Hi, everyone.

In June I wrote here that the autonomous agent had left the server and moved into your laptop. In September it moved into your ChatGPT plan.

On September 29, at DevDay, OpenAI launched Dots. On August 11, xAI had opened the beta of Grok Bot. Both sell the same thing under different names: an AI agent that stays on, has a computer of its own in the cloud, and works while you do something else.

I have been running this at home since June, with open-source tools called Hermes and OpenClaw. So this edition is a two-level tutorial. Level 1 is using what OpenAI and xAI just put on the shelf. Level 2 is building your own, your way, with your data.

First, what an always-on agent is

The ChatGPT you know answers when you ask. You type, it replies, done.

The always-on agent is a different animal. It has:

1. A goal, not a question. "Keep my inbox at zero," instead of "summarize this email."

2. Its own computer in the cloud, with a browser, files, and access to your tools.

3. Memory of what it already did and how you like things.

4. A way to reach you when it needs your approval.

In the loops edition I explained the idea behind this: you stop prompting and start designing who does the work. Dots and Grok Bot are that idea, packaged for sale.

Level 1 · Dots, by OpenAI

What it is. An agent you create inside ChatGPT, give a name, and send to work. It runs on GPT-6 Astra, on its own cloud computer, with access to more than 4,000 apps. You talk to it through ChatGPT, Slack, or Teams. Text messages and phone calls are coming next.

Who gets it. The first Dot is included in the Pro and Business Premium plans. Enterprise, Edu, and Healthcare customers get the beta once an admin turns it on. It has reached Brazil, where I live: it showed up in my Pro account this week.

Mine is called Amora. She has already sent an email through my assistant's inbox with a partner's correct contact, logged into a portal to check a registration nobody had answered, and pinged me with the link when the confirmation arrived. All of that while I was recording.

The example OpenAI showed. A Dot named Alfred ended a phone call with "Done, you're live," because its owner's website had just been published with the approved design.

Tutorial, step by step, on screen.

1. Open ChatGPT on your computer, on a Pro or Business Premium plan. In the left panel, right below "New chat," your Dot appears with the name you gave it. If you don't have one yet, the first conversation is called "Talk to Dot": that is where you explain who you are, how you work, and what it may or may not do. Then you name it.

2. Click the Dot. The screen looks like a normal chat, but look at the top right corner: three buttons. The profile icon opens its panel, with Call, Slack, the Dot's Computer, Recent activity (every task it created and ran), and Results. The phone icon actually calls it. And the three dots, "Your dot's actions," hold Rename, Pause, Restart, and Delete.

3. Give it a goal you can measure. Not "help me with marketing." Rather: "every Monday at 8 am, send me on Slack the 5 mentions of my brand from the past week, with links." You can type it or say it out loud.

4. Connect only what it needs. Start with one tool. An agent with access to everything is an agent that makes mistakes everywhere. Let it run for a week, read Recent activity, and only then give it more power.

Copy and paste, in the CRTA format:

Context: you are the Dot responsible for monitoring mentions of the brand [NAME] on the web and social media.
Role: reputation analyst, direct, no adjectives.
Task: every Monday at 8 am, gather the 5 most relevant mentions from the previous week, with link, source, and one line on why it matters.
Activity: post it in the #brand channel on Slack. If any mention is a crisis (public complaint, negative press), alert me immediately, don't wait for Monday.

Level 1 · Grok Bot, by xAI

What it is. xAI calls them "AI teammates you can give real work to." Each Bot signs into your tools with your login, uses them the way you would, and comes back with the work done. All your Bots share one cloud computer, so one can hand work to another.

The differentiator. It is called "Teach a task." You screen-record yourself doing a workflow once, and the Bot saves it as a routine and runs it on its own next time. No API, no integration, no developer.

Who gets it. It is included in SuperGrok, SuperGrok Plus, and Heavy, and in Cursor's Pro, Pro+, and Ultra plans, plus teams and enterprise. It runs on Mac, Windows, and iPhone.

A detail that says a lot. On the same day OpenAI launched Dots, xAI pointed the domain dot.com at the Grok Bot page. I tested it, it still redirects. Big players fighting over the same customer.

Tutorial, in 4 steps.

1. Download Grok Bot for Mac or Windows and sign in with a SuperGrok or Cursor plan.

2. Create your first Bot with a job title, not a task. "Inbox manager." "Proposal analyst." xAI suggests a few.

3. Use "Teach a task" on a routine you do every week. Record your screen once, start to finish, saying what you are doing.

4. Turn on "Auto Review" for sensitive actions. Sending email, paying, publishing: it goes through you first.

Level 2 · Hermes and OpenClaw, the agent that is yours

Now the part I live.

Dots and Grok Bot are great for starting. But they run on their computer, with their model, with your data inside their house. In June, when the US government switched off Claude Fable 5 in three days, I wrote about it: whoever depends on a single vendor also depends on that vendor's government.

There is a path to owning the agent. Two names:

OpenClaw. Open source, created by Austrian developer Peter Steinberger, it went viral in February and he was hired by OpenAI within two weeks. It runs on any system, talks through WhatsApp, Telegram, Slack, and does real things on the computer. The mascot is a red lobster. It is the most used tool in the world in this category.

Hermes Agent. By Nous Research, launched in March. The difference: it learns from its own work and rewrites its own instructions after every task. It is what I use today.

How mine actually works. I have a small server in Brazil. It runs four agents on Hermes, each with a role: a chief of staff who reads my email and calendar, a sales agent who negotiates and drafts proposals, a finance agent who tracks receivables, and an auditor who reviews the others' work every week. I talk to them on Telegram. They coordinate through a queue on GitHub, not by chatting with each other. Every night one of them writes the day's journal.

None of this requires being a programmer. It requires understanding four things: where the agent runs, with which model, with access to what, and how it reaches you. If you read Portuguese, I run a live, hands-on program where everyone builds their own agent in week 3: IA na Prática. I love watching each face when the agent sends its first message on its own.

The rule I use

The report. I started with OpenClaw and moved to Hermes because mine went down four times in fifteen days. That is not criticism, it is a report. Open-source agents are freer and more fragile. Factory agents, like Dots, are more stable and more locked in. Amora, for instance, has not gone down once. But she lives in OpenAI's house.

The rule. Anything that touches client data, contracts, or money runs on mine, on my server, with the model I chose. Research, monitoring, email triage, and first drafts can run on a Dot or a Grok Bot without guilt.

The line. You don't need more prompts. You need someone who stays awake for you. Now that someone is for sale. And you can also build it at home.

One thing to try this week

1. If you have ChatGPT Pro, open it and look for Dots. Create one with the prompt above.

2. If you don't, pick a task you repeat every week and write its goal in a way that can be measured. That is your first agent, even before you have the tool.

3. If you want the agent to be yours, with your data, on your server: the open-source path is open. Start with OpenClaw's documentation and give it one job.

BrAIn Dispatch is reader-supported. If you want practical AI for real work, without the hype, consider becoming a free or paid subscriber: braindispatch.substack.com/subscribe

See you in the next edition.
Amanda

P.S. Built your first Dot or Bot? Tag me on Instagram, @gracianoamanda. I want to see the name you gave it.

The best use of AI in sales may not be writing another prospecting email.

It may be stopping a good salesperson from starting the day by hunting for information across four different systems.

In May 2026, Travis Bryant, Anthropic’s Head of US Mid-Market GTM, explained how he uses Claude Cowork to run a 4,000-account book.

He automated three operating rhythms:

• daily customer-call preparation;

• the weekly forecast;

• quarterly account scoring and territory prioritization.

The principle behind all three is simple:

Claude builds the what. The human brings the why.

AI assembles the evidence, updates the format and surfaces signals. The human still decides where to invest time, which conversation needs to happen and what to do with context no CRM can fully capture.

BrAIn Dispatch is reader-supported. If this kind of field guide helps you put AI to work without outsourcing judgment, subscribe for free or choose a paid subscription to support future editions.

AI assembles the data. You make the decision.

Bryant runs a go-to-market operation spanning mid-market technology companies and other industries. Before Claude Cowork, assembling data, reformatting reports and re-baselining numbers consumed much of his week.

Now the repetitive work runs on three cadences: daily, weekly and quarterly.

This is not a story about replacing salespeople.

It is about giving them back the time that belongs in customer conversations.

Play 1: automated daily meeting preparation

Preparing for one customer meeting can take 15 to 30 minutes. You open the CRM, search recent emails, check opportunities, look for contract dates, identify participants and try to reconstruct the story so far.

In Bryant’s workflow, a scheduled skill checks his calendar, Salesforce and BigQuery and prepares the brief before he opens his laptop.

The important unlock is not only the skill. It is the schedule.

When preparation stops being a command you need to remember and starts running on its own, it stops competing with the rest of the day.

Play 2: the weekly forecast

Forecasting is often high-frequency work with low strategic value during the assembly stage.

The data already exists. The format repeats. What changes are the numbers, the movement and the interpretation.

A weekly routine can pull opportunity records, team commits, usage, notes and recent changes into one page with:

• top-line metrics;

• the most important deals;

• movers and decliners;

• the current forecast snapshot.

Then the leader steps in.

Claude shows what happened. The leader explains why it happened and what the team will do next.

Play 3: scoring 4,000 accounts overnight

The most ambitious project was prioritizing all 4,000 accounts.

At other companies, work of this size could consume hundreds of hours across RevOps, finance and marketing. Bryant defined the criteria, tested a territory, adjusted the weights and let Claude Cowork run the rest overnight.

Each account received a score and a written rationale for every dimension. The result then became an interactive dashboard for account executives.

That second part matters.

A ranking without an explanation looks like one more number invented by a machine.

When the seller can see the signals and rationale, they can agree, challenge and correct it.

The meeting-preparation prompt

The prompt below is a practical adaptation based on the Anthropic case. It is not a transcript of Bryant’s actual prompt.

Before using it, connect only company-approved sources and decide which data may enter the workflow.

You are my Meeting Preparation skill. Every weekday morning, create a one-page brief for each customer meeting on my calendar today.

Follow this order:

1. REVIEW MY CALENDAR

Check today’s calendar. Remove internal meetings, focus blocks and personal events. Keep only external customer meetings.

2. FOR EACH MEETING, CREATE A BRIEF IN THIS EXACT FORMAT

MEETING: [customer name + time]

PARTICIPANTS: [name + title + role in the conversation; include members of my team]

STAGE: [new customer / expansion / renewal / churn risk]

LAST CONTACT: [date and one-line summary of the latest interaction in the CRM or my recent email]

REVENUE HISTORY: [CRM value for this year and total value, only when I am authorized to access it]

NUMBERS THAT MATTER:
- [open opportunities, values and expected close dates]
- [contract renewal date, if applicable]
- [available usage or adoption metrics]

THE STORY SO FAR: [three sentences: what they bought, why they bought it and what has happened since]

THEIR LIKELY OBJECTIVE FOR THIS MEETING: [your best hypothesis based on stage, last contact and recent activity. Label it clearly as a hypothesis.]

MY ONE OBJECTIVE: [the single outcome I need from the meeting: a decision, expansion conversation, renewal commitment or another concrete next step]

TWO QUESTIONS I SHOULD ASK: [specific to their situation; nothing generic]

ONE RISK TO WATCH: [a realistic objection or warning sign supported by recent evidence]

SOURCES USED: [list the systems, documents and dates consulted]

3. DELIVER

Place all briefs in a shared document labeled with today’s date, or in the approved channel configured for this workflow.

Do not contact the customer, change CRM records or make commitments on my behalf. I review and approve every external action.

Be specific. Every brief must pass this test: does it say something an AI could not know without access to my real, authorized business data? If not, rewrite it.

A simple account-scoring rubric

Anthropic used two different rubrics for the segments Bryant’s team serves. Below is a general-purpose version you can test in your own context.

Score each factor from zero to ten:

1. Recent buying signal: engagement, webinar attendance, target-role hiring or market expansion.

2. Fit: similarity to your best 20 percent of customers across size, industry, stack and use case.

3. Relationship stage: new prospect, former customer, expansion candidate or renewal risk.

4. Company moment: funding, product launch, leadership change or relevant public activity.

5. Reachability: an internal champion, warm introduction or recent contact.

The total reaches 50 points. But the score alone is not enough.

Also ask for:

• a three-sentence rationale;

• the sources and dates used;

• confidence level;

• missing evidence;

• a recommended next step that always requires human approval.

Before connecting it to your CRM

Start small.

Choose ten accounts, run the process manually, compare the output with the team’s assessment and adjust the criteria. Only then should you consider running it across the full book.

Define:

• which systems are sources of truth;

• who may access each kind of data;

• what AI may only suggest;

• what requires human review;

• how corrections are recorded;

• when the automation must stop and ask for help.

Automating a bad decision only makes the bad decision happen faster.

One thing to try this week

Do not start with 4,000 accounts.

Start with your next meeting.

Build the brief using the prompt, check every source and mark everything you had to correct. Those corrections are the beginning of your skill’s instructions.

The goal is not to enter the meeting with more data.

It is to enter knowing which conversation needs to happen.

See you next week,

Amanda

Subscribe to BrAIn Dispatch for practical, human-first AI systems for real work.

Primary source: How an Anthropic sales leader uses Claude Cowork to run a 4,000-account book, by Travis Bryant, Anthropic, May 20, 2026.

When your competitor can generate the same post in three seconds, what still makes you impossible to replace?

That was the question I took to the stage at Gramado Summit. The answer became a framework: Voice, Lived Experience, and Volume.

BrAIn Dispatch is reader-supported. If you want practical AI for real work, without the hype, consider becoming a free or paid subscriber.

Everyone became a content machine

Generative AI made production cheap. The result is not automatically better communication. It is more communication competing for the same attention.

People may prefer an AI-generated piece in a blind test and still trust it less when its origin is revealed. Quality alone is not enough. Origin, intent, and relationship now shape how content is received.

This is the trust penalty: the same sentence can feel useful when it comes from a person and disposable when it feels mass-produced.

The three Vs of authenticity

1. Voice

Voice is what makes you recognizable without a logo. It is your editorial accent: tone, vocabulary, point of view, and the ideas you will defend even when they are inconvenient.

AI can imitate a tone. It cannot choose what you believe.

Build your voice file:

• List five statements you defend publicly.

• List five popular statements in your industry that you refuse to sign.

• Identify three words you use in a way that feels distinctly yours.

• Ban three pieces of jargon that flatten everything you write.

• Add two examples of writing that sound unmistakably like you.

Give this file to every person and every AI that produces work for your brand.

If a post could have been written by any competitor, it is volume without voice. Volume without voice is slop.

2. Lived experience

Lived experience is the material nobody else can generate because nobody else was in the room.

It is the real client conversation, the decision behind the scenes, the mistake you can admit, the detail you noticed, and the opinion you earned by doing the work.

This is not a talent problem. It is a capture problem.

Create an experience log:

• Keep one running note open during the week.

• Record decisions, mistakes, customer questions, and surprising moments.

• After a useful meeting, write one sentence before opening the next task.

• Ask permission before sharing or screenshotting a conversation. Anonymize it.

• Every Friday, turn two entries into rough drafts.

If none of your content uses anything you actually experienced this week, you are operating like a commodity.

3. Volume

Volume is where AI belongs. It can turn one strong idea into a post, email, carousel, script, summary, and internal note. It can test headlines, shorten explanations, and adapt formats.

The mistake is starting with volume.

The right order is: voice first, lived experience second, volume last.

• Define the voice before automating production.

• Feed the AI your voice file as permanent context.

• Start from one real story or argument.

• Use AI to adapt that source into several formats.

• Keep a human review before publication.

• Measure replies, saved posts, conversations, and trust, not output alone.

The architecture

Think of three circles.

Voice is the core. It contains what you believe and how you sound.

Lived experience surrounds it. It supplies facts, stories, tension, and earned perspective.

Volume sits on the outside. It distributes and adapts what already has a human center.

When a team starts from the outside, it produces more and means less.

A five-question authenticity audit

1. Voice: If the logo disappeared, would anyone know this came from us?

2. Experience: Did we publish anything this month that only we could have created?

3. Process: Does AI-generated work pass through a real human filter?

4. Metrics: Do we measure qualitative engagement, not just reach and output?

5. Relationship: Do customers trust us more than they trust the content machine?

If you answered no three or more times, increasing production will make the problem more visible.

Where AI stops

The strongest AI professional is not the person who automates everything. It is the person who knows where automation should stop.

Humans: choose the position, inject perspective, make ethical calls, respond to the community.

AI: researches, organizes, drafts, adapts, compares, and scales.

The boundary is not fixed. But it must be chosen deliberately.

One exercise for your team

Take the last ten pieces you published. Mark the source of each one:

• an original opinion

• a lived event

• customer evidence

• external research

• AI-generated filler

If the last category dominates, do not buy another content tool. Rebuild the center.

I use AI for research, structure, editing, and adaptation. The voice here is still mine.

See you next week,
Amanda

Liked this framework? Paid subscriptions keep BrAIn Dispatch independent and make practical guides like this possible.

There is an easy way to spot writing made by AI: it sounds correct, organized, and completely ownerless.

BrAIn Dispatch is reader-supported. If you want practical AI for real work, without the hype, consider becoming a free or paid subscriber.

It is the text that “dives into” a topic, “unlocks potential,” “drives impact,” and turns every ordinary idea into a “game changer.” You finish reading without finding a major mistake. You also finish without remembering a single sentence.

The problem is not using AI to write. The problem is letting the average vocabulary of the internet think and write in your place.

This is a guide you can save. It includes the words and phrases that give AI away, the three-minute setup that fixes most of them, and a block you can paste into ChatGPT, Claude, Gemini, or any other tool.

Why every AI sounds a little alike

Claude, ChatGPT, Gemini, Copilot, Perplexity, Grok, and DeepSeek learned from enormous collections of text. When your request is generic, they choose statistically likely combinations. Those combinations often arrive with the same vocabulary, transitions, and sentence shapes.

That is why you keep seeing words like delve, leverage, transformative, robust, and seamless.

It is not a machine conspiracy. It is statistical average.

The coffee test

Before accepting a sentence your AI wrote, ask one question:

Would I say this to someone over coffee?

You probably would not say, “We need to leverage a transformative journey to unlock cross-functional synergies.” You would say, “We need to choose what to do and start.”

Voice does not come from decoration. It comes from choices.

100 words and phrases that make writing sound like AI

None of these words is forbidden by nature. The problem is using them by reflex, especially when they are not part of your normal vocabulary.

Classic AI vocabulary

delve, unlock, leverage, resonate, resonance, journey, landscape, realm, tapestry, testament, beacon, cornerstone, linchpin, synergy, synergistic, holistic, transformative, transformational, empowering, impactful, innovative, authentic, resilient, elevate, transcend, reimagine, redefine, revolutionize, foster, catalyze, emerging, burgeoning, crucial, pivotal, paramount, fundamental, multifaceted, nuanced, intricate, meticulous, masterful, vibrant, bustling, daunting, robust, seamless, dynamic, comprehensive, cutting-edge

Empty phrases

it is important to note, it is worth noting, needless to say, without a doubt, in today’s fast-paced world, in the digital age, in an ever-changing landscape, this article will explore, at its core, when it comes to, not only... but also, more than just, serves as a reminder, stands as a testament, plays a vital role, paves the way, sets the stage, drives meaningful impact, delivers value, unlocks the potential, elevates the experience, navigates the challenges, embraces change, pushes the boundaries, bridges the gap, sheds light on, takes it to the next level, a game changer

Autopilot transitions

moreover, furthermore, additionally, consequently, subsequently, notably, essentially, fundamentally, arguably, undeniably, undoubtedly, strategically, proactively, effectively, in conclusion, in summary, on the other hand, meanwhile

Punctuation habits

• an em dash in every paragraph

• dramatic ellipses at the end of sentences

• too many exclamation marks

• colons that introduce nothing

• headings that repeat the sentence below them

Fix it in three steps

1. Open your permanent instructions

Look for Custom Instructions, Personal Preferences, Saved Info, AI Profile, or the equivalent setting in your tool.

2. Paste the block below

Add it to your existing instructions. Do not delete the context that already explains who you are, what you do, or how you communicate.

3. Start a new conversation and test it

Ask for a short email, post, or summary. Read it aloud. If one of the habits comes back, point to the exact phrase and ask for a rewrite.

The block to copy

VOICE AND STYLE: NON-NEGOTIABLE RULES

Write like a specific person speaking to another specific person. Prefer concrete nouns, active verbs, short sentences, and examples from the context I provide.

Do not use em dashes. Use commas, periods, parentheses, or a colon when it introduces a real list.

Avoid the following words unless I explicitly use them first or the subject truly requires them: delve, unlock, leverage, resonate, journey, landscape, realm, tapestry, testament, beacon, cornerstone, linchpin, synergy, holistic, transformative, empowering, impactful, innovative, authentic, resilient, elevate, transcend, reimagine, redefine, revolutionize, foster, catalyze, emerging, burgeoning, crucial, pivotal, paramount, multifaceted, nuanced, intricate, meticulous, vibrant, daunting, robust, seamless, dynamic, comprehensive, cutting-edge.

Avoid generic openings and transitions such as: it is important to note, in today’s fast-paced world, in the digital age, this article will explore, at its core, when it comes to, moreover, furthermore, additionally, consequently, notably, essentially, fundamentally, arguably, undeniably, in conclusion, and in summary.

If a sentence could appear in almost any article, replace it with a specific point. If you used any item from this list out of habit, rewrite it before showing me the final answer.

What this list cannot do

Banning words removes some autopilot. It does not create your voice.

Your AI still needs raw material from you: examples of writing you like, opinions, stories, rhythm, favorite words, subjects you avoid, and the level of formality that fits each situation.

The list cleans the surface. Your experience builds the voice.

One thing to try this week

Choose one text AI wrote for you recently. Read it aloud. Mark every sentence you would never say in a conversation. Ask for a new version without those phrases, then turn your notes into permanent instructions.

Instead of correcting the same problem ten times, teach it once.

Save this edition for the next time you configure a new tool. And send it to the person who is still “delving into the ever-evolving landscape.”

See you next week,
Amanda

Liked this guide? BrAIn Dispatch brings practical AI for real work, without hype or panic. Paid subscribers keep this work independent and make guides like this possible.

It’s Amanda. And this is the last edition of Brain Dispatch for 2025. Yes, I know, it’s December 23rd. If you’re reading this today, either you’re incredibly dedicated, or you’re hiding from family to get a moment of peace. Either way, I’m with you. It’s almost a habit of ours, right? Grabbing a coffee and catching up.

Year-end is retrospective time. And what a year 2025 was for Artificial Intelligence. It was the year AI definitively left the lab and invaded the boardroom, the factory floor, and most importantly, our news feeds. It was the year AI got rich, dangerous, frighteningly intelligent, and at times, completely stupid.

So, before you dive into the holiday feast, let’s recap the 5 moments that, in my opinion, defined the year AI broke free.

2025 Retrospective: The 5 Moments That Defined the Year of AI

1. The Money Game Went Insane

If 2024 was the year of hype, 2025 was the year the money arrived by the truckload—trillions of it. Nvidia became the first $5 trillion company (read on CNBC) by selling the “pickaxes” in this gold rush. We saw deals that looked straight out of Monopoly: OpenAI closing $300 billion partnerships with Oracle, which in turn bought chips from Nvidia, which in turn invested in OpenAI. The word “bubble” became the term of the year, and even Google’s CEO admitted that if it bursts, “no company will escape.”

Why it matters to us: Money validates technology, but it also inflates expectations. Our job is to separate real innovation investment from pure speculation.

2. AI Started to “Think” (For Real)

Until now, AI models responded. In 2025, they started to reason. With “Reasoning Models,” AI began using a “chain of thought” to solve complex problems. The result? Models from DeepMind and OpenAI won gold medals at the International Mathematical Olympiad (read in The New York Times). And even scarier: Gemini Pro started helping accelerate its own training. This is the beginning of AI self-improvement.

Why it matters to us: AI is moving from being a glorified search tool to becoming a reasoning partner. The challenge is learning to ask the right questions.

3. China Entered the Game (And Kicked Down the Door)

Until January, China was a supporting player. Then came DeepSeek, an open-source model that reached second place in global rankings at a fraction of the cost of its rivals (read in Nature). It single-handedly caused Nvidia to lose half a trillion dollars in market value and forced OpenAI to launch its own open-source model. The AI race, which was an American monologue, became a tense dialogue.

Why it matters to us: Competition accelerates innovation and makes technology cheaper. High-quality open-source models mean more power in our hands, without depending on 3 or 4 companies.

4. AI Agents Left Science Fiction

If 2024 was the year of chatbots, 2025 was the year AI Agents were born (read on InfoWorld). These are AIs that don’t just respond—they do things. They can navigate websites, fill out forms, and execute multi-step tasks. Microsoft has already embedded this into Windows 11, and the promise is that they’ll revolutionize work the way apps revolutionized phones.

Why it matters to us: This shifts the game from “I use AI” to “AI works for me.” The professional of the future won’t be the one who knows how to use ChatGPT, but the one who knows how to orchestrate a team of agents.

5. The Real World Reacted (For Better and Worse)

2025 was the year AI collided with reality. We saw Alexa+ finally launch, while Siri remained a joke. We saw Grok, Elon Musk’s chatbot, spew Nazi apologetics (read on NPR). And, at the most tragic extreme, we saw the first lawsuits linking ChatGPT use to suicides (read on NPR). AI stopped being an abstract concept and started having real consequences—shaping elections, enabling crimes, and unfortunately, costing lives.

Why it matters to us: We can no longer ignore this. Using AI today requires not just technical skill, but responsibility, ethics, and sharper critical thinking than ever before.

What We’re Doing About It in 2026

Phew. It was an intense year. And the pace is only going to increase. Staying updated has become a second job. And that’s why Brain Dispatch is evolving in 2026.

Based on everything I’ve learned writing for you and analyzing the market, the newsletter will evolve to be even more useful, practical, and to the point. Because in the middle of this chaos, we need a filter, not more noise.

What’s changing starting in January:

1.Consistent Cadence: Every Tuesday at 8 AM, you’ll have a new edition of Brain Dispatch in your inbox. Without fail.

2.Clear Formats: We’ll alternate between 3 types of content:

•🛠️ How-To: Practical tutorials and step-by-step guides so you can actually do things with AI.

•🔬 Deep Dive: In-depth analyses on topics that truly matter, answering “what does this mean for me and my business?”

•📰 Curated: A quick, smart summary of the most relevant news, with tools and prompts worth your time.

What’s NOT changing:

The core. Our mission to be “the newsletter for those who want to use AI without becoming a machine.” The direct tone, the hype-free curation, and the focus on what’s practical and applicable.

Want to Go Deeper in 2026? Meet Brain Dispatch Pro

Beyond everything happening in the free plan, in 2026 we’ll have Brain Dispatch Pro, our paid community.

It’s the space for those who want to not just read about AI, but accelerate their career and business with it. Pro subscribers will have access to:

•WhatsApp Community: A private group to ask questions, share experiences, and network with other professionals on the same journey.

•Live Meetups Throughout the Year: To discuss trends and help each other grow.

If you feel like 2026 is your year to go beyond with AI, consider becoming a Pro subscriber. It’s an investment in your professional development.

I Want to Be a Pro Subscriber!

Help Me Build Brain Dispatch 2026

Now, the most important part. I want to hear from you.

What do you most want to learn or see in Brain Dispatch in 2026?

Reply to this email. Seriously. I read everything. Your feedback will shape our content for the next year.

Finally, I want to thank you immensely for being part of our community in 2025. It’s been a privilege to have your attention.

Wishing you an excellent end of the year, happy holidays, and a 2026 full of incredible projects (with plenty of AI to make the work easier).

Warm regards,

Amanda Graciano

😬
That’s not an agent.
That’s an assistant. Maybe a chatbot.

Let’s be real:
We’re calling everything “agent” now.
It’s like calling every soda “Coke.”
Confused yet? You’re not alone — from scrappy startups to CEOs at major banks, everyone’s mixing it up.

But here’s the thing: if you don’t know who’s on your AI team,
you won’t know who’s dropping the ball.

⚽ Football Explains It All

Three types of players on every field:

🧠 The Commentator = Assistant

Knows the game inside out.
Explains every play — but only when asked.

🎥 The VAR = Automation

Sees an event. Runs a fixed protocol.
No flair. No flexibility.

🧢 The Coach = Agent

Observes the whole match.
Makes decisions. Adjusts strategy.
Acts — even when no one asks.

Sound familiar?

In your company:

• Assistant = reactive

• Automation = repetitive

• Agent = autonomous

👉 Call your assistant a coach… and you're setting yourself up for disappointment.

💥 Why Everyone's Misusing “Agent”

It’s marketing, baby.
“Agent” sounds smarter, more advanced, more… fundable.
Just like “metaverse” did in 2022.

So now:

• Add ChatGPT to your landing page?
  ➡️ Call it an agent.

• Wrap it in a workflow?
  ➡️ Say it's autonomous.

Result:
A whole lot of hype — and not much action.

If your “agent” can’t make a decision without you…
You’ve got a glorified FAQ bot.

🧠 Assistants: Smart, but Passive

Think: ChatGPT, Claude, Gemini.
They wait for your command — and deliver, brilliantly.

They DO:

✅ Write killer emails
✅ Summarize meetings
✅ Create pitch decks
✅ Analyze data
✅ Highlight contract risks

But they DON'T:

❌ Send the email
❌ Update CRM
❌ Schedule anything
❌ Notify your team

💡 Example:
A sales director uses Claude to process pipeline reports in 2 minutes instead of 2 hours.
But still has to copy, paste, send. Every week.

🎥 Automation: Rigid, Reliable

Think: Make, Zapier, n8n.
When X happens, do Y.
Perfect for recurring tasks.
Useless for improvisation.

They DO:

✅ Onboard clients automatically
✅ Send nurture sequences
✅ Build folders when meetings are booked

But they DON'T:

❌ Read context
❌ Improvise
❌ Adapt to behavior
❌ Think

💡 Example:
A VP of Sales builds an automation:
Lead downloads eBook → 5-email sequence →
Opens 3+ → SDR gets pinged
= 40% more meetings booked

🧢 Agents: Finally, Someone Who Thinks

This is the game-changer.
Agents make decisions.
Act without being micromanaged.

They DO:

✅ Track your inbox → book meetings → alert the team
✅ Monitor KPIs → detect issues → propose solutions
✅ Handle complaints → offer refunds
✅ See bottlenecks → reassign leads

But they DON'T:

❌ Fire people
❌ Handle team drama
❌ Build creative strategies from scratch
❌ Run unsupervised in critical areas

💡 Example:
A CEO uses Manus:
Agent reads tickets → replies to simple ones → escalates complex ones → updates metrics
Result: Response time cut from 4 hours to 20 minutes.

👥 Multi-Agents = AI Teamwork

Picture a content team run by agents:

• Research Agent: spots trends

• Writing Agent: drafts LinkedIn posts

• Tone Agent: ensures brand voice

• Distribution Agent: posts at peak time

💥 Boom: 15 hours saved per week for the marketing lead.

🚨 When Hype Sells More Than Substance

Remember Builder.ai?
“AI-built apps,” they said.
Raised $1.5B. Then collapsed.

Turns out… 700 humans behind the scenes.
Investors pulled the plug. In May 2025, it all crumbled.

📌 Lesson:
If they can't show the AI working — live — it's probably smoke and mirrors.

🧭 TL;DR — Know Your AI Roles

Need support now?
💬 Assistant (ChatGPT, Claude) = "Do it for me"

Repeatable task?
🔁 Automation (Make, Zapier) = "Do it again"

Real business outcome?
🧢 Agent (Manus, CrewAI) = "Handle it for me"

✨ Want to see your AI actually work?

Brain Dispatch Premium is where the magic happens:

🔓 Step-by-step tutorials
📦 Monthly prompt packs
📰 Daily news drops

All for the price of your Monday coffee.

👉 Join here

💬 Or hit reply and tell me:
Where does your assistant stop — and your agent begin?

Forward this to a friend still thinking ChatGPT is running the show.
Let's raise the bar. Together.

Usually, I’d open 37 tabs, binge-search with ⌘ + F, and pray my memory holds up during the actual meeting.

But this time? I tried something else: NotebookLM, Google’s AI-powered notebook powered by Gemini.

I expected help with organization. What I didn’t expect was this: it turned my docs into a podcast.
Yes, really. Clear, human-like audio that summarized everything I needed. I listened to it on my commute — and caught a key point I had completely missed.

Here’s how it works — and how you can do it too in under 5 minutes.

🎯 The Prompt in Practice: PDF to Podcast with NotebookLM

Step 1 – Start Your Notebook
Head to notebooklm.google.com, sign in, and create a new notebook. Name it something like “Board Meeting Q3” or “AI Strategy Deep Dive.”

Step 2 – Drop Your Materials
Drag and drop everything you’ve got:
— PDFs (slides, reports)
— Docs or Word files (meeting notes, articles)
— YouTube links (it reads transcripts!)

The more context you give, the smarter the responses.

Step 3 – Ask Smart Questions
Now chat with the AI. Try prompts like:

• “Summarize the top 3 challenges mentioned across these docs.”

• “What were the key decisions made in meeting X?”

• “Suggest 5 quick actions based on bottlenecks in report Y.”

• “Generate a 30-min meeting agenda with 5 questions the leadership team might ask.”

Step 4 – Turn It Into a Podcast
Once the insights are ready, look for the 🎧 icon or Generate Podcast option (interface may vary). Click — and boom.

In seconds, you’ll get a natural-sounding audio summary, in Portuguese (or English), ready to listen on your drive, at the gym, or while doing dishes.

In my case, I caught a subtle tone in the language that helped me reframe my talking points before the meeting. It felt like having an assistant doing prep and giving me auditory feedback.

⏱️ Time saved: 2 hours of reading and searching collapsed into 15 minutes of AI chat + listening time.

Why NotebookLM Deserves a Spot in Your Stack

✅ It only reads what you upload — no random internet guesses.
✅ You can download the audio and listen offline.
✅ It helps anticipate questions (perfect for rehearsing big meetings).
✅ It extracts the signal from noisy documents — fast.

⚠️ One catch: long PDFs (500+ pages) may trip it up. Break them into smaller chunks for best results.

Want More Like This? Try These AI Cousins:

Subscribe now

• Perplexity Pages – Ask a question, get a researched report with sources.

• Fireflies / tl;dv – Record & summarize meetings (Zoom, Meet, Teams).

• Notion AI Q&A – Ask your workspace questions and get contextual answers.

🔁 The playbook: Feed it good context → Ask specific questions → Review the output.

Your Turn to Try It

What project, meeting, or reading stack will you throw into NotebookLM first?

Start small: upload prep docs for your next big presentation. Ask it for a summary, meeting questions, and yes — turn it into your own private podcast.

Tried it already? Hit reply or tag me on LinkedIn — I’d love to hear what worked for you.

—

📬 Enjoyed this prompt?

This kind of hands-on AI experiment is what Brain Dispatch’s paid edition is all about. Tutorials, behind-the-scenes tests, and practical workflows — no hype, no filler.

👉 If you want more practical AI like this, consider becoming a subscriber.
Let’s take your AI skills from talk to transformation.

See you next week,
Amanda Graciano

This isn’t another hype machine. This is curation with context, strategy, and a human touch.

And today’s topic? The biggest AI drop of the year: GPT-5.

THE ESSENTIAL IN 30 SECONDS

OpenAI launched GPT-5 on August 7. It's their most advanced model yet — and finally, free users get access to real reasoning. It makes 45% fewer mistakes and can build apps from just one prompt.

“It’s the first time it truly feels like you’re talking to a PhD expert.”
— Sam Altman, CEO of OpenAI

WHAT CHANGED (AND WHY YOU SHOULD CARE)

1. Fewer “Hallucinations” = More Trust
— 80% fewer factual errors compared to GPT-4
— Admits what it doesn’t know
— Hallucination rate: 4.8% (down from 22%)

2. One Prompt = Real Code
— Build full apps from a single command
— Real examples: Twitter clone, CRM, 2D games
— 74.9% accuracy in real-world coding tasks (GitHub)

3. It's a System, Not Just a Model
— Smart router picks fast vs deep thinking
— Three versions: GPT-5 (full), Mini (fast), Nano (ultralight)
— 256k token context (≈ 600 pages of memory)

WHAT TO WATCH

✅ Exciting
— Free access to advanced reasoning
— Gmail + Calendar integrations (AI time management!)
— Safer responses that help, not block
— Swappable personalities: Cynic, Listener, Robot, Nerd

⚠️ Ground Yourself
— Still not AGI
— Claude & Gemini are strong competitors
— Niche tasks = still better handled by specialists

OUR TAKE

GPT-5 isn’t magic — it’s better.
Think 2020 car vs 2025 car: smoother, smarter, just more satisfying.

It’s worth your time if you’re:
✅ A dev who needs clean code
✅ A data pro
✅ A creator who wants consistency
✅ A free user ready for more power

Maybe wait if you’re:
❌ Happy with Claude or Gemini
❌ Using AI for quick tasks only
❌ Still holding out for AGI (not yet)

FINAL THOUGHT

AI doesn’t have to be magical. It just has to work.
And GPT-5? It works. Really well.

700 million people use ChatGPT weekly.
Maybe now’s your time to use AI like it’s meant to be:
As an ally. Not a substitute.

☕ Already tried GPT-5? Hit reply — what did it do that GPT-4 couldn’t?

📬 Like this kind of drop?

AI Drops are part of the paid Brain Dispatch editions — built for leaders who want sharp, practical takes on what matters in AI.

👉 If this first edition hit home, consider becoming a subscriber.
The best of AI, curated with strategy and clarity, lands here every week.

More soon,
Amanda Graciano

What to expect?

✅ Tools and tricks I’ve actually tested — no fluff, just what works.
✅ Sharp reflections on what AI is changing (and what it’s not).
✅ Weekly provocations to help you stop just consuming content and start applying it.

All of it with a dose of humor, a Brazilian lens, and my own unapologetically human voice — because learning can (and should) be light, but it also needs to be urgent.

💡 First edition lands in your inbox soon. No BS, no buzzwords — just real, useful AI.

🧠 Want to get ahead of the curve? Subscribe now and come along for the ride.

P.S.: Know someone who still thinks AI is just for Silicon Valley or some engineer’s playground? Forward this to them — and let’s expand what AI can mean.