what the first pass produced

I approved it with: “yes go and build it this way.”

I checked the build. Claude restored names that went blank during cleanup. After I flagged #REF! errors, its next scan found none. We also checked which data was missing and which still needed research.

The filter found 867 Self-build, 802 Partner and 74 Dual-track candidates. That is 1,743 people, with 204 in Batch 1.

Of 7,819 rows, 7,672 have a profile link. The other 147 need review. They are not repeat rows.

Another 2,376 went to Review, 100 to Peers and 3,453 to Out. The dashboard says “Qualified,” but here that only means they passed the first filter.

Now I had a small group to test. Research would show where the filter was wrong.

6. I tested LinkedIn lead enrichment on 10 people

Before spending credits, I asked another question:

“I want to see the full enrichment process, what will it cost, what data will we have. Before you do anything, tell me your plan.”

All 10 test leads came from Partner. The test could only tell me about that group.

Trigify checked 10 profiles, 7 company pages and posts from 9 people. Claude saved each person’s final group, verdict and reason in Batch 1.

4 qualified. 5 maybe. 1 not fit.

After research, 3 moved to Self-build and 1 to Dual-track. Another sold a service like mine. Job titles had missed this.

Company data showed team size but no income. Claude’s income bands were guesses. Next time, I would leave them blank without a trusted source.

One owner wrote that work kept getting stuck with them. Another said their team built campaign groups by hand. These posts gave me a reason to ask a question. They did not prove a budget or a wish to hire me.

The first batch after research. Names and estimated revenue are blurred for privacy. The original group and the final verdict remain visible.

7. I finished the verdicts for the first batch

After the test, I told Claude to continue through Batch 1.

Claude checked profiles first and stopped paying for clear non-fits. Only the better matches got company and post checks.

Here is where the 204-person batch stands:

• 187 profiles researched.

• 17 obvious mismatches skipped, with a reason recorded and no profile lookup.

• 69 marked Qualified, meaning they fit the current criteria well enough for further review.

• 51 marked Maybe, because the available information did not settle the decision.

• 84 marked Not fit, including peers who may be better referral partners.

Every row now has a verdict. About 1 in 3 passed as Qualified. A founder title alone was not enough.

The fit rate is 36.9% of the 187 checked profiles, or 33.8% of all 204 leads. These are fit rates, not sales rates.

Current dashboard: 69 Qualified, 51 Maybe and 84 Not fit. The live Tier A count is now 202; the batch remains the 204 people frozen on October 12. This is why I kept a fixed batch.

Review, Peers and Out still need their own checks. I have not researched the whole network.

8. I continued in a fresh chat, with a clear handover

The first chat got too large while checking posts. Claude stopped to avoid mixing up rows.

I asked for a handover prompt so a new chat could pick up the work and use the skill.

It listed the sheet, fixed rows, work done, next steps and credits spent. It said: check each name and profile link, work in groups of 5, and stop at about 25.

The order was Qualified Partner/Dual-track, then Qualified Self-build, then Maybe Partner/Dual-track. Batch 2 and outreach stayed on hold.

Before switching chats, ask for a handover: what is done, where it is saved and what comes next.

25 more people got post checks over 14 days: 6 Strong, 3 Medium, 7 Weak and 9 None. Their fit verdicts stayed the same.

Each useful signal has a topic, rating, short quote, date and post link. I can check it before reaching out.

The batch now has 9 Strong ratings. These are Claude’s views, not proof someone wants to buy. A good fit may have no useful recent posts.

The fresh session checked 25 people, recorded its work and reported the cost.

Lead enrichment costs and the remaining research

The log shows 412 Trigify credits: 188 for profiles, 11 for companies and 213 for posts. The last run used 129, all on posts.

The profile logs differ by 1 credit, so 412 is the logged total. Claude’s plan costs extra. Apify, Bright Data and looot need their own price checks.

Posts drove up the cost. The first 20-person estimate was much higher than the later profile checks. “Cost per lead” means little unless you know what was checked.

The Dashboard’s “37 credits to finish” leaves out post checks. It is not the full budget.

All fit verdicts are done. Post research is still in progress.

51 people still need post checks: 5 Qualified Partner/Dual-track, 32 Qualified Self-build and 14 Maybe Partner/Dual-track. This is not the same group as the 51 Maybes.

I will finish those checks, review the calls and fix the competitor words before the next group. I also need to check past outreach. These sessions sent no messages.

The skill now keeps raw names if cleanup fails, checks profile links before writes, saves small batches and logs what was checked.

Where this is going

After the first batch, I asked Claude to step back:

“Can you recap what we did so far, and what are the next steps until the final destination? What will be the final outcome once we enrich and qualify all leads, and how will we achieve that?”

I wanted this sheet to help me start talks with the right people.

The goal is a ranked list I can check before reaching out. For each person, I want:

• A likely track: Self-build or Partner, with uncertain cases flagged for a conversation.

• A reason they fit, plus an intent rating and a dated quote with its source link when relevant evidence exists.

• A personalised opening question, in their language where appropriate.

• A checked outreach status, so I can see who is ready, who has been contacted and what happened next.

If there are no recent posts or useful signals, say so. A LinkedIn connection is not proof of a warm contact. An AI post is not proof they want to buy.

I also get a list of people who serve the same buyers and may send work my way.

How I will get from Batch 1 to the full list

First, close the first batch properly. Finish post checks, review Maybes and write opening questions. Mark Ready only after checking the facts, message and past outreach. Keep unclear cases in review.

Then, fix what the first batch taught us. 84 of 204 leads did not fit, about 41%. I will use those misses to fix Rules before the next batch. The saved batch stays fixed.

Repeat the same process on the remaining groups.

For Tier B, count and save a fixed batch. Check the cost, remove clear non-fits, then research profiles.

Save company and post checks for better fits. Check the Review queue’s titles and company names before paying.

For Tier C, decide whether to check again, research a few or leave it out. Every group needs a choice. Every row does not need paid research.

I still need to pick the tool, set a budget and decide if all Self-build leads need post checks.

Claude put the last 51 post checks at 250 to 270 Trigify credits. That only applies while my access works. A new tool needs a new price check. Skipping Self-build posts saves money but leaves me with profile data alone.

After my review, move approved prospects into outreach.

I plan to use Aimfox for LinkedIn campaigns, split by group and language. Start with the best evidence, review the messages, then approve the campaign. This skill does not launch it.

Finally, use the replies to improve the next round.

Track messages, replies and follow-ups in the sheet. Later, add cost per qualified lead and reply rates by group and signal. Today’s Dashboard only shows filtering and fit results.

Each new batch can learn from the bad matches and what people tell us.

How to install Claude in Google Sheets, Docs and Slides

Claude for Google Workspace launched on October 6, 2026. Its beta adds a sidebar in Docs, Sheets and Slides. In Sheets, it can edit formulas and add tabs. Official announcement.

It is available on Pro, Max, Team and Enterprise plans. One installation covers all 3 apps.

Install Claude from Google Workspace Marketplace. Reload the file. Go to Extensions → Claude → Open Claude and sign in. Your admin may need to allow it. Setup guide.

You can also work on Workspace files from Claude chat. I used the Sheets sidebar, with Trigify linked for this run.

Use the LinkedIn connections qualifier skill

I turned this process into a skill: linkedin-connections-qualifier.

A skill tells Claude what to ask, what to do, what to check and when to stop for your choice.

The original session saved the process as linkedin-connections-qualifier.

You can find it in the community under:

Free Resources → AI Skills → Sales → linkedin-connections-qualifier

Use it with your LinkedIn export in Google Sheets. Add your offer, buyer rules and who to leave out. Ask for a plan first. It filters your own data, then checks which API or MCP tool can fetch the rest and at what cost. Apify, Bright Data and looot are options. You approve the tool and test budget before it spends.

After adding the skill to Claude, your starting instruction can be:

/linkedin-connections-qualifier

My offer is [what you help people do]. My buyers are [who fits and why]. Exclude [who does not fit]. Inspect my Connections tab first. Show me the data gaps, proposed rules and expected outputs before building. Check my available enrichment providers, show the data coverage and test cost, and ask before spending.

Check the results against your own business. Replace my buyer groups and rules with yours.

Join the free Unstuck community and look for the skill in Free Resources/AI Skills/Sales - it is uploaded.

The prep made this work: define the job, check the data, review the plan and test a small group. Now I have a list with reasons and sources I can check.

Start with your offer and your list. Learn from one batch before running the rest.

To your next client,

Sabahudin

P.S.

Before you start, name one person who fits your offer and one who does not. Use both to check the filter.

Last Thursday, I showed the launch video for Unstuck in a live workshop.

I had finished the presentation 3 hours before the call. The video engine had been running for weeks. Then I played the video and forgot to share the sound.

I had spent all that time getting the cuts to match the music. Nobody could hear it. 😅

But the reason I showed that video was bigger than the edit. It was something I could use for my own business, and the same engine could make videos for another brand.

If you are a consultant, coach or business owner, start with a job you already need done. Build it for yourself, then learn how to do it for a client.

That is why I am opening Unstuck: a free Community to get started, and a paid Academy to build the full system with my support.

Start with a video you need to publish

You record something useful. Then it sits in a folder because you still need to cut the pauses, add the right visuals and make it fit the places you post.

A video you never post brings you nobody.

I built the AI Video Editing Engine to take one idea and turn it into a finished video in your colours and your style. You still bring the message, record yourself when needed and review the work.

You can make launch videos, edit your talking-to-camera recordings, cut podcast trailers or explain a product with a walkthrough. Each comes in wide, vertical and square formats.

Start with the one your business needs this week.

The part I would copy first

In the workshop, I explained why I review the frames before I let AI render the whole video.

If something does not make sense in a screenshot, I fix it there. It makes more sense to fix the frame than wait for an entire video and start again.

Here is the order:

1. Give it your brand assets and the message you want to get across.

2. Pick one video type and one editing style.

3. Review the script, music and preview frames.

4. Make your changes, approve the preview and let it render.

Your idea → script → preview → your approval → finished vide

On the first few videos, you need to guide it. Save the good examples. When an edit is wrong, tell it what went wrong and keep that as a bad example.

You need to train it on your own taste, on what is good.

This is not fully autonomous. You need to be there, review it and tweak things. Once the preview is right, leave the laptop open and do something else while it finishes.

Watch the full workshop for the examples, the build process and the walkthrough of the engine.

How this becomes a client service

In your case, you could start with a business that already records videos but rarely gets them published.

I shared the earlier version in my video-editing breakdown: the setup took hours, then one podcast brought me 2 inbound leads and a sponsorship deal. That was my result, not a promise. It is why I would start by publishing your own sample.

Make one useful sample in their brand. Once you can repeat the result, discuss a monthly service with a clear number of videos, what they provide and what you deliver.

The engine can hold more than one brand. Which means you can use it for your own content and to make videos for clients.

You are selling the finished work. The unchanged engine is not yours to repackage and resell.

I have spent over 3 years working in AI, from solopreneurs to my current work with enterprise businesses. The approach I keep coming back to is finding one small problem, solving it and starting the conversation there.

You do not need to present yourself as someone who knows everything about AI. Show a useful result you know how to deliver.

Unstuck is open: choose your starting point

I am launching the free Unstuck Community and the paid Unstuck Academy so you can take that first step with me.

This is the launch video I made with the engine:

Unstuck Community: start free

Get the free skill that studies the editing style of a video you find online, so you can use that style as a reference for your own work.

Start here if you want to try one useful job before building the full system. The Community also has free AI skills and guides you can work through at your own pace.

The free style skill is one part of the process. The full Video Editing Engine is in the Academy.

Get the free video-style skill →

Unstuck Academy: build the full system with support

Join when you want to install the full engine, make it work for your business and learn how to offer the finished work to clients.

You get the engine and setup instructions. When you get stuck, bring your questions to the Ask Sabahudin channel. Weekly office hours are planned so we can work through problems live.

The 4-hour Claude Code Masterclass and workshop recordings are included, too, so you can work through the foundations as you build.

You will need your own AI tools; their subscriptions and any generation credits are separate from Academy membership.

The first 30 places are €17/month. After those, the joining price becomes €99/month.

A sale is not promised. You will need to practise, check the quality and talk to potential clients.

Your first job inside: make one video you would be happy to publish for your own business.

Join Unstuck Academy and install the video engine →

To your next client,

Sabahudin

The last few days show how powerful AI models have become.

If you just take a look at your feed, you will see an amazing motion graphic that Opus 5.5 was capable of creating.

The only thing that is left is how to actually create scripts that convert and speak the language of your potential buyer so they can actually do certain actions while they are watching those videos.

That is the exact topic of today’s newsletter - convert with the video.

My recent 3 YT videos brought me now 3 inbound leads and one sponsorship deal with the only 50 YT subscribers.

So I tried to analyze why it is like that - and what I found is that good editing is not enough anymore; you need to sync it with a script that keeps them listening and energy that keeps them hooked.

So I do some playarounds and actually build the system that starts with those 2 main components - script & sound.

How should any video system be structured?

We need to define 2 things:

• Category

• Style

By the category I mean what type of videos you will record, what their main purpose is, and where you actually use them.

So I created my video system on 4 main categories that can be expanded.

I started with these 4:

• TALK - this is when you talk on your script and Claude does the rest of the editing in your style. It is good for short videos, tutorial videos, etc

• HOST - if you have a podcast, or you have case study videos, this category is the best. I create the trailer for the full episode or case study with the best highlights of the podcast or your recent call and turn them into the trailer that gets them hooked

• SHOW - If you want to create AI recordings, show your process and system; this category is a perfect fit. It can work across the classic step-by-step AI tutorials, curated news, and similar. The main point is to position your video as a walkthrough of the actual problem

• LAUNCH - the most interesting ones since we can create launch videos that biggest companies pay tons of money for it.

So let me actually show you instead of talking to you, and you can decide.
Watch this:

If you want to build your own video engine - you can start from my own system and adjust it to your categories, style and motion effects

We are starting today at 5:30 PM (CET) - Edit Your Videos With Claude/Codex

Join Here

To your next client,

Sabahudin, ‘close more & edit less’ Murtic

P.S.

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form.

This is a focused 30-minute call where I map your workflows, tools, and daily reality—then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step—not a vague plan, but an actual action

Welcome to the Unstuck newsletter….

Guys, before we go into the main topic, I NEED TO SAY THIS….

Whoever read this newsletter, thank you so much.

I mean, this is my favorite part of creating the content - writing the Substack edition, where I want to focus on delivering as much value as possible for free.

For all of us, online journeys always have some ups and downs. The ones who actually succeed are the ones who always come back from those hardest times. This even breaks you to give up or makes you successful.

Personally, for me, it was not easy. So many times I ask myself, “WTF am I doing here?” Am I posting into the void, or does someone from the other side actually love this? Is this newsletter overwhelming or confusing?

If you love my newsletter, I would love to reply to this email - it means a lot. If I get 1 reply, I will be more than motivated to keep sharing the value.

There are a lot of questions that sometimes you don’t know the answer to now, but by the time it comes, you get the answers.

If you are reading it for the first time - well, thank you, stranger… would love to hear your feedback, and feel free to visit outher editions HERE

Today is 2 years since I started my newsletter. It has an amazing journey. Some of the readers come from my first publication, AI Case Study; some of them come from reborn of Unstuck.

Today we are on 9300 people with open rate of 30%. This means every week 2500+ people open my email. Wow, this is actually like one big conference where everyone wants to hear something from my side.

Of course, sometimes an edition is amazing and sometimes it is a total fail.

But if I fail, I make sure to fail BIG (and smile along the way)

I am not someone who pushes those topics: “I made $10K; here is how you can do the same” type of content. Probably that cost me more money, but I am raised to not talk about the money in this way.

I believe that if you are doing good for a long time, everything will start compounding. That is actually what started happening to me.

If you can notice while you are reading my previous editions, it is that I don’t pitch that much. Usually you have one CTA at the end of each edition, and it is for my AI audit call.

Some people will say I am crazy, but I play a long-term game. I only sell when I feel it is actually worth selling, not just nonsense I refer to previously above.

The entire internet scheme is about that line, “I made X - here is how you can do the same”.

But in reality - they are actually selling you dream of those stripe screenshots. (LoL, I know this is deep - but if you understand, you understand).

It is like an endless loop of "everyone is succeeding,” but in reality everyone is struggling- but some people are using shortcuts where they are lying about their incomes.

I can screen-shoot my earnings where I can say, “ChatGPT, create for me my Stripe dashboard of 50K/m.” So guys, be careful who you trust. And I am not saying you should trust only me. Nope. I earn trust by being fully transparent with you in every step of my journey.

And a question will also be: Would you trust me more if I posted those earnings?

My goal was always humble: to earn enough so I can get my freedom and help others to succeed along the way. And freedom for me means

• work for myself on my own terms

• help my family and friends when they need it the most

• work with amazing people around the globe

• enjoy every moment of building my business

But a lot of people never ask themselves, “What is actually enough money?”

Because if you know that line - you will never fall into the trap of chasing money while your life is moving too fast.

Sorry for the long intro, but I need to say all of that.

Now to the main topics of today’s edition…

You can register for the upcoming free workshop:

The topic is video editing, and we know the drill… It is time-consuming and can be expensive to pay someone else to do that, right?

You saw that people edit videos with AI, so I want you to show my editing style

Here is the example of what you expect you can build with my video editing system:

Join below:

How to edit your videos with Claude/Codex

You leave with the working video editing system that:
• builds more trust of your ICP
• save you time on editing
• Save you money by not paying video editors
• You can deploy it right after it

20 people alrady register.

Unstuck Podcast: How To Run 30+ Clients With a 5-Person Team on Claude Code

The new Unstuck Podcast episode is out, and this one is packaged with absurd value you can implement into your business (especially if you run an agency)

The guest is Faris (Founder of Paradox Media)

I have known Faris Biogradlic for 6 years. In that whole time I never once asked him how he actually runs Paradox Media, his agency in Sarajevo. This episode is the first time I found out, and it changed what I thought 5 people could do.

Faris has been running Paradox Media for 11 years. They are crushing it on the Balkans. They worked with hundreds of clients in the last 11 years.
Faris is not a beginner playing with a new toy (AI).

So when AI showed up, he did not start from an empty prompt the way most of us did. He already had the SOPs, the approved client work, the context, and a team that knew what good looked like. He spent the last stretch turning all of that into a system his own team could run on.

Here is the one line from the recording I have not stopped thinking about:

The model is rented, the wiring is ours.

That is the whole episode in 8 words. The model is a subscription. Anyone can rent it. What Faris built is the wiring around it: the skills, the agents, the client hub, the checks. Rent the model, own the wiring, and the wiring is what actually runs the agency.

Here is exactly what you get.

I hate emails that promise a feeling and never say what is actually inside, so both lists are on the table before anything else.

Out of this email, in the next 4 minutes:

• the one sentence that explains how a 5-person agency runs 30-plus clients, and the screen he showed to prove it

• what he automated, what he still checks by hand, and where the line sits between the two

• the one number from his own outbound dashboard, and what it means for a team that size

Out of the hour, if you watch it:

• the full walk-through of the 10 specialist agents, what each one does, and how the copywriter agent was trained on his own team’s approved work, not a generic prompt

• the scheduled automations that run every 60 seconds through the day and once more at 7am, and what each one is checking for

• the client hub, screen by screen: how a new client gets onboarded without anyone on his team touching a back end

• the moment he stops explaining the system and starts running it live, Meta ads built and paused for a human to launch

• his own honest limit, said on the recording: where AI output is not good enough yet, and why

That last one is why I am sending this at all. Faris is not showing off a finished machine. He is showing the order he built it in, and being specific about where it still needs him.

You can watch it here:

Here are some highlights we cover on this episode:

The setup: one repo, 5 people, 30-plus clients

His own stats screen. Look at the 4 numbers in that top row: 62 skills, 10 agents, 30 clients, 5 people. That is the whole claim in one screenshot, on his own screen, not a slide he made for the podcast.

Everything runs from one repository, hosted on his own machine, not a cloud dashboard he pays extra for.

Every SOP the team already had is a file in it now. That is the part most people skip: he did not start by asking what to automate. He started with the work that already existed and made it a file Claude could read.

10 specialists, none of them a generic prompt

The agents grid: copywriter, designer, creative director, media buyer, analyst, sales ops, verifier, diagnostician, campaign copywriter. The copywriter agent was trained on his own team’s approved client work, which is why it sounds like Paradox Media and not like every other AI copywriter.

Checks that run while nobody is watching

His scheduled jobs, running every 60 seconds through the day, with one more sweep at 7am. This is the part that does the quiet work: catching a stalled campaign or a client task nobody assigned, before it becomes a problem on a call.

One hub, every client, no back end to touch

A new client gets a hub, not a shared folder and a hope. Onboarding is steps in this screen, the same steps every time, so the result does not depend on which person at Paradox Media happens to be free that week.

The number that stood out to me

37.5K sequences sent, a 3.09% reply rate, 281 booked opportunities. On autopilot, on his own dashboard, not a claim on a slide.

Here is my own take on that screen. A 5-person team running 30-plus clients is usually a 10-person job. That gap, 5 people doing what normally takes 10, is not the model being smarter. It is the wiring: the same specialist checking every campaign the same way, every time, instead of whoever on the team has 20 minutes free.

That is the part worth taking from this, whatever size your business is. The model is the same one everyone else can rent. The wiring, the actual SOPs turned into something Claude can run, is the part nobody can copy from you.

If you only have a few minutes, start with the agents grid and the cold outbound number above. That is the shape of the whole hour.

To your next client,

Sabahudin “the wiring is ours” Murtic

P.S.

What is the one job in your business that still only runs because you personally show up for it? Reply with it. That is usually the first thing worth wiring.

P.P.S.

Where to find Faris Biogradlic:

Paradox Media: https://paradoxmedia.agency/

LinkedIn: https://www.linkedin.com/in/farisbiogradlic/

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form.

This is a focused 30-minute call where I map your workflows, tools, and daily reality - then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step—not a vague plan, but an actual action

Fast forward today: I reached 20K followers, work with over 50+ clients, and coach over 1000 people (group & 1:1 combined).

I love to be there. Reach was generous; it was fun and exciting every time I wanted to create the content. I suppose because it was all new for me.

Don’t get me wrong, I still love LinkedIn. It is my first love. But I need something new and challenging. That’s why I am doubling down on YouTube.

The reason I love YouTube is long-form content that builds trust faster than any other platform.

And my entire content strategy will be focused around one video weekly, which I can repurpose further to the Substack article, notes, LinkedIn post, and X thread, and now I’m also thinking about expanding on Instagram too, with shorts.

I realized, if I put my soul and hearth into the one video per week, I know I can just simply use that value and repurpose it across different platforms and formats.

My YouTube strategy will be simple: post 4 videos per month and add 1-2 podcast editions as extra.

Topics are no-brainers:

• AI system breakdown

• AI offers & monetizations

• Podcast: How my network runs their business with AI

So I didn’t expect I would get my first wins with YouTube that fast. The reason for this is that for my LinkedIn journey I needed 6 months to close my first clients & sponsorship deals. With YouTube I needed 2 months for these results

Subscribe now

YT podcast video that brought me 2 inbound leads & 1 sponsorship deal

Let’s talk about the recent results:

My recent podcast edition brings me 2 inbound leads and 1 sponsorship deal

Now let me break down what happened with that video and how I achieved those results…

1) How do I select my next guest?

My first 3 episodes somehow become all about content systems.

The topics were repeatable and I need something new. (I love all 3 previous editions; they have a special place in my heart, especially because those people give me their time and believe in my podcast - thank you guys… they know who they are.)

So for the next edition I select a guest specifically who runs his AI agency from every possible level (from beginner to pro).

2) I invest a few hours to set up my video system.

I realize I need a pattern of how my podcast video will look.

My first 3 episodes were testing. The 4th episode is where everything clicked.

I needed a better structure for my podcast video process

The first and most important thing that I do is invest additional time to set up my video & editing system.

I realize, if I want to leverage YT…

I need something that:

• looks decent

• keeps the attention of my viewers with

• a better system that creates a video outline and intro scripts

• provides me better angles to ask better questions.

Here is the video system process that led to new inbounds & my most “viral” video so far:

👉 I make myself a better host. I ask good questions, which leads to better conversation (I know I am learning as I go—the best way to improve)

👉 I realized I need a better intro and hook, so I turned the raw podcast video into the edited trailer:

👉 After the trailer, I realized I need something that makes my podcast more memorable, so I created a signature template for every new podcast:

👉 Finally, I finish my process with:

• a catchy title that sounds good

• thumbnail that catches attention

• and a video description that is optimized more for SEO

Here is my time I spend to achieve these outcomes:

1. To create my video editing style I needed roughly around 5-6 hours. I needed to understand how to set up my beatmaps, video transitions, animations and my overall branding style

2. For the cartoon template, I needed to create 10 variations of my avatar and set up my SeeDance API so I could generate stunning videos. I also need a better prompt that will create the scene and deliver something totally unique. Here I spend 1 hour extra

3. I didn’t spend time to prepare what I will talk on the podcast; I let it go with the flow (so here was 0 hours spent)

4. Finally, once I have it ready, I run one command to edit my videos, and Claude finishes the entire process in 15 minutes. Luckily, it was done from the first shoot, which is telling me I do good preparation before I edit anything.

So now I can roughly spend 15-30 min to edit my entire video and schedule it on my YouTube. For scheduling I use Blotato

Why stop on one video style format?

I realized, okay, now I have my podcast video system, but why stop on that one workflow? I start exploring other options with the same video editing style but a different format.

So I add one more workflow, which I love, called “talking head videos.”

I simply prepare my script for 10-15 min and occasionally check while I am speaking. Once I record myself, I add my entire video editing style library that matches with what I say in my talking head video.

Here is the example of the video I will post soon on YouTube:

Leave a comment

Do you want to create your own video system?

From what you can see, you can achieve the same or better results from me.
I am not saying I have the best video system. Obviously, it could be improved, and I can add more video formats as well.

But I create a strong base and foundation; I can build on top of that.

And this is exactly what you can build with me on my next free workshop:

EDIT YOUR VIDEOS WITH CLAUDE CODE/CODEX

I will show you exactly how I edit my videos now, how you can create your own system, and how you can start building from my own video editing library so you don’t need to spend extra time and stop paying expensive video editors where AI can do almost the same output in less time than you expected :)

To your next client,

Sabahudin, ‘Stop paying your expensive video editors’ Murtic

P.S.

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form.

This is a focused 30-minute call where I map your workflows, tools, and daily reality—then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step—not a vague plan, but an actual action

Before we go into the main topic….

Guys, I owe you a big apology for not being consistent in last month when it comes to the newsletter.
There was no clear schedule when the newsletter was live, which days, etc.

I took a small vacation as well to recharge myself.

After that I also have the pleasure of helping my friend Jasmin to organize his live meetup.

Even I manage to share the stage with him and Chris Do

With that being said…

I need one favour from you so I can know does Unstuck goes into the right direction:

Now, onto the main topic and today's guest...

A few days ago I sat on a call and watched a guy talk into his phone while his company kept working behind him.

That was his actual agency. Real clients, real money moving.

The problem that most of us have is that we spend too much time waiting and reviewing the answers of our LLMs.

You type a prompt. Claude starts working. And now you are sitting there watching it, because you cannot really start anything else, because in 4 minutes it is going to stop and ask you something, and if you are not at the desk it just waits. So you come back. You read it, you correct it, you prompt again. Then the next job, same thing. By 6pm you have 5 things half done and none of your own work touched.

And then we go and blame the model.

Well, that is solved today’s edition.

Walid Boulanouar runs an agency on Claude Code and a second company on top of it. He spent an hour on the podcast with his screen shared, walking through how he got himself out of being a person who prompts or review all the time.

He said this:

“You are the bottleneck of your flow.”

To give you more credibility about our guest, Walid is the person who has a 132 public repos, works with more than 30+ B2B clients and have team of 20+ engineers.

His agency, Ay Automate, is ranked 1st on Google for Claude code services

He was not saying our prompts are bad. He was saying the work is sitting there waiting on us. Every loop in the day passes through your hands. Which means the whole thing runs at your speed.

Here is exactly what you get in today’s edition

I hate emails that promise a feeling, so let me put both lists on the table before the levels start.

So here is what we cover…

Walid will walk us through the 7 levels of running a company on AI agents, from one Claude Code prompt to running the whole company by voice.

About the levels:

→ the first 3 levels of his ladder, the actual boards he drew on screen, so you can point at the one you are standing on today
→ what changes at level 3, and the number it moved for one of my clients
→ where to start this week if you are on level 1 right now, for free

Out of the hour, if you watch it:

→ every step he went through to reach the setup he runs today, in order, and why each one is good. It is a walk up from level 1, not a tour of the finished thing

→ how he uses each tool for the one job it is genuinely good at instead of forcing one tool to do all of it. Claude Code for the building, Codex for the voice and the browser

→ levels 4 to 7 on his screen: the loops, the agent that prompts his other agents, and the board where he watches which ones stopped

→ at 52:41 he stops talking to me and starts talking to his agents out loud, and you watch the work start

→ and his own rule, said out loud: do not copy his setup. Take the part that works for you and drop the rest

That last one is why I am sending this at all. Walid is not showing you a finished machine to admire. He is showing you the order he built it in, so you can find your own next step in it.

Then he drew the 7 levels. Each one is him removing himself from one more place.

Watch the full podcast edition here:

Pick the level where you are now and go from there

If you are on level 1 right now, start there.

In the meantime…

Walid also has his Claude code challenge; you can join for free

One lesson a day, in order, and you do not need to finish it to get value out of it.

Day 34 is running a company from a terminal. Day 35 is hooks. Day 36 is coordinating multiple agents. That is levels 3 to 5 of the ladder above, broken into evenings.

His GitHub, where the frameworks he actually uses are open and free.

Not a portfolio. The agent swarm framework, the sandbox, the session porter, all public. And 3,994 contributions in the last year.

And his agency itself is at ayautomate.com if you want to see what level 7 looks like as a business rather than a diagram.

To your next client,
Sabahudin “one level higher” Murtic

P.S.

Which level are you on today? Reply with the number. I read every one, and it is almost always the same 2 or 3 jobs holding people at the same place.

P.P.S.

Where to find Walid Boulanouar:
LinkedIn: https://www.linkedin.com/in/walid-boulanouar
His agency: https://ayautomate.com
His GitHub: https://github.com/walidboulanouar

If you missed some of the previous podcast guest editons here are the links:

Leave a comment

Claude moved into your browser last week.

Most owners will scroll past the news. That would be a mistake, because this is the rare AI update that asks nothing new from you. No new tool to learn. No new tab to live in. No course.

What actually changed

On 12 August, Anthropic put Claude into the Chrome side panel. The plain-words version: Claude now sits next to whatever page you are working on. It can read that page, click, type and fill forms, with your logins, while you watch. And your skills come with it.

A skill, in plain words: your way of doing one job, written down once. How you follow up. How you prep for calls. What your offer is and who it is for. Claude reads it and does the job your way, not the internet’s way.

I have sat through 150+ discovery calls with business owners in the last 2 years, and the same sentence shows up in almost every one: “I paid for the tools, and my follow-ups still die, my content still does not get written and my numbers still get typed in by hand.”

This update is for that sentence. The 5 jobs you keep avoiding can now run in the tab where the work already lives.

Getting in is 3 steps:

→ Install Claude from the Chrome Web Store (paid plans: Max and Team today, Pro rolling out)

→ Sign in

→ Click the Claude icon. The side panel opens next to your work.

That’s it.

Below are the 5 jobs I would point to at first.

Every screenshot is from my own business, today.

1. Follow up where the conversation lives

This is my LinkedIn inbox this morning. A good conversation, a real question, asked at 12:35. The old way, this sits until “later”, and later never comes.

My client Dan used to spend 2 hours a day on outreach and follow-ups. Most of that time was not writing. It was re-reading old threads to remember where he left off. After we built his system, the same work takes a few minutes.

Back to this use case, in the browser:

1. Open the conversation

2. Open the side panel

3. Say: “Read this thread and draft my reply the way I write. Keep it under 60 words.”

4. Edit one line. Send it yourself.

It works like the barista who knows your order. You walk in, the coffee is already being made. You never re-explain it.

The skill holds how you write and what you offer, so no reply starts from zero.

In your case: start with the 3 conversations you keep postponing. Run it on those first.

2. Turn today’s work into tomorrow’s post

You already did the work today. The call happened. The problem got solved. Then the day ends and the post never gets written.

This is my content board right now: 15 drafts, 32 posted. Almost none of it started as “sit down and write a post”.

It started as calls.

In the browser:

1. Open the notes from today’s client call

2. Open the side panel

3. Say: “Read these notes. Find the one thing a business owner would want to know, and draft a short post in my voice. No names.”

4. Read it tomorrow morning, fix what is off, post.

A carpenter never buys wood to make sawdust. By Friday the workshop floor is covered in it. Your calls are the wood. The content is the sawdust. You are not creating anything extra, you are collecting what the week already produced.

In your case, one call this week has a post in it. The skill’s job is to find it.

3. Walk into every client call prepared

This is the prep page my system built for a call I have tonight at 18:15. Who it is, why the meeting exists, what to read before it, and where I could lose it.

The old way is winging the first 10 minutes and hoping you remember what they said last time. On a sales call, that costs you the deal.

On a client call, it costs you trust.

In the browser:

1. Open the client’s email thread (or their notes page)

2. Open the side panel

3. Say: “Build me a 1-page prep for my call with this person: where we left off, what they asked for, what I promised, and 3 things worth raising.”

4. Read it in 2 minutes before the call.

If we compare this use case to the real life analogy example here is what that mean:

You read the guest list before the dinner party, not at the door. Same thing here: 2 minutes of reading before the call beats 10 minutes of guessing inside it.

In your case: run it before your next call, then compare how the first 5 minutes feel.

4. Stop typing numbers into spreadsheets

One of my clients runs a property management company. Before we started, 200 invoices a week were typed in by hand, across a whole team. Not read. Typed.

Here is a month of my own inbox: invoices, receipts, renewals. Every one of those numbers used to end up in a sheet, moved there by a human.

In the browser:

1. Open the portal or invoice tabs you normally copy from

2. Open the side panel

3. Say: “Go through these tabs and collect every amount and date into one table: vendor, amount, date, due date.”

4. Check the table. Claude reads the pages the way you would, with your logins.

Anthropic’s own launch example is exactly this job: collecting amounts and dates from vendor portals into a budget.

To picture this problem better… It is like emptying every pocket onto the table at once, instead of digging for coins one by one. The numbers were always there. The typing was the job.

In your case… whatever gets retyped weekly in your business is the first candidate.

5. Research a prospect before you pitch

Let’s say this is the public website of your client company.

The old way: an hour of clicking around their site before a call, and you still miss the thing that matters.

The new way should be Claude’s skills in the Chrome browser:

1. Open their website

2. Open the side panel

3. Say: “Research this company for my call: what they sell, who they serve, what changed recently, and 3 questions worth asking. Use my discovery format.”

4. Skim it. Walk in knowing more than the person who spent an hour.

In a real-life example, this will be like you checking the menu before you walk into the restaurant. Nobody sits down blind and hopes. A pitch works the same way.

In your case: run it on your next 2 calls and keep the notes. That pile becomes your market map.

The reality

A skill is only as good as what it knows about your business. In reality, this is the whole game.

You can’t automate what you haven’t mapped.

A generic skill writes generic follow-ups. The reason mine work is not the browser. It is that each one knows my offer, my clients, my voice, and my way of doing that one job. The Chrome update moved that system into the room where the work happens. My whole business runs through 9 of these routines every day.

The short of it: you do not need to learn this update. You need your 5 jobs written down once. The browser part is 3 clicks.

If you need help to set up this for your business - reply with ‘CHROME’ and I will send you the discovery call form. From there we can do the proper audit and adjust my working skills to your skills

To your next client,

Sabahudin

P.S. Which of the 5 would buy you back the most time this week? Hit reply and tell me the job. I read every reply.

Leave a comment

Nadia had not posted in 14 days.

She has too many ideas. The problem is what happens when she opens the app to pick one. 40 minutes later she has read 6 lead magnets, 4 screenshots of somebody’s Stripe dashboard, and one guy who booked 20 calls in 3 hours with a model that launched yesterday. She closes it with nothing.

Do that 3 days running and you have skipped a week. Skip 2 weeks and the DMs go quiet with you.

She is a software engineer with 10 years behind her, so she did what engineers do. She fixed it with 2 free tools and about 3 cents a week.

I recorded the whole hour.

Who this is for and what it fixes

Read this if you sell your own expertise and your leads come from being visible. Coach, consultant, advisor, solo operator. You already know posting works. You also know the week it does not happen.

3 things get fixed here. The third one is the one that matters.

→ You stop opening the feed to look for ideas. The ideas arrive instead.

→ You stop reposting things that turn out to be false, which is the fastest way to burn trust you spent 2 years building.

→ Every idea comes back already pointed at the thing you sell.

That last one is why this is not a content tip.

Why I asked her

Nadia Privalikhina spent 10 years as a software engineer, including a stretch at the International Atomic Energy Agency, before she went out on her own. Now she runs AI Insiders: workshops, coaching, and an n8n & claude code course thousands of people have been through.

I have been reading her for about 2 years, back from the n8n and Make era. She tests things and then tells you what broke. That is rarer than it sounds.

Here is the proof, and it is the reason I wanted this on record.

A job ad went round X asking for 10 years of Claude Code experience. Claude Code is 17 months old. Smart people reposted it as a Polymarket screenshot and laughed at a recruiter who never existed. It was a Yahoo job ad, written that way on purpose, to get attention.

She never saw the funny version. Her system caught it first, marked it partially accurate, and told her where it actually came from.

She did not have to be careful. The system was careful for her.

What she built

2 pieces. Both free, or close enough that it does not matter.

The first one runs while she sleeps.

At 6am, Grok searches everything that happened on X in the last 24 hours and emails her the summary. She is on the free plan. Free accounts get 2 scheduled tasks, and she has run this one over 50 times.

At 7am, Claude picks it up. It reads the research, then sends out a set of cheaper sub-agents to check every claim against official blogs, changelogs and press releases. They come back with real sources and any contradiction they found.

Then Claude writes her the version she actually reads. 5 items. A verdict on each one. And what to do about it.

→ Anthropic cut 80% of the Claude Code system prompt. The angle it handed her: Anthropic just proved less prompt engineering wins. The action: run /doctor on your own rules and see what is now dead weight.

→ 19 year old turns $600 into $10,000 in 84 hours with a trading bot. Partially accurate. Real capability, invented profits.

→ Yahoo wants 10 years of Claude Code. Partially accurate, and here is where it really came from.

She does not throw the false ones out. Those are the ones she posts about.

The click almost nobody adds

This is the part I would have got wrong myself.

Grok’s email does not hold the research. It holds a teaser and a “continue reading” button. Everything real sits behind that click.

Nadia:

You only see a snippet of the news in your original email. If you tell your system to read that email, it will not read 95% of the actual research.

She has watched other people build this exact thing and stop right there. Their system reads the teaser, summarizes it with total confidence, and nobody ever notices. A confident summary of 5% looks the same as a confident summary of all of it.

So she added one step. Claude opens the email, clicks the button, reads the whole page.

Her reason: “My technical brain cannot just keep it as is.”

One click is the difference between a research system and a machine that guesses out loud.

The part that makes it a business system

This is where it stops being about content.

Her second skill plans the week. It reads the news, and then it reads 4 things no scraper can reach:

→ her Notion page of raw ideas

→ her recorded client calls

→ her project folder and git history, so it knows what she actually built

→ her old posts, so it can surface a winner she never reused. One had 1,000 saves and had never been reposted.

3 cents a run.

Every slot comes back with the idea, the angle, and a line pointing at what she sells. One of them took Zuckerberg saying billions of people will have a personal AI agent within 5 years, and handed her this: it is not 5 years away, you already built yours.

That is one trending topic doing 2 jobs. It reaches people who have never heard of her, and it lands on the exact thing she gets paid for.

The best idea it produced had nothing to do with news at all.

It pulled a line out of her Notion that said: women never question my AI price, men push back. That came off her own discovery calls. It is nowhere on the internet. No model was ever going to find it. It was sitting in a note she had forgotten she wrote.

What I took from it

I already run my research on a schedule, for the same reason she does. So I stop opening the feed.

The part that got me was not the automation. It was the input.

I have 150+ discovery calls with business owners sitting in my recordings from the last 2 years. Not one of them is on the internet. Every honest sentence in there is worth more than any trending post I could hijack, and I was mining maybe 10% of it.

The news half keeps you current. Your own half is the only reason anyone reads you instead of the news.

Build your own version this week

You do not need exact stack. You need the shape.

1. Pick where your news actually breaks. AI breaks on X. Your field breaks somewhere else, and LinkedIn is always a few days behind it.

2. Schedule one search there, daily. Grok’s free plan gives you 2 of these.

3. Put a checking pass in front of your eyes. Every claim against the original source before you read it, not after you posted it.

4. Point it at your own material. Your calls, your notes, your old posts. This is the half nobody copies, and it is the only half that makes the output yours.

5. Make it name the offer. Every idea comes back with one line on how it connects to what you sell, or it is just news.

Steps 1 to 3 take an afternoon. Steps 4 and 5 are the ones worth your Saturday.

Where to find Nadia

She shares her screen for the full hour. The Grok automation, the command file, the skill running live, and the plan it built for her week while we watched. She also offered to hand the Grok skill to anyone who asks for it.

Message her on LinkedIn. She answers.

→ Nadia on LinkedIn

→ AI Insiders, her site

→ Her YouTube

→ Book a call with her

To your next client,

Sabahudin “stop opening the feed” Murtic

P.S.

The workflow is copyable in an afternoon. What is not copyable is that she refused to accept a summary she could not check, and one browser click was the entire fix. Most broken AI systems break like that. Not at the model. At something small nobody looked at.

P.P.S.

Reply with the one thing you know about your own clients that you have never posted. That is your version of her Notion note. I will tell you what I would do with it.

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

We ask them, “Are you selling fog?”

Fog looks like it exists. You can see it sitting there. Then you reach for it and your hand comes back empty.

I have used that line about other people for years.

This week I pointed it at myself, because when I wanted to know whether a tool was any good, here was my entire research process: search it, read two articles, decide. And I would have told you, with a completely straight face, that I had “looked into it”.

I was buying fog and calling it research.

Then someone sent me a video about a free thing sitting on GitHub.

I installed it, asked it one question, and four minutes later it put a sentence in front of me that I have been saying to clients for a year. Not my sentence. A Reddit thread. With a comment section full of strangers describing my clients’ exact problem, in words I could not have invented.

There is one thing that run changed for me and it has nothing to do with research.

I will get to it at the end.

The same question, asked two ways

The tool is called /last30days.

Free, open source, MIT licensed, 55,000 stars on GitHub.

It installs into Claude Code in two lines and starts working immediately.

There is an API key question and I answer it properly further down, because “no keys needed” is not quite true and I would rather you hear the honest version from me.

Here is what I did with it.

First I asked Claude Code the normal way. What are people saying about n8n?

It ran a web search and came back with Wikipedia, n8n’s own homepage, four “what is n8n” explainer posts, two blog reviews, and a review page for a Nokia N8 phone (LOL)

Nothing in there is wrong. It is just the marketing layer. Almost every source was written by somebody who wanted something from me, and the one page that wasn’t was about a mobile phone from 2010.

Then I asked the identical question through the skill.

3:42 later:

A Reddit thread asking whether learning n8n is still worth it now that AI can build the automations for you. A Hacker News post about an account takeover flaw, CVE-2026-59208. Someone who opened their own n8n logs and found real customer names, emails and payment details sitting there in plain text.

And at number 7, this:

“Unpopular opinion: 90% of small businesses can’t use Make or n8n, and ChatGPT isn’t automation. So what are they supposed to do?”

That is the reason I am writing this edition.

I have been saying a version of that sentence for a year. I thought of it as my opinion, earned across 150-plus discovery calls. Turns out it is also a Reddit thread with a comment section full of owners describing the same wall, in their language rather than mine.

That thread was sitting there the whole time. Web search never found it. Not because it is hidden. Because nobody wrote it to rank.

What this skill is actually doing

You ask a question in normal words.

It goes to six places at the same time: Reddit, YouTube, Hacker News, GitHub, Polymarket and the web.

It reads the comments under the posts, not only the headlines.

Then it ranks by whatever keeps showing up in more than one place, and hands you one brief.

That ranking step is what makes the thing usable rather than just noisy. One angry comment on Reddit is one angry person. The same complaint appearing on Reddit, then on Hacker News, then under a YouTube video, is a pattern, and a pattern is something you can price against, pitch against, or build against.

The API key question, answered properly

I said "free" a few times up there and I want to be precise, because you will install this and then wonder whether you missed a setup step.

Six sources work the second you install it.

Nothing to sign up for, no card, no key, no config file: Reddit, YouTube, Hacker News, GitHub, Polymarket and the web.

Every screenshot in this edition came off those six. I have not paid a cent and I have not filled in a single form.

X is the only one that costs real money, roughly ten cents per run. It also has a free way round: if you are already logged into x.com in Firefox or Safari, the tool borrows that browser session and reads X without any key at all. If you would rather use a proper key, it takes one from api.x.ai or from xquik.com.

TikTok and Instagram share one key, free, from a service called scrapecreators.com. The project says its setup step hands you a few thousand free calls to start with. I have not needed it yet, so take that number from them rather than from me.

Keys, if you ever add them, live in a small settings file at ~/.config/last30days/.env. That is the whole configuration story.

My honest read: leave all three off. The six free sources are where the useful stuff is anyway, because that is where people go to complain and ask questions. X is fast for breaking news and that is not what most of us need this for.

Five jobs it did for my business today

I ran it 15 times. These 5 earned their place.

1. Before a call with a company

I ran beehiiv, as if I had a call with them tomorrow.

“I’ve been lost in the beehiiv dashboard for days and I’m this close to quitting.”

And from a self-described fan: “the product is caught in a cycle of constantly shipping new features while leaving existing ones unfinished.”

Now imagine walking into that call already knowing their users find onboarding painful. You are not pitching at them any more. You are talking about something they already know is true.

2. Before you buy a tool

I asked it about HubSpot, because half the owners I speak to are either on it or about to be.

Top result was a Reddit thread: is anyone else experiencing HubSpot emails going to spam? The comments underneath explain that HubSpot does not automatically give you a dedicated sending IP, so you are sharing with everyone else.

Then an agency owner posting “stay away from HubSpot contracts” after being locked in while downsizing.

Then a Hacker News post titled “I rage-built email automation because HubSpot wanted me to talk to sales.”

And at number six, a YouTube video called “HubSpot Review: does this all-in-one CRM really deliver?” with affiliate links disclosed in the description.

That is the whole argument in one screenshot. The review is paid. The complaint is free.

3. Before you write your next post

I asked what people are saying about AI automation for small business.

Look at the bottom of that output. There is a section called Audience Questions, and it gave me two:

Is AI consulting actually worth it for small and mid-sized businesses?

How do I price a CRM audit plus optimisation plus automation plus SOP drafting for a service business?

Those are not content ideas I invented at my desk. They are questions real owners typed out this month. One of them is my next edition.

4. Before you talk about your own category

This one stung, and I am glad it did. I asked what people say about AI automation agencies. My category.

Number 1: was a guy who made $400,000 last month building automation systems, arguing that learning automation in 2026 is one of the worst career moves you can make. Number 8: “most people trying to start an AI or automation agency have zero credibility.”

If you sell something, go read what your market says about people like you. It will change how you open your next call.

5. Before you go looking for leads

I ran Zapier, expecting a boring tool report.

Instead:

“60 days left until RingCentral contract expiration. We started looking at alternatives 3 months ago and now we are at the evaluation stage.”

And separately, somebody publicly posting that they need help building out Notion, Zapier, Docusign and Microsoft 365 together.

Those are two people announcing, in public, that they are ready to buy something. This month. No cold list, no scraper, no tool subscription.

Few things I actually think about this

A free repo does not get 55,000 stars because it is neat.

Here is why I think this one is everywhere right now.

1. It quietly does the job of collection of fully paid tools

I have paid for scraping tools. I have paid for social listening dashboards. Most of them charge a monthly fee to show you what people said on Reddit and then put a chart around it.

This does the useful part of that for nothing.

You still pay for the Claude subscription you already have, so it is not magic money. But there is no extra tool fee, no seat pricing, no annual contract, no sales call to see the good tier.

If I ran a small scraping or listening tool right now, I would be watching this repo very closely. Some of those products are one free skill away from having no reason to exist.

2. It is a small preview of where agents are going

Look past the search for a second and look at how it gets the data.

For X, it does not scrape anything. It uses your browser session, the one you are already logged into. The agent inherits your access instead of trying to break through a wall.

That is the part I keep thinking about. The limit on what an agent can do for you is mostly the list of doors it can walk through. Every time that list gets longer, the agent gets closer to doing the whole job instead of a slice of it. This is one small repo built by one person. It is pointed exactly where everything else is heading.

3. It is the best hype filter I have used

This is the one I actually care about.

There are two internets.

The first one is written to be found: blogs, homepages, top 10 lists, comparison pages, launch threads. Search engines are very good at showing you that one, because it was built for them.

The second one is written to be understood. Comments. Replies. Someone typing a question at eleven at night because they are stuck. Nobody optimised it for anything.

People who want to sell you something write the first internet. People who have your problem write the second one. This tool reads the second one.

So the next time a new AI model drops and your feed goes loud for a week, run it. You will find out what the people who actually used it are saying in week three, once the launch-day excitement has worn off and they are dealing with the thing in real work. That gap between the launch post and the week-three comments is where most bad buying decisions get made.

4. It stops the scroll and hands you one page

The normal way of researching something is to open eleven tabs, read four of them, get pulled sideways by a fifth, and come out with a vague feeling instead of a decision.

This gives you one page. Ranked, with the sources attached. You read it once, pick the two things worth acting on, and close it.

That is not a small thing. Whatever you are working on this week, a piece of content, a new offer, a product decision, the hard part is almost never finding information. It is stopping. This makes stopping easy, because there is nothing left to scroll.

Where it falls over

I ran one test that failed properly and you should see it, because the limits matter more than the wins.

I asked: small business owners AI automation frustration.

Zero Reddit threads. A Hacker News story about New York cutting red tape. Something about video settings in a computer game.

Vague in, junk out. It works on nouns: company names, tool names, products, specific arguments. It does not work on moods. If your question could be a feeling, rewrite it as a thing.

It also picks up false positives. A comedy YouTube channel turned up in my beehiiv research purely because the host runs his mailing list on beehiiv.

Last thing: do not use it for everything. Normal web search is still right for most questions. Save this for the ones where being wrong is expensive.

The thing it actually changed

Here is what I promised you at the start.

The research was never really the problem. The problem is that for three years I have been handing clients my opinions as though they were facts, because I had no cheap way to check them. Not lying to anyone. Just confident, with 150 calls behind me and no way to know whether the other ten thousand owners out there felt the same way.

Now I can check. Before I write the offer page, before I get on the call, before I tell somebody their real problem is X, I can spend four minutes reading what a few thousand people actually said and find out whether I am right.

Sometimes I am. Twice today I was not.

It was never about how confident you sound. It is about whether you bothered to check.

Try it on yourself

Install it, then ask it one question you are already sure you know the answer to. Your industry. Your main competitor. The tool you are about to renew for another year.

If the answer surprises you, you were guessing.

GRAB THE FREE REPO HERE

And if reading what your market really said leaves you staring at a problem you now want properly fixed, that is what my free AI Tools Assessment is for.

45 minutes, I look at where your daily work is stuck, and you leave with a short deck naming the one problem worth solving first. Nothing to buy at the end of it.

To your next client,

Sabahudin “read the comments” Murtic

Leave a comment

P.S. If you run it and it tells you something uncomfortable about your own category, reply and tell me. That is the useful kind of uncomfortable.

I run this whole business on my own. One person. Every discovery call I take is a good sign, someone wants to work with me, and for the longest time I still quietly dreaded the part that came after it.

The call itself is the fun bit. Someone walks you through their problem, you talk it out, you hang up and you feel good for about ten minutes. Then you remember what is sitting there waiting for you: the notes nobody has written, the proposal that needs to go out before they cool off, the CRM that is three deals behind, and the follow-up email you keep meaning to send and never do, until the lead has already gone quiet on someone else.

So most of my week was never really the calls.

It was the tax I paid after every single one. And I told myself for years that this was just what a one-person business feels like, that the admin was the price of not having a team, and I slowly made peace with handing my evenings over to it. Solo means you do everything yourself, right? That was the deal I thought I had signed.

Then one week I got stubborn about it. Instead of doing those four jobs myself, half-asleep, one after the other, I built four little workers to do them and put one boss on top of them. I dropped a single call transcript into the thing and walked off to make a coffee, honestly half-expecting to come back to a mess.

It worked. By the time I sat back down there were four finished files waiting in a folder: a summary of the call, a proposal I could actually send as it was, an updated CRM record, and a follow-up email that honestly sounded like me. I had not written a word of any of them.

That was the day it changed for me, and it comes down to one thing almost nobody using AI has clicked onto yet. I will tell you what it is at the very end. First let me show you how the thing actually works, because you can build your own version of it this week.

One helper is not a team

Here is the mistake I made for a year, and you might be making it too.

I would open Claude, ask it for one thing, get a decent answer, and close the tab. Then open it again for the next thing, and again for the one after that. That is not a team. That is just me doing the work one chat at a time with a very fast helper sitting next to me. Handy, sure. But I am still the one holding every piece together in my own head.

A team is a different setup entirely. You get one lead agent that looks at the whole job and hands each piece to a specialist underneath it.

One digs through the transcript, one drafts the proposal, one updates the record. Each keeps its own notes and its own tools so they never trip over each other, and they run at the same time instead of politely queueing up behind one another.

The clearest example I have: back when I needed to score a list of a hundred leads, handing all hundred to a single Claude was painfully slow, it started making things up somewhere around lead sixty, and I could never predict what it was going to cost me. Ten agents each taking ten leads finished while I made lunch. Same idea, only bigger.

Switching it on took me about thirty seconds. Open the terminal, type agents, turn Agent Teams on. That was the whole setup.

The exact prompt that ended my post-call admin

The part everyone braces for is the wiring.

People assume building a team of agents means config files, connecting things up, an afternoon lost inside documentation. It really doesn’t.

You describe what you want in plain English, and Claude works out the roles, decides which tools each worker is allowed to touch, and sets up how they hand the work down the line.

Below is the actual prompt I used, in full. Take it, swap my job for yours, point it at one of your own transcripts, and run it.

The prompt (copy-paste, then change the job to yours):

You are Claude Code.

Create a multi-agent team that turns a sales/discovery call transcript into useful sales assets.

When I drop a transcript into /input, the system should generate:

/output/insights/client_summary.md
/output/proposal/project_proposal.md
/output/crm/deal_update.json
/output/emails/follow_up_email.md

Create these agents:

Agent 1 — Call Insights Agent
Analyze the transcript and extract:
- client/company summary
- core problem
- pain points
- desired outcomes
- timeline and budget signals
- objections
- key quotes
Output: /output/insights/client_summary.md

Agent 2 — Proposal Builder Agent
Use the insights to draft a structured proposal with:
- problem
- proposed solution
- scope of work
- deliverables
- timeline
- next steps
Output: /output/proposal/project_proposal.md

Agent 3 — CRM Agent
Create a structured JSON CRM update including:
company, contact, main_problem, pain_points, outcomes, budget_signal, timeline, objections, deal_stage, summary.
Output: /output/crm/deal_update.json

Agent 4 — Follow-up Email Agent
Write a concise personalized follow-up email summarizing the call and suggesting the next step.
Output: /output/emails/follow_up_email.md

Workflow:
1. Read transcript from /input
2. Run Insights → Proposal → CRM → Email
3. Save outputs in their folders.

Read it once and you see the shape: a lead, four specialists, one clear output each. That is the whole trick. You are not coding. You are describing a team and the result you want.

Build it once, use it forever

Getting a team to run once is easy. Anyone can watch it work a single time and feel clever about it for an evening.

The thing that actually handed me my time back was making it repeatable, so I never have to build it again.

Once the team did the job properly, I locked it in. I wrote the roles out clearly. I set the permissions so the researcher can read but never send, and the writer can draft but never delete anything (you learn to care about that line the first time an agent gets a little too ambitious).

I ran it on one real call to be sure nothing was off. Then I saved the whole thing as a single command.

Now it just sits there in my system. Every discovery call, I run that one command and four files appear. The work dropped from about two hours to about two minutes, and honestly the time saved was not even the best part of it.

What got me was the consistency. The team does it the same careful way every single time, and it never runs out of patience at 9pm the way I always did.

Now take your hands off it

There is one more level, and it still makes me grin a bit. You stop pressing the button yourself.

You hand the team a trigger instead. Maybe a set time each morning, maybe a single click, maybe just a new file landing in a folder.

The trigger fires, the whole team runs, and the finished work gets pulled together and dropped exactly where you need it, a Notion page, your inbox, a doc, wherever. Nobody is standing over it while it happens.

I have a couple of these running while I am asleep now.

I wake up, open a folder, and the work is already sitting there finished. That, to me, is the part almost everyone misses. They get AI to shave a few minutes off one task and stop right there, when the very same setup could be running a whole slice of the business for them while they are not even in the room.

So, the one thing I promised you back at the start.

The shift almost nobody makes is simple to say and oddly hard to actually do.

For most of us, AI is a helper we lean on to get through the work a little faster. The real move is to build something that does the work in your place, start to finish, while you are off doing something else entirely. The day that finally clicked for me, my evenings came back, and the business kept moving on the nights I wasn’t touching a single part of it.

I record the whole thing up properly, for free, as a step-by-step guide: what the real difference between one agent and a team actually is, the exact prompt from up above, how to deploy it, and how to put it on a trigger so it runs without you, with a diagram and a short video for each part. If those four files in a folder sounded good to you, this is how you go build your own.

📘 Watch the full Agent Teams guide (free, step by step)

To your next client,

Sabahudin ‘team player’ Murtic

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

I lost about 200 of you in the last 40 days.

I saw the subscriber number drop. Then drop again. And I did the thing every creator does. I told myself it was the algorithm, the season or the unsubscribe button being too easy to find.

Let me be real. It was me.

I think I buried you.

Too many ideas, too many builds, too much “look what I made this week.” I was so busy proving what was possible that I forgot the person on the other end of the email has a business to run and forty minutes, tops.

Here is the uncomfortable part. This very edition is the proof of what went wrong.

Over the last two weeks I executed a pile of plans. Big ones. I upgraded the brain of my whole system, rebuilt my outreach, shipped a design system, started a product, redid my video setup. Every single one of them worked.

On paper, it was the most productive stretch I have had all year.

And it cost me the three things I tell you to protect with AI every single week. My time. My focus. My energy.

I had a call recently that put words to it. I will not name names.

The realization was simple and it stung: I was all over the place. One problem at a time, then move on. That is the whole game. I know this. I preach this. And I still got pulled into optimizing a system that was already working, because a new model dropped and I wanted to use the shiny thing.

Meanwhile the people I actually build for are still learning how to make their first skill.

They are busy delivering real value to real customers. For most of them, learning AI is a luxury they cannot afford, because they do not have the time.

That is the whole reason I am building Guideloop, which you will see below.

My promise to you is one sentence. I help you forget the weight of your weekly, repetitive, tedious tasks so you can focus on what you love in your business. Using AI the way every non-technical operator should.

And in two weeks of nonstop building, I broke that promise for myself. I built the systems. They run. But I have to ask the honest question: did I really need to spend that much of my life optimizing what already worked?

Not everything was a mess. My cohort is running, and the small group in it is genuinely happy with their progress. I manage to deliver some great results for my current client, and I am also waiting for some big deals to come in as well.

That is the part that reminds me what this is actually for.

So here is my apology, in the most useful form I know.

Below are the 11 fable plans I actually executed. For each one I will give you the problem in plain words, what I got out of it, a simple visual of the plan, and a master prompt you can copy to run the same play in your own business.

Steal any of them. But take the real lesson too. I did 11 things at once and it nearly cost me my focus and a chunk of my audience. You do not have to make that trade. Pick one. Finish it. Then pick the next.

That is it. Let’s go.

1. I taught the cheaper AI to work at my standard

My whole business runs on one master instruction file that tells the AI who I am, what I sell, and how I like things done. I was about to switch off the expensive top model and let cheaper ones do the daily work. The worry was obvious. Would the cheaper AI still sound like me and hit my quality?

So I rewrote that master file so a weaker, cheaper model could read it once and produce near the same quality. I fixed the parts that were five weeks out of date, and I wrote down the rules that used to live only in my head: how I want work delivered, when to pause, what “done” actually means.

This brings me a lower AI cost, same standard, no rework tax. One document that carries my judgment so I do not have to babysit every task.

The lesson: write down the standard once, so anything cheaper can hit it.

This is exactly what you would do before handing work to a new assistant or a VA.

Master prompt you can steal:

Act as my operations documenter. I want one master instruction file that lets any
assistant, new hire, or cheaper AI tool produce work at my standard from that file
alone. Interview me about: what my business sells today and to whom, the five things
I most often have to correct, how I like work delivered, and what "done" means to me.
Then write one document, under two pages, in plain language. Flag anything you guessed.

2. I cut my daily AI bill in half by deleting duplicate notes

I have problem that before every task, my AI loads a folder of background files about my business. Over months that folder got fat.

The same facts were copied across three or four files. Two files even disagreed with each other. I was paying for that bloat on every single run, and worse, an out-of-date copy could feed the AI a wrong fact.

What I do?

I gave every fact exactly one home file. Everywhere else, a one-line pointer instead of a full copy. I slimmed what loads at the start and let the AI pull the rest only when a task actually needs it.

The final output was the fixed load dropped from around 18,000 to 22,000 words down to roughly 9,000 to 10,000. Cheaper and faster on every run, with no more contradictions.

The lesson: one fact, one home. Delete the copies.

Master prompt you can steal:

Audit the background documents my AI assistant loads before every task. List every
file. Find every fact that appears in more than one place, and every fact that looks
out of date. Give each fact one home file, and replace the copies elsewhere with a
one-line pointer. Show me how much text loads at the start, before and after. Flag any
file that contradicts another.

3. I audited my whole system against how I actually use it

I had built a lot of tools and routines. But I had no idea which ones I actually used and which were dead weight sitting there, quietly making everything harder to navigate.

I studied 30 days of real usage, 236 work sessions. Then I gave every tool a verdict: keep, fix, merge, or delete. I also wrote down the tasks I still do by hand every time, so I could turn the worst two into their own helpers.

The output is actually a clear picture: keep 104, fix 24, merge 4, retire 28.

Nothing deleted for real, everything archived so I can undo it. The clutter that was slowing me down got a plan to disappear.

The lesson: audit against real usage, not against how proud you are of what you built.

Master prompt you can steal:

Act as my systems analyst. Look at the last 30 days of how I actually work: which
tools and routines I use, which I never touch, and which tasks I still do by hand.
Give me a keep / fix / merge / delete verdict for each tool, one line of reasoning
each. Then name the two manual tasks I repeat most and design a simple helper for each.
Change nothing until I approve each list.

4. I hunted the drift across everything that carries my brand

The problem. I make a lot of visuals: decks, posts, graphics, documents. Over time they drifted apart. Different fonts creeping in. Rules being broken. A few pointing at files that no longer existed. Any one of them could go out looking off-brand and I would not catch it until it was public.

What I did. A read-only audit of every visual tool I own. I lined them up against my one style guide and listed every place they broke a rule, with the exact file and the exact quote. No fixing yet, just an honest map of the mess.

What I got. The headline problems, found and ranked: three tools using the wrong body font, two that contradicted each other, and the root cause, a brand file that was pointed to but never actually existed. A checklist I can now fix from, in order.

The lesson: line your brand up side by side and log the drift before you fix a thing.

Master prompt you can steal:

Line up every piece of my brand output side by side: decks, posts, graphics,
documents. Check each one against my style guide for colors, fonts, and tone. List
every place they drift apart, with the exact file and the exact rule it breaks. Do not
fix anything yet. Hand me a ranked checklist of what to fix first, with a rough effort
estimate for each.

If you want to build claude code Content OS & design system that does animations like these on the newsletter, plus it covers your entire branding from pitch decks and proposals to landing pages, then you can join my waitlist for my next cohort

Join the wait list here

5. I packaged my content machine so students can install it

The problem I had was that I taught my content system to the people in my cohort. But teaching it and handing it over are different things. I wanted a student to install my entire content machine into the system they already built and be producing content in under 45 minutes, fully in their own brand.

I built a clone-and-install kit. Every skill ported over. Everything personal to me stripped out: my names, my IDs, my colors, my file paths. In their place, a short setup interview that fills in the student’s own details and repaints every template in their brand.

What I got. A teachable, install-able version of my own system. My teaching stops being “watch me” and becomes “run this yourself.” That is the difference between a course and a product.

The lesson: package a process you have perfected so someone else can install it, not just watch you do it.

Master prompt you can steal:

Help me package a process I have perfected so someone else can install it and make it
their own in under an hour. List every piece the process depends on: files, templates,
tools, accounts. Then find everything personal to me that must be stripped or swapped:
names, IDs, colors, file paths. Design a short setup interview that fills in the new
person's details. Give me the install steps and a checklist that proves nothing of
mine leaked through.

6. Guideloop and the pivot I am weighing out loud

Learning to build with AI takes time. My clients do not have it. They are busy running their businesses. So the skill they could build to save four hours a week never gets built, because learning how is a luxury. I wanted to kill that barrier.

So I built Guideloop. You record yourself doing a repetitive task once, screen and voice, and it turns that recording into an installable AI helper that can walk anyone through the same job forever. Then I wrapped it in a playable game so scoping your first workflow feels fun instead of technical.

Here is the pivot I am thinking through, honestly, in public. I first imagined Guideloop for solopreneurs and teams, everyone. Now I am leaning toward one focused thing: a tool that coaches non-technical small teams to build better skills. Narrower. Clearer. One job done well.

The final output was a working prototype and a playable front door that hands a stranger a real skill file and a “record it” call to action. And a clearer sense of who it is truly for.

Btw if you remember my previous editions I won 2nd place at Lovable X Paddle hackathon for this idea which I am working to bring to life soon.

The lesson: Record the task once and turn it into a guide anyone can follow.

Master prompt you can steal:

I do a repetitive task on my computer that I want to hand off. Walk me through
recording myself doing it once, screen and voice, in a single take. Then turn that
recording into a step-by-step guide anyone could follow: what to click, what to type,
how to know each step worked. Ask me a few questions first so the guide fits my exact
situation, not a generic template. End with the finished guide as one file I can share.

7. I built a CEO dashboard I can run my whole business from

I have a problem that my business runs on dozens of routines. But firing them from a terminal and then checking a separate app for the results is friction. I wanted one screen. Press a button, the work gets delegated. Glance at a tab, my live numbers are right there.

I realized I need to built a local command center. About 92 buttons across 9 tabs. Each button runs one of my routines quietly in the background. Each tab shows my live business data pulled in from where it lives. Some runs can even pause and ask me a question, and I answer right there in the screen.

What I got? One place to delegate my entire back office: email, CRM, content, outreach, finance. I press, it runs, I step in only when it asks. This is the closest thing I have to a chief of staff.

The lesson: build one screen with a button for every routine you delegate.

Master prompt you can steal:

Help me build one simple screen where each button runs a routine I normally do by
hand, and each tab shows me my live business numbers. Start by listing the 10 tasks I
ask for most often. For each, write the exact instruction the button should run. Then
list the data I check most and where it lives. Keep it local and simple. Show me the
button list before building anything.

8. I built a design system, and made Excalidraw its heart

Anything the AI makes for me, a deck, a post, a graphic, has to look unmistakably like my brand.

Not generic. Not “could belong to any startup.” I needed one file that any tool reads before it makes a single visual.

So I wrote one front-door brand file: 6 colors, 3 fonts, a grain texture on every background, one signature hard-offset shadow, and a short list of hard rules. And I put hand-drawn Excalidraw at the center of it.

The sketchy, handwritten look is not a limitation. It IS the brand. It is the opposite of clean corporate SaaS, and that is the point.

My output was that any tool I own can now produce something that looks like me, across every channel, without me art-directing it by hand. (these animations are all built with new content OS)

The lesson: define your colors, fonts, and five hard rules in one file every tool must read first.

Master prompt you can steal:

Help me write one brand file that every tool must read before making anything visual
for me. Pin down: my exact colors and where each is allowed, my three fonts and their
jobs, and five hard rules that make my work look like mine and no one else's. Write it
so a stranger could produce something on-brand from that file alone. Then give me a
10-point checklist to catch anything off-brand before it ships.

9. My website, the honest one that is still not live

Here is the one I am least proud of. While I rebuilt the brain of my business, redid outreach, and shipped a design system, my actual front door sat unfinished. The public website that would carry my offers. Still “building.” For weeks.

It kept slipping down the list because the deeper system work felt more exciting.

The copy is blocked behind finishing my positioning, and I handed the build itself to someone helping me. It is moving, slowly.

So what I got? A lesson, mostly.

This is the clearest proof of the whole edition. I optimized the machine behind the counter while the front door stayed locked. The back office was never the thing keeping me from more clients. The front door was.

The lesson: ship the front door before you optimize the back office.

Master prompt you can steal:

Act as my honest advisor. I keep improving the systems behind my business while my
public front page sits unfinished. List what my front door actually needs, in order:
the one sentence that says who I help, the proof, the offer, the call to action. Tell
me the smallest version I could ship this week. Then push back: is any back-office work
I am planning more important than getting the front door live? Be blunt.

10. I gave my video system a director

My video tools each did their own thing.

The preview lived somewhere I could not easily edit. The style shifted from tool to tool. And there was no step at the start that planned the whole video before production began. So videos came out inconsistent and slow to make.

So I added a director.

Before anything gets built, it writes a one-page plan: the type of video, the one thing that stays on screen to anchor it, and a beat-by-beat map of what happens and when. Then it hands me an editable, hand-drawn storyboard I can mark up and approve before a single frame is produced.

My outputs get consistent, on-brand video across every format, with me approving the plan up front instead of fixing the result at the end. Less rework. One style everywhere.

The lesson: write the creative plan first and preview it as a draft you can mark up.

Master prompt you can steal:

Before I make any video, act as my director and write a one-page plan first. Ask me:
where will this live, who watches it, and what should they do after. Then decide the
type of video, the one thing that stays on screen to anchor it, and a beat-by-beat map
of what happens and when. Show me the plan as a rough draft I can mark up. Do not start
building until I approve the plan.

11. I rebuilt outreach to sort people, not score them

I have a problem that my old outreach was a machine that filtered everyone out. Across every run: 165 leads gathered, exactly 1 qualified, and that 1 never even got contacted. Zero conversations. It was mechanically perfect and completely useless. The scoring was so strict that warm, real people got disqualified just under the line.

I realized I need to threw out the scoring. I kept only the hard rules that rule someone out for certain. Everyone else gets sorted into two simple groups by what they want from me: people who want to learn, and people who want it built for them. Each group gets a short, no-pitch first message. Three simple buttons feed two standing pipelines.

My final output was a system that actually starts conversations. The new rule is simple: filter people by how they reply, not by a score before you have even said hello.

The lesson: if your funnel filters everyone out, sort people instead of scoring them, and qualify on the reply.

Master prompt you can steal:

My outreach filters out almost everyone before I ever talk to them. Redesign it. Drop
the scoring that disqualifies people early. Keep only the hard rules that rule someone
out for certain. Sort everyone who survives into two simple groups by what they want
from me, and write each group a short, no-pitch first message. The goal is to start
real conversations, then qualify people by how they reply, not by a score before
contact.

The one thing to take from all of this

Eleven builds. Every one of them works. And I would trade half of them back for the focus and the 200 of you I lost while my head was in fifteen places at once.

That is the real lesson, and it is the one I keep having to relearn. The ability to do many things at once with AI feels like a superpower. For most of us, it is a trap. You get lost. You confuse motion for progress. You optimize the thing that was already fine and let the thing that actually matters, your front door, your customers, sit and wait.

So do not copy my mistake. Copy one plan. Finish it. Feel what it is like to close a loop completely before you open the next one.

Pick the one above that made you nod. Run its prompt. That is your one thing this week.

To your next client,

Sabahudin “one thing at a time” Murtic

P.S.

I want to make this up to you, so tell me: which of the 11 do you want the full behind-the-scenes build on? Hit reply with the number. The one that gets asked for most becomes the next edition, and this time I will go deep on one thing instead of skating across eleven.

P.P.S.

If learning to build your first skill feels like a luxury you cannot afford the time for, that is exactly the problem I am solving with Guideloop. Reply “record” and I will show you how it works.

To your next client,

Sabahudin ‘one command, 5 channels’ Murtic

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

If you think the fix is the next AI feature, I need to disappoint you. It isn’t.

In more than 150 discovery calls with local and small business owners over the last two years, I have watched almost none of them get better by adding another AI tool to the pile. The ones who actually got better did something far less exciting first.

You can Google “15 AI agents for small business” right now and get a list that reads like every other list: plug-and-play copy with the personality of a spreadsheet.

What you cannot Google is which of those agents an owner actually kept running after we hung up and which one they quietly abandoned by week two. That only shows up after you have sat across from 150+ owners and watched what happens once the call ends.

That is the part I am giving you here. Not another menu of features.

New AI features keep launching. The problems inside local businesses never changed.

For a long time I believed what most of the AI industry is still selling you: that a new feature, the latest model, one more tool bolted onto the stack would fix a business.

It doesn’t.

Most local businesses have the exact same problems they had before ChatGPT existed. Slow follow-up. Leads going cold in the CRM. Invoices chased by hand. Content that only happens when the owner has a spare hour. None of that is new.

AI did not create these problems, and on its own it does not solve them either.

What is genuinely new is quieter than a smarter chatbot.

AI can now plug into the tools a business already runs on, read the company’s own data, and act inside that context instead of sitting beside it in a separate tab. That is the real shift. A system that connects to the CRM, the inbox, the calendar and quietly does the repetitive work, so the business saves time and makes more money without hiring anyone new.

I chased shiny features, and it cost me real money

Let me be real, because the honest version is the only one worth reading.

For a long stretch I chased the feature, not the problem. Every new model, every new agent framework, I wanted to walk into a discovery call holding something shiny.

I tried to impress owners with capabilities they never asked for and, if I am honest, mostly did not need. I lost real money and real time letting the hype steer me, instead of just asking the person in front of me the only question that mattered: what is actually broken in your business right now?

That one is on me.

Nobody forced me into that loop. I did it because demoing a clever feature felt more impressive than sitting with a boring, unglamorous admin problem for an hour.

The bucket has holes: why chasing more leads never fixes a leaking business

This one stings. Most owners are convinced their real problem is not enough leads. So they go hunting for a better marketing strategy, a new ad account, another lead source. Almost none of them look in the one place that already holds the answer: their own CRM, full of people who raised a hand once and never got a proper follow-up.

I have sat on that exact call more than once.

The owner tells me the problem is leads, and while they are saying it, I am looking at their CRM open in the other tab: hundreds of names that said yes to something once and never heard from the business again.

To give you better perspective on this problem in business, let’s frame it through life-analogy example:

Picture two kids racing to a fountain, each with a bucket. First one back with a full bucket wins. The boy sprints. His bucket is full of holes. He gets back first, but half the water leaked out on the way. The girl takes her time. Before she even fills up, she patches the holes. She arrives later, with a full bucket.

That is conversion and retention versus acquisition.

Chasing new leads is the sprint to the fountain. If the bucket leaks, slow follow-up, no re-engagement, no review or referral system, it does not matter how fast you run or how many trips you make. Most of the water is gone before it gets home.

Patch the holes before you run back to the fountain

I did not land on this order because it sounded clever in a framework. I landed on it because I got it backwards with a few early clients. I built the acquisition piece first, because it felt more exciting, and watched them still drown a month later because nothing on the inside had actually changed.

After enough of these calls, I stopped starting with acquisition. The order that works is conversion and retention first, then the admin drag sitting underneath it, then acquisition last. Fix the leaks, clear the drag, then pour in more water.

So here are the 14, in that order, pulled straight from what kept coming up on real calls. Not from a menu I wrote in advance.

Conversion and retention

1. Lead re-engagement

“I have leads, I just never got back to them.” I hear some version of that on nearly every call with a service business. So the first workflow I usually build goes into the CRM, pulls the leads that went cold, works out what has changed for each one since the last conversation, and writes a re-opener personal enough to earn a reply. It books the appointment off the back of it. No ad spend, just revenue pulled out of a list the business already paid for once.

2. Reviews and referrals

Home service owners in particular almost never ask for the review. They finish the job and quietly hope one shows up. This one closes that gap. The moment a job wraps or a customer sounds happy, it sends the review request by text and slips in a referral offer at the same time. Instead of hoping people remember you, you get a steady trickle of reviews and warm introductions running on their own.

3. Website lead nurturing

Once, on a screen-share, I watched a form submission sit untouched for two days while the owner and I talked about something else entirely. The fix here is boring, and it works. Every form that comes in gets a reply inside five minutes, every single time, whether the owner is up a ladder or asleep. The lead gets caught while it is still warm, not whenever a free minute finally shows up.

4. AI sales coaching

Every owner is convinced their close rate is a script problem. Sit through the calls and it is almost always a prep problem. So this one listens back through recent sales calls, holds the full offer in its head, and marks the exact moments deals started slipping. Then it hands the owner tailored notes and a script before the next call, so they walk in ready instead of improvising.

5. Competitor intelligence

An owner can describe a competitor’s van wrap down to the font, and tell you nothing about what that competitor is doing online. This agent watches the things that actually move the needle: their pricing, who they are hiring, what they change on their site. It turns that into a simple read on where to compete and who to go after next, so the business stops guessing about where it is quietly losing ground.

6. Follow-up and pipeline health

There is always one proposal that should have closed by now, and it sits in the back of an owner’s mind through the whole call. This build chases every open conversation and every proposal on a schedule, then runs a health check on the pipeline once a week. Nothing gets to die quietly in a thread somebody forgot to reply to.

Admin tasks

7. Data triage

I once watched an owner scroll past forty unread emails mid-call, hunting for a single invoice. Every day, this one sorts, tags, and files whatever comes into the inbox, the drive, and the CRM, and sends each thing where it belongs. Nothing slips through the cracks because someone was too busy to file it.

8. Invoicing

Ask an owner which task they hate most and it is usually this one: chasing money they have already earned. It is also the one they put off the longest. So the workflow builds the invoice straight off the completed work, chases the overdue ones in a tone that sounds like the business and not a debt collector, and tells the owner where the cash actually stands. The part everyone dreads just handles itself.

9. Client onboarding

More than one owner has admitted they are the bottleneck between a signed contract and a job that has actually started. The moment a new client signs, this one kicks off the intake. It collects the assets, sets up the folders and the kickoff docs, and puts the first call on the calendar. The work is moving before the owner has even opened their laptop.

10. Proposals

I have seen a proposal sit half-written in a drafts folder for a week, purely because there was never a clean hour to finish it. Feed this one the discovery call and it turns the conversation into a branded proposal draft, priced off the business’s own rate card. What used to wait a week goes out the same afternoon.

Acquisition

11. Paid ads

The thing owners complain about with ad agencies is rarely the spend. It is the monthly report nobody actually reads. This agent sits on the ad account, flags where money is leaking, writes fresh ad copy off whatever is already winning organically, and reports back once a week in plain English a human can follow. You stop paying a retainer for a PDF you never open.

12. Organic content

Every owner I talk to is sitting on opinions worth posting and has zero time to write any of them down. This workflow takes what is already there, the voice notes, the sales calls, the small wins, and turns it into a week of content written in their voice, queued up for a yes or a no. They stay visible without moonlighting as a content creator.

13. AI outreach, from cold email through to a full SDR

This is almost always the first thing an owner asks for, and almost never the thing that should be built first. When the rest is in place, it earns its spot. It builds a prospect list against the ideal customer, opens each one with a line pulled from a real signal instead of a template, and sorts the replies into a simple pipeline. As the business is ready for more, the same workflow grows up: watching inbound signals like site visits and content engagement, qualifying them, and dropping booked meetings onto the calendar. New conversations start every week, and the owner is not the one sending them.

14. AEO and SEO

Almost none of these owners had ever stopped to check whether ChatGPT even knows their business exists. This one audits how findable they actually are, both in Google and inside AI answers from ChatGPT and Claude, then drafts the pages and FAQ content built to get quoted back. So when someone asks an assistant for a recommendation, the business is in the answer, not buried on page two of a search.

One lesson from 150+ calls, and my own expensive mistakes

If you take one thing from all of it, take this. The question is never which AI feature to add next. It is which of your own internal loops is leaking water right now.

Nine times out of ten it is not the marketing. It is the follow-up that never happened, the lead that went cold, the invoice nobody chased, the onboarding that only moves when you personally have a free hour. Fix that first.

Then, and only then, go get more leads, because now the bucket actually holds water. That is the whole lesson, and it is the one I wish someone had handed me before I burned money on shiny features nobody needed.

To your next client, Sabahudin

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

Over the last few editions I showed you the foundation, the morning routine, the content routine, the outreach routine.

If you missed them, start with first edition here

This week, the last piece.

The one that decides whether any of it lasts.

Here is the moment nobody warns you about.

The course ends. The calls stop. You are sitting in front of your system, alone, and one question shows up.

What should I build next?

I know how this usually goes because I lived it for months. You build the exciting thing.

Or the thing some guru posted about that week. You add it to the pile. And the pile gets fatter while your business still runs on you.

A folder of skills you never open is not a system. It is a graveyard.

So Workshop 4 in my cohort is not another routine. It is the thing you do after the cohort ends, on your own, so the system keeps getting sharper without me.

It starts with one reframe.

“What do I build next” is the wrong question.

The right one is, “how do I decide what to build next,” every single time.

Because if you have the loop, you never run out of answers.

Here is the loop.

5 steps. It always starts with your time, never with a tool.

1) Find the leak.

Look at where your hours actually went last week. Not where you planned them to go. The repeating work that eats your week and never gets finished. That is your raw material.

2) Ask the value question.

For every leaking task, ask one thing: what does building this give back to my business? If the honest answer is “not much,” it comes off the list. This is how you kill the fear of missing out. You are not automating everything. You are automating the few things that buy back real hours.

Then you split what is left into 3 piles. Automate the boring, rules-based work. Keep your taste and your judgment manual, on purpose. Drop the rest.

3) Pick 3 priorities, not 30

Score the “automate” pile, rank it, circle the top 3. Three on purpose. Not 9. Three is what you can actually build, trust, and finish before you move on.

4) Build one, fix what it gets wrong.

Build the number 1. Run it by hand for 7 to 14 days. Every time it gets something wrong, you correct it, and the brain remembers, so it stops repeating the mistake. You do not get the good version on day 1. You get it after the system has learned you.

5) Schedule it, or keep your hand on it.

Here is the only rule you need. Let a routine run on its own when the work is safe to undo: reading, sorting, drafting, prepping. Keep your hand on it when it sends, posts, pays, or deletes. The system does 90% and stops. You press send.

That is the loop. Re-run it any week you have a free afternoon, forever.

One thing I have to flag, because it is where most people get stuck.

When I ask someone what they want to automate, and then ask how the manual version works, most of them go quiet. They cannot explain their own process. That is not a Claude problem.

If you cannot say the steps out loud, the system cannot do them for you yet. Map it first. Then it becomes buildable.

This is exactly what we do together in Workshop 4

You walk in with 3 routines running.

You walk out with your own ranked roadmap: the next 3 things to build, in order, and the rule for which ones are safe to let run. Live, with me, on your real business.

Not theory. Your leak list, your piles, your 3.

That is the whole point of the cohort.

Not to hand you my system. To hand you the loop, so you never need me to tell you what is next.

I could keep doing this 1:1. That is my 4000+ EUR work. But more people need it than I can reach one at a time.

Cohort status: 15-20 founding seats. A few are taken. We start July 8.

To get the most out of it, you will want a Claude Pro plan to start, your current tool stack connected, and 3-4h a week for the workshops.

Grab a founding seat here

To your next client,

Sabahudin ‘clarity first’ Murtic

P.S.

If your business ran itself for a week, what is the first thing you would build to make it run for two?

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

“ AI can clone your output.

It can write in your style. Match your tone. Produce at your speed.

What it cannot do is be you.

The perspective. The lived experience. The judgment calls.

That is the only moat worth building right now. “

Source → Charlie Hills’ substack note.

And he is so damn right.
Because in the era of AI, where content becomes a commodity, your lived experience is only what is left to stand out.

This is exactly why my content system has two halves.

And why I automated one of them and will never automate the other.

So let’s cover part that is automated and why make sense to be like that

/content-research

Finding what to talk about can be really consuming.

Especially because I get addicted to scrolling + I always forget to collect some ideas.

On top of that, the problem is that this scrolling gets me dumb, and that’s why I started creating content at the first place, to avoid this saga.

So I realised, let’s build something about that.

My system scans five platforms. LinkedIn, Substack, YouTube, X, Reddit. 200-300 posts in a run.

It scores every one against my pillars and the people I serve, and hands me about 10-15 winners in 10 minutes.

I automated this part fully.

Finding raw material is a search problem.

A machine does it faster than me, and it n

ever gets bored.

I talk about this part of the routine on my webinar

I give the full research repo to webinar atendees, but I will give it to you here as well.

Clone the repo HERE

/content-production

Half two is production. Turning an idea into a finished piece.

This part I will never automate.

Because production is where the moat lives.

The perspective. The judgment calls. The part that is me. AI can match my speed. It cannot be me.

So I keep my hands on it. The system drafts, I rewrite. It proposes, I decide. My taste stays mine.

Here is what makes production fast without losing me.

First, the design system. Every visual and every post pulls from one place. My colors, my fonts, my voice. So everything I ship looks and sounds like me, not like AI.

Second, context. Before anything runs, the system reads who I am and who I am talking to. The exact pain points of the people I serve. So when it researches or drafts, it is already matching real topics to real problems. Not generic content. Content aimed at one person.

Third, my calls. The system reads what I say on calls, and what my prospects say. The objections. The questions. The exact words they use. That becomes content too.

The best posts I write are just clear answers to something a real prospect said out loud.

Here are the infographics outputs you can build with me on my next cohort as part of this content system:

This is what we building together in Workshop 2 in my next cohort

Your content routine, both halves. The research that finds your winners. The production setup that keeps every piece in your voice. Your design system. Your context. Even your calls, turned into content.

You walk away posting consistently, in your voice, without it eating your week.

Cohort status: 15-20 founding seats. A few are taken. We start July 8.

To get the most out of it, you will want a Claude Pro plan to start, your current tool stack connected, and 3-4h a week for the workshops.

Grab a founding seat here:

Join the cohort

To your next client,

Sabahudin “your taste stays yours” Murtic

P.S. What is the one thing your best prospect said last week that should have been a post?

To your next client,

Sabahudin ‘one command, 5 channels’ Murtic

P.S.

Want to work with me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you:

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

Last edition I showed you why one system beats 14 tools.

This week, the part underneath it.

The structure that makes the whole thing work.

Here is the whole structure.

8 folders.

Each one has a job.

- context: who you are. Your offers, your voice, your clients, your priorities. Short memory, loaded before every task.

- brain: long-term memory. Every run logs what it learned and gets sharper next time. The system writes this, not me.

- areas : your playbooks. How you actually run outreach, content, delivery, finance.

- design-system: your brand. Every visual pulls from here, so nothing looks off.

- inbox: everything the system produces, sorted by type.

- projects: each client in their own folder, their context kept separate.

- research: inspiration you are not using yet. You ask, “does this fit my system?” Half the time the answer is no. That is the point.

- .claude: the engine. Your commands, your skills, your agents.

One file on top, CLAUDE.md, tells the system what every folder means.

What makes it a system, not a pile of folders?

When I prep a client call, the system reads my context, how I work, and that client’s project folder, where we are, at the same time. It combines them. The brain adds what it learned last time. The design-system keeps it on brand.

Nothing is an island.

Every folder feeds the next.

And every run, the brain remembers a little more.

That is the whole foundation.

Get this right, and everything you build on top gets sharper instead of heavier.

This is exactly what we set up together in my cohort at the first live build.

We build this structure with you.

Your folders.

Your context filled in.

Your brain is wired to compound every run and remember what you corrected, so it stops repeating the same mistakes.

Then we add the Claude code for your morning routine on top.

So your very first morning, tomorrow, is already planned. Full access to all your inboxes and your last 24h of data triaged the proper way customized to how you actually work.

That is Week 1. One workshop.

You walk away with the foundation and your first routine already running.

To get the most out of it, you will want a Claude Pro plan to start, your current tool stack connected, and 3-4h a week for the workshops & implementations.

Grab a founding seat below 👇

Join The Cohort

P.S. What is the one folder your business is missing, the one that would finally make everything talk to each other?

To your next client,

Sabahudin “structure beats willpower” Murtic

Want to work with 1:1 me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you.

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action

Life gets busy, so if you missed some of my previous AI workflow builds, you can watch and deploy for free here:

1. Build agent teams that watch your competitors’ next move

2. Book more meetings with simple outreach flow

3. Filter 5000+ LinkedIn connections with one claude skill

4. Run your content with 2 commands

Now to the main topic…

If we calculate what value AI brings to our business and life, we can put it into the 3 buckets:

• time

• money

• focus

If you have time, you can make more money and have a better focus

If you have more money you can spend more time with people you love

If you have more focus, you can organise both time & money at the same time.

All 3 buckets are interconnected and can work from both sides.

Now let’s start with the value of having more time with AI.

My screen time notification said ‘9h/day’.

9h/day to run one business. Mine.

I had the tools. 14 of them.

A folder of prompts I never opened.

3 subscriptions I forgot to cancel.

And I still felt behind.

You know this feeling.

Every week there is a new tool.
A new guru. A new thing you should be doing.
You buy it. You try it. It sits there.

The thing is, more tools never gave me my time back.

They just gave me more tabs.

Here is what I got wrong

I thought the next tool would fix it.

So I kept collecting.

But a folder of tools/skils/prompts is not a system. It is a pile.

The people who actually get their time back are not the ones with the most tools. They are the ones who built one thing that runs without them.

One routine. Then another.

So I stopped collecting and started building.

One routine at a time.

Skill by skill, step by step. Not the whole thing on day one.

That is the mistake almost everyone makes. They try to build everything at once. Then they get confused and quit.

I started with the task that buried me most.

I built one routine for it.

I used it for two weeks. I fixed what broke. Only then did I build the next.

Today core routines run my business.

My mornings. My content. My outreach. My CRM. My Finance

My screen time went from 9h to 5h. I got 17h-26h back every week.

I run the whole thing in 4-5 hours a day.

Same business. Half the screen.

+ more time to enjoy in your life

Why does this matter to you?

I could keep building this 1:1 with my current and future clients.

That is my 4000+ EUR work.

But more people need this than I can reach one at a time.

And something happens in a group that never happens solo. You hear other people’s questions. Their ideas. Their builds. Everyone gets sharper, faster.

So I opened a founding cohort.

Four weeks. We build it together.
One routine each week:
- Week 1: Setup + Morning routine → organize your day and makes me responsible for my business
- Week 2: Content routine → brings more focus, clarity and visibility to your business.
- Week 3: Outreach routine → nurture your prospects and make the right time to reach out again.
- Week 4: Troubleshooting → 2nd-brain OS working and a full roadmap for what you build next.

+ you will learn how to create animated infographics like these below (in this case I visualize my morning routine) 👇

You do not sit through lessons about AI.

You walk away with 3 core routines running in a 2nd-brain OS that you own, powered by Claude Code that runs on 4 core layers:

LAYER 1: Folder structure and rich context around each business area

LAYER 2: Notion dashboard where you can track triaged data optimized to your business needs

LAYER 3: Brain that spot the patterns and learn where to apply them

LAYER 4: Delegate and schedule it

Current cohort status:

• Launched 4 days ago

• 15-20 seats.

• 4 are taken.

• We start July 8.

Grab a founding seat here:

One Claude Code system -> Built with you in 4 weeks

These are just a few testimonials from my clients who work 1:1 with me:

If one routine got your mornings handled, would that be worth it?

That is the whole bet.

To your next client,

Sabahudin “one routine at a time” Murtic

Before I show you how to use this for your own business (or sell it as a service to B2B clients + you can clone this system for FREE)…

Here’s what’s happening in my cohort:

Agent teams are the engine behind this. One Claude Code system that knows your business inside out, grows daily, brings clients in, and remembers everything after each run. That’s what I’m covering in depth inside the cohort.

There are only 15 spots available, and today marks the first day of registration. You can join until July 2.

Join Here

Here are a few of the many testimonials I've received over the last four months about the impact I had on my clients' businesses by assisting them in setting up the Claude code system.

You can watch yesterday’s webinar recording:

Watch Webinar

Now to the main topic…

What if your competitors watched themselves for you?

Well, it is possible.

I built a small team of AI agents that watch my competitors for me.

Every week. And every single thing they tell me comes with a source link I can click.

Okay, okay, I know what you will say: “Why should you care?” - fair enough

But…

Competitive research is a job nobody ever finishes. It never ends. So it always loses to the work that pays you today. You skip it, you skip it again, and then a competitor changes their pricing and you find out from a client.

The thing is, this is exactly the kind of job AI should be doing. Not writing your emails. Doing the boring, endless, important work you keep putting off.

I think people point AI at the wrong job.

They ask it to think for them.

The real win is letting it run a process for you, the same way, every week, without you.

Let me show you what I actually built.

It runs on two engines…

The first engine is a team of scouts.

One agent per competitor, all running at the same time. Each one pulls live data off the web using BrightData agentic scrapers (you can use ‘sabahudin’ promo code and get $20 of credits for use the). Pricing, news, hiring, what people are saying on Reddit and X. They never talk to each other. They just gather, fast, and report back. Every claim carries a source.

The second engine is a war room.

This part is different. It’s a real team of agents that take the findings and argue. On purpose. A pricing analyst. A product-gap analyst. A devil’s advocate whose whole job is to challenge the others. They go back and forth until they agree on what I should actually do next.

That was the one design choice that made the whole thing work.

Two jobs. Two different tools.

For gathering, you want cheap, fast workers that run in parallel and don’t chat. For thinking, you want a team that disagrees with each other and pushes past the easy answer. Use the right one for the right job, and the output stops being generic.

Then it all lands in three places.

Files on my machine. A Notion database I can share with my team.

And one clean dashboard I can open in a browser and hand to anyone.

I ran it live on six Notion competitors. It came back with one rising threat I wasn’t watching closely enough, a full breakdown of who’s hiring for what, and every claim sourced so I could verify it myself.

Start to finish, it’s about a 15 minutes run.

Watch the full build below

This part you’ll like.

I packaged the whole thing into a repo you can install for your own company. Point it at your own competitors, run it, and let it work.

And I’m giving it away free to everyone reading this.

FREE REPO

Reply to this email with the word RADAR if you need extra help installing it and running it on your market, and if you already install, tell me what it surfaces.

I read every reply.

That’s it for this week.

If you want to build systems like this with your own hands, the Unstuck Hub Cohort closes soon. Same method, small group, four weeks. You can start with the free repo first. See what it does for you. In cohort you have full Claude Code Masterclass course + 4 live workshops where we build and ship agentic workflows.

To your next client,

Sabahudin “every claim carries a source” Murtic

P.S.

The agents found something on a competitor I’d half-written-off. That’s the whole point. The system pays attention when you’ve stopped. Reply RADAR and go find your own blind spot.

Subscribe now

Want to work with 1:1 me?

Fill out the form for a FREE AI AUDIT CALL (30 min) and see if I can genuinely help you.

Fill out my qualifications form

This is a focused 30-minute call where I map your workflows, tools, and daily reality – then show you where AI would actually make a difference right now.

After this call you’ll have the following:
→ A clear picture of which parts of your business are ready for AI
→ The 2–3 highest-impact workflows to automate first
→ A high-level overview of how your second brain should operate.
→ A concrete next step - not a vague plan, an actual action