Single learning path · 9 lessons
Build an AI assistant in Portuguese, without programming, that remembers you, reaches your tools, and works on a schedule — one piece at a time, at your own pace.
Lessons
Leave knowing how to identify what your own chat is missing to work for you, using a test—and how to name what an AI OS adds.
Leave able to diagnose any assistant: which of the three parts is missing and what symptom that gap causes.
Leave with a repetitive work task rewritten using the four Cs—and side-by-side proof that the answer improves.
Leave with your assistant’s introduction dossier written and tested—who you are, what you do, and how you want things done.
Leave with a tool you already use connected to your assistant—with limits you chose yourself.
Leave with your first recipe named, tested, and saved—a win you’ll never lose in your chat history again.
Leave with the work order for the first task your assistant will run on its own, at a scheduled time.
Leave with your assistant’s knowledge folder started and a one-page dashboard it delivers every morning.
Leave with next month’s plan written and saved—one change per week, without bloating the system.
Single learning path · Lesson 1
By the end of this lesson, you’ll be able to show, with a test in your own AI chat, what it needs to work for you — and name the pieces an AI OS adds.
You already use AI and it helps—but every conversation starts from zero. What you explained yesterday is gone today, and the cost of repeating context is yours, every day. This lesson shows why that happens and what changes when the assistant gets a home.
↓ role for studying
Think of the AI chat you use today as a brilliant professional who works well—but whose memory lasts one day. Each new conversation starts with no history: it doesn’t know who you are, what you agreed on, or how you like things done.
Attorney Renata feels this every Monday. She asks for a case status summary to send to a client and has to explain again that the client has no legal background, that the text must fit on half a page, and that the firm’s tone is direct, without legalese. By Tuesday, the explanation has vanished—and she types it all again.
Notice the detail: the AI didn’t make a mistake because it lacked ability. It made a mistake because it lacked context—and today, you’re the one carrying that context.
There’s a name for the solution: AI OS — an AI operating system. The name sounds technical, but the idea is domestic: instead of a front-desk attendant who forgets you, an assistant with a desk, notebook, and keys inside your office.
An AI OS has four parts, and you'll build each one in this course: the assistant (the chat you already use), the memory (what it remembers about you), the reach (the tools it can see) and the routines (what it does on its own, at a scheduled time).
Accountant Túlio didn’t switch apps to get his—he organized these four parts around the chat he already used. AI OS isn’t a product you buy; it’s a way to put things together.
before
Renata spends about 20 minutes a day re-explaining the office, tone, and format—and still reviews generic responses.
afterward
The assistant opens the conversation already knowing all this; she asks in one line and reviews it in one pass.
Balance: about 1h40 returned to you each week—just from context you no longer have to retype.
Here’s the good news from this lesson: the building material for an AI OS is Portuguese. The pieces you’ll create are texts — an introduction of yourself, written preferences, task recipes — the same kind of thing you’d write to train a new team member.
Accountant Túlio started with five lines: how he likes a deadline reminder to be written, what kind of client it’s for, and how far in advance it should arrive. No computer language, no hidden screen—just clear instructions saved in the right place.
If you’ve ever trained an intern, a secretary, or a new partner, you already have this course’s core skill.
Test yourself
What turns an ordinary chat into the beginning of an AI OS?
An AI OS doesn’t make a difference in a demo—it makes one on an ordinary Tuesday. Attorney Renata starts the day with a ready-made summary of what’s due that week. The message to her client with no legal background comes out in the firm’s tone on the first try. A deadline that’s easy to forget is watched by someone who never sleeps.
None of this is new magic: it’s the same old chat, surrounded by memory, reach, and routine. Each lesson in this course turns on one of these pieces—and you feel the benefit each time, without waiting until the end.
A chat answers questions. An AI OS takes care of tasks.
Practice now 0/3 done
Measure the starting point of your system in your own chat: what it already knows about you and what’s missing — ~10 min.
It's a read-only conversation: nothing is changed, deleted, or purchased. If the response is odd or empty, great — that's exactly what we're measuring.
Without me telling you anything now: what do you know about me, my work, and my preferences? List what you’d use to help me tomorrow morning. If you don’t know, clearly say what’s missing.
You've just measured your system's starting point with evidence. Each lesson from here closes part of that gap.
Summary
Single learning path · Lesson 2
By the end of this lesson, you’ll be able to examine the assistant you use today and say which of the three pieces is missing — Model, Memory, or Hands — and point out the symptom each gap causes.
When AI disappoints, the common reaction is to blame its intelligence—or your question. Almost always, the problem is something else: an entire part of the system doesn’t exist yet. Knowing how to name the three keeps you from spending months adjusting the wrong part.
↓ role for studying
The first piece is the model: the engine that reads, reasons, and writes. It’s the part you get ready-made when you open any AI chat—and for most uses, it’s already excellent.
What the model can’t do is guess. Accountant Túlio asks it to "write a deadline reminder" and gets a flawless text — for a nonexistent client, about a generic tax, in a tone that isn’t his. The engine worked; everything around it was missing.
Keep this distinction in mind: writing quality comes from the model; accuracy about your reality comes from the other two components.
The second piece is memory: the assistant’s notebook. It holds three kinds of notes—facts about you and your work, preferences for format and tone, and decisions already made (“client X prefers email,” “never use legalese with clients who have no legal background”).
With the notebook filled in, the same one-line request comes back right because it already knew the rest. It's the cheapest piece of the system—it only costs the work of writing—and the one that most changes the day-to-day results.
before
Túlio’s deadline reminder comes out generic; he rewrites the request four times until the text is ready to send.
afterward
With the client's name, lead time, and tone saved in memory, the reminder is ready in the first draft.
Balance: from 4 versions to 1—about 15 minutes saved per client, every time a deadline gets tight.
The third piece is the hands: what the assistant can reach outside the conversation—your calendar, your documents, your email. Without hands, it only talks about the world; with hands, it checks the world before answering.
Attorney Renata let the assistant see her calendar. The difference was immediate: instead of taking a guess and suggesting “how about Thursday at 10 a.m.?”, it proposes times that really work, avoiding the hearing she’d forgotten to mention.
Hands have two levels, and the distinction matters: see (read the calendar, read a document) and tweak (mark, send, change). You’ll always start with read — lesson 5 covers that carefully.
Test yourself
Your assistant writes well, but makes up your free time. What’s missing?
With the three pieces named, your assistant's flaws become a diagnosis instead of a frustration. Generic response that could work for anyone? It lacks memory. Confidently invented data—a time, an amount, a law article? It lacks a hand to check the source. Confusing text even with complete context and the right access? That's when it's the model—the rarest case of the three.
That’s how attorney Renata stopped rewriting requests for no reason: before changing the question, she asks herself which component the error is pointing to. Most of the time, the answer isn’t "ask better" — it’s to build the missing component.
The Model thinks. Memory remembers. Hands reach.
Practice now 0/4 done
Leave with an assessment of your current assistant filled out: which part is there, which is missing, and one example of each—~10 min.
You just observe and take notes; no changes are made in the app. There’s no wrong answer here—it’s a snapshot, and an incomplete snapshot is information too.
You've just completed an audit that most users never do. Now you know which piece to tackle first — and that's exactly where the course goes next.
Summary
One track · Lesson 3
By the end of this lesson, you’ll be able to take a recurring task from your work, rewrite it using the four Cs — Context, Command, Capability, Cycle — and compare side by side how much the response improves.
You already know which pieces make up the system; what you need is a way to use them without starting from scratch with every request. Without a script, every conversation becomes a negotiation: you ask, the result is off, and you fix it. The four Cs method replaces that negotiation with four questions you answer once.
↓ role for studying
The first C answers a simple question: what would you tell a capable colleague who just joined before delegating this task? Who will read the result, what it’s for, and what must not happen.
Attorney Renata wants a case status summary. The context that changes everything fits in three lines: the reader is a client with no legal background who’s anxious; the goal is to reassure without promising an outcome; the limit is half a page.
Context isn't decoration for the request—it's half of it. And starting with lesson 4, much of your context will stay saved for good instead of being typed every morning.
The second C is the request itself — and a good request has three hallmarks: a clear verb (list, summarize, compare, draft), a named format (table, list, text of a certain number of lines), and a limit (what to leave out, maximum length, what to avoid).
Compare the accountant Túlio's two requests during tax season:
before
"Can you help me with the clients’ deadlines?" — and the response is a motivational piece about financial organization.
afterward
"List the July deadlines for the 12 clients below in a table, from most urgent to least urgent, with no extra comments."
Balance: from three back-and-forths to one complete, usable answer—the cost was a longer sentence.
The third C looks to the next day. A request that worked — refined context, polished prompt — is a small asset. If it dies in the conversation history, next week you’ll have to reconstruct it from memory, and do a worse job.
Capability is that named, saved success: attorney Renata named hers “client summary” and saved it outside the chat. When the situation comes up again, she doesn’t draft it — she calls it by name.
For now, keep the principle in mind; lesson 6 is all about how to write and use these recipes.
Test yourself
One of your requests worked for the second time this month. What does the method tell you to do?
The fourth C recognizes a pattern: some of your work isn’t an event; it’s a calendar. Every Monday there’s a list to check; at the end of every month, a notice to send. Cycle is the decision to take these repetitions out of your memory and put them on a schedule.
Accountant Túlio identified his in one minute: every Monday at 7 a.m., the week’s deadlines, organized by client. He hasn’t scheduled anything yet—that’s for lesson 7—but he already knows exactly what he’ll put on the clock.
Notice how the four Cs fit together: context and command improve today’s response; capability and cycle build the system that works tomorrow.
The four Cs also work in reverse: when a result comes out wrong, one of them is missing. The question shifts from "why is AI so bad?" to "which C did I skip?".
A real office case: accountant Túlio asked for "a text about taxes to send to clients" and got a formal, three-page article. He took a breath and applied the yardstick. Context was missing — the readers are shop owners with no time — and the instruction was missing — a short notice, not an article. Two rewritten lines, and the response was ready. No drama, no starting over.
Context first. Clear command. Saved capability. Cycle on the clock.
Practice now 0/3 done
Compare, side by side, the response to your usual request and the response to the same request rewritten using the method — ~12 min.
These are two test conversations: nothing gets saved, changed, or scheduled. If the second answer isn’t better, the adjustment is almost always in the Context—add a line about who will read it and run it again.
Contexto: sou <sua profissão> e este pedido é para <quem vai ler ou usar>. O que importa aqui: <dois ou três fatos que mudam o resultado>. Comando: <verbo + o que fazer>, no formato <tabela / lista / texto de N linhas>, com no máximo <limite de tamanho>. Evite <o que não pode aparecer>. No fim, me diga: se eu repetir este pedido toda semana, que nome curto você daria a ele como receita?
You've just applied the full method to one of your own cases — and seen the difference with your own eyes, not based on a course promise.
Summary
One track · Lesson 4
By the end of this lesson, you’ll leave with your assistant’s introduction dossier written and tested — who you are, what you do, and how you want things done, on one page that works for you in every conversation.
All the context you type and retype—your profession, clients, tone, quirks—fits on a page you write carefully. Until that page exists, you pay the same explanation toll in every conversation. Today, you put it in writing: it’s the first memory piece in your system.
↓ role for studying
When someone new joins a good team, their first day isn’t a guessing game: they learn who we are, what we do, and how we speak to customers. This practice has a name in the workplace— onboarding — and that's exactly what your assistant has never received.
Attorney Renata realized this with her intern: an afternoon of onboarding on the first day led to months of aligned work. With AI, she’d been doing the opposite—thousands of little explanations, none of them permanent.
The dossier for this lesson is that afternoon introduction, written down once and for all.
A dossier that works isn't an autobiography — it's a page with five sections, each answering a question the assistant would ask itself if it could.
The five parts, in order
Accountant Túlio filled in all five sections in twenty minutes. The part that paid off most was the one he almost skipped: “Never”—three lines that saved him from his costliest mistakes.
The classic mistake with a first dossier is writing it like an introduction letter: elegant paragraphs, adjectives, life story. The assistant doesn’t need prose—it needs verifiable facts and actionable preferences, one per line.
A short case to calibrate: in her first draft, attorney Renata wrote, "I value human communication with my clients." Nice — and useless. In the second, it became: "Clients aren't experts; I explain things in plain language; no more than 10 lines; I always end by stating the next step." In the first test conversation, the difference was clear: the new text was ready to send, without edits.
The yardstick is simple: if a sentence in the dossier wouldn’t change any answer, it’s just decoration—cut it.
Once written, your dossier needs an address. The AI apps people use most today have a place for it — look in the menu for names like "memory," "custom instructions," or "about you." Anything you paste there applies to every new conversation without repeating it.
If your app doesn’t have this feature (or you can’t find it), there’s a method that works in any app: save the dossier in one of your documents and paste it as the first message in the conversation. Less automatic, same effect.
Accountant Túlio also set a quarterly reminder: the dossier is a living document—a major new client, a change in service, or a change in tone gets added to the page, while everything else stays as is.
Practice now 0/4 done
Leave with the complete dossier (five parts), approved in a real test in your chat—~15 min.
The dossier is your own text, saved wherever you choose—nothing is published or sent to anyone. If the test comes out strange, edit the page and try again; mistakes here cost nothing.
QUEM SOU: <nome, profissão, cidade — 2 linhas> MEU TRABALHO: <o que faço, para quem, volumes — 3 linhas> COM QUEM FALO: <tipos de leitor e o tom para cada um> COMO QUERO: <tamanho padrão, língua simples ou técnica, o que sempre incluir> NUNCA: <o que jamais fazer sem me perguntar antes> Guarde esta apresentação. Agora, usando só o que está nela, escreva <uma tarefa típica sua> como se você fosse meu assistente.
Your assistant just received its first day of work. From now on, every conversation starts with it already knowing who you are.
Summary
One track · Lesson 5
By the end of this lesson, you’ll be able to connect your assistant to a tool you already use — calendar, files, or email — and decide for yourself what it can see and what it can’t touch.
Your assistant already knows who you are (the dossier), but it’s still blindfolded: it can’t see your calendar, documents, or inbox. That’s why it still guesses times and makes up details. Giving it reach is the scariest step—and this lesson makes it controlled and reversible.
↓ role for studying
In the most commonly used AI apps, there’s a feature called connector (sometimes shown as "integrations" or "connected apps"). Before opening that menu, here’s a translation: each connector is a door to your office, and you hold the key — you hand it over when you want to and take it back with the same gesture.
Attorney Renata opened just one door: her calendar. From then on, “what times can I offer for Thursday’s meeting?” got answers with her actual availability—taking the hearing into account and keeping lunch free.
Nothing gets installed on your computer and nothing changes in your tools: the connection happens inside the AI app, with an on/off button.
Every access has two levels, and the difference between them is your peace of mind. See: the assistant reads your calendar, a document, an email—and nothing in the world changes. Edit: it flags, edits, sends—and then something happened in your name.
Accountant Túlio applied his firm’s rule: for the first month, everything stays at the see step. The assistant read the deadlines spreadsheet and drafted reminders—but Túlio sent them after reviewing them. He considered moving up a step only after the reminders had been right for weeks.
Common mistake
Connect everything at once on the first day. This happens because the screen offers several connections at once, and enthusiasm makes you select them all. The problem: you lose track of what it can reach, and every strange answer becomes a mystery. How to avoid it: one tool per week, starting with the one that causes the most pain — and only at the see step.
The criterion isn’t technical—it’s where your week hurts. Ask yourself: What information do I copy and paste into the chat most often? That’s the first door.
For attorney Renata, it was the calendar (rescheduling all day). For accountant Túlio, the files — the deadline spreadsheet he pasted in piece by piece. Email is usually the third door, not the first: it’s the most personal and the least necessary at the start.
Test yourself
You copy your list of commitments into the chat three times a day. Which door should you open first?
Connecting is just a tour through a menu: the app asks you to confirm with your tool account and shows what will be shared. What makes this process secure is following the steps in order—always the same for every door.
The roadmap, door by door
That’s the process accountant Túlio followed with the files — and waiting a week kept him from opening the email on impulse when he didn’t need to.
Practice now 0/4 done
Leave with a tool connected at the view-only level—and verified with a real question—~15 min.
At the see step, the assistant only reads: it doesn’t check a box, send anything, or make changes. The button that turns the connector on also turns it off—you can undo it at any time without leaving a trace in your tools.
Your assistant just gained its first bit of real reach—chosen by you, limited by you, and reversible by you.
Summary
One track · Lesson 6
By the end of this lesson, you’ll be able to turn a prompt of yours that worked into a named recipe — written, saved, and tested — and reuse it in seconds without rewriting anything.
You've already refined requests that worked — and they're rotting in your history, mixed in with everything else. Next week, you'll reconstruct a worse version from memory. The recipe ends this waste: it's the third C (Capability) moving from theory into your drawer.
↓ role for studying
Every month, accountant Túlio would gradually put together the perfect prompt for his deadline reminder: the right tone, the right order, the legal notice at the end. And every month after that, he’d scroll through his history looking for that conversation—or give up and recreate it from memory, forgetting a detail.
Chat history is a diary, not an archive: things appear in chronological order and disappear beneath new conversations. Storing valuable information there is like keeping an important document in a pile of old newspapers.
What slips through that pile is anything with its own format and destination — exactly what a recipe has.
A work recipe has four fields — no more, no less. The technical term you'll see for it is skill, but it works like a recipe notebook: anyone who reads it can recreate the dish.
The four fields: name (short, memorable: "client summary"), when to use (the situation that triggers it), the steps (the method, numbered) and how to make it good (two lines showing what the right result looks like).
before
Attorney Renata spent 15 minutes recreating the request for a contract summary—and each version came out differently.
afterward
She writes "use the client summary recipe" and attaches the contract; two minutes later, the text comes out in her usual style.
Balance: from 15 minutes to 2 per use—and a pattern that doesn’t vary with how much you remember that day.
A written recipe needs to be taught once. The process is the same as for the dossier: paste it into the app’s memory (or your system document) with a simple instruction—“save this recipe; when I type its name, apply it to the material I send you.”
From accountant Túlio's test in the first week: "use the deadlines reminder recipe" plus the client list—and the notice came out in his usual style, without any extra explanation. The name became a button that exists just for him.
Test yourself
You wrote the recipe, but the following week the assistant didn't apply it. What's the most likely cause?
One recipe is convenient; five recipes make a work system. After two months, attorney Renata's drawer: client summary, first response to a proposal, meeting outline, progress update. Each one came from a repeated success — none came from imagination.
That’s the golden rule of the drawer: don’t write recipes for tasks you imagine you might have; wait for the repetition to show up. The signal is clear, and you already know it from the third C.
If you’ve asked twice, write the recipe before the third time.
Practice now 0/3 done
Leave with a four-field recipe written, taught to your assistant, and verified in a test with real material—~12 min.
The recipe is your text: if it turns out badly, you edit it and teach it again—nothing breaks, nothing gets stuck. The test material stays intact where it’s always been.
RECEITA: <nome curto — ex.: resumo para cliente> QUANDO USAR: <a situação que dispara — ex.: cliente pede notícia do caso> PASSOS: 1) <primeiro passo do método> 2) <segundo passo> 3) <terceiro passo> COMO FICA BOM: <duas linhas com a cara do resultado certo> Guarde esta receita. Quando eu escrever só o nome dela, aplique os passos ao material que eu enviar.
You've just created your system's first custom button — one of your own successful approaches, now with a name, an address, and no expiration date.
Summary
One track · Lesson 7
By the end of this lesson, you’ll be able to write the complete work order for a task the assistant runs on its own at a set time — and leave it scheduled or ready to schedule.
Up to this point, everything in your system waits for you to ask. But some of your work is purely calendar-based — the same review every Monday, the same reminder at the end of each month — and you’re the one carrying that clock in your head. This lesson hangs the clock on the system and gives you only the result.
↓ role for studying
A routine it's the simplest part of the entire course to explain: one of your tasks, scheduled for a specific time, that happens without a prompt. An alarm clock that works instead of ringing — and what lands in your hands is the finished result.
In today’s most popular AI apps, look for features called “scheduled tasks” or “automatic reminders”—the feature still varies from app to app, and some offer it only on paid plans. If yours doesn’t have it, the lesson still applies: the work order you’ll write also works as a two-minute manual routine.
Accountant Túlio’s first routine: every Monday at 7 a.m., a list of the week’s deadlines by client, with the most urgent first. He reads it over coffee and starts the day already knowing where the fire will break out.
Not every repeated task deserves to become a routine. Use these three questions as a filter: Does it happen on a predictable day or time? Does the result have a consistent format? Have you completed this task manually and successfully more than once? Three yeses: put it on the clock. Any no: wait.
The third question is the one that saves you. Attorney Renata only scheduled the weekly case update summary after requesting it by hand for three weeks, adjusting the format each time—when she scheduled it, she scheduled something that worked.
Common mistake
Schedule a task that has never worked right when done manually. This happens because automation seems like a shortcut to fixing what’s broken. The effect is the opposite: a one-off defect becomes a one-off defect at 7 a.m. every Monday. How to avoid it: get it right by hand first — routines are for multiplying success, not hiding mistakes.
There’s a second kind of automated task, smarter than an alarm clock: the cycle with a gate (you’ll see the name "loop" around). Instead of always delivering, it checks and only contacts you when something has changed — like a security guard’s patrol, who doesn’t call the owner every time around, only when they find an open door.
Attorney Renata set hers up for case docket updates: the check runs every day, but she only gets a message when something with a deadline comes up. Silence is information too: it means “everything checked, nothing urgent.”
The gate is what separates useful automation from a flood of notifications — without it, your routine becomes one more inbox to clear.
Test yourself
Your daily routine sends a report even when there’s nothing new, and you’ve stopped reading it. What’s missing?
Every well-run routine starts with a four-line work order—the same document you’d give a trusted employee for a recurring task.
The four lines of the work order
Notice how the lessons fit together here: the recipe (lesson 6) says how to do; the work order says when e what to bring me. Accountant Túlio wrote his in five minutes — because the earlier pieces were already in place.
Practice now 0/4 done
Leave with the work order for your first routine written in four lines—scheduled in the app or saved and ready to use—~15 min.
A scheduled routine can be paused or deleted where it was created — nothing runs in secret. And if your app doesn't support scheduling yet, a written instruction still works: two minutes of copy-and-paste per week, same result.
Your system just worked at a time when you weren’t asking it to do anything. That’s the line between an assistant and a system.
Summary
One track · Lesson 8
By the end of this lesson, you’ll put together the first version of your assistant’s knowledge folder — with an address and labels — and define the one-page dashboard it delivers every morning.
Your system already has a dossier, recipes, and a work order—but they’re scattered. Scattered knowledge comes with a cost: you forget where it is, duplicate things, and let them go stale. And you’re missing the habit that brings all of this into your day effortlessly: a dashboard to start your morning. This lesson puts things in order and establishes the ritual.
↓ role for studying
Everything your assistant knows deserves to live in one place — a wiki, which here is just a short name for "the knowledge folder." No new technology: it can be a folder of documents in the service you already use.
Three drawers are enough: who I am (the dossier from lesson 4), how do I do it (the recipes and work orders) and what I know (facts that change slowly: list of important clients, fixed deadlines for the year, document templates).
Attorney Renata put hers together in ten minutes by moving what already existed. The payoff showed up with the first question: instead of searching in three places, she opened one folder.
The dashboard is the most visible deliverable in the whole system: one page every morning with what the day requires — schedule, deadlines, pending items, and one priority line. Think of the bulletin board at the entrance to a doctor’s office: anyone passing by can read it in thirty seconds and know what matters.
The content changes by profession; the format doesn’t: three short blocks and a final priority line, always in the same order, always on one page.
The same dashboard, two jobs
A system without a habit gets rusty. The ritual that keeps yours running takes ten minutes a day, in two steps: in the morning, read the dashboard and choose your priority (five minutes); at the end of the day, add one line to the wiki—what changed, what was added, what is no longer relevant (five minutes).
A first-week case: accountant Túlio skipped his afternoon check-in for three days. On Thursday, the dashboard showed a document as pending even though it had already arrived — and he nearly followed up with the wrong client. The afternoon entry he didn't write cost him exactly the embarrassment it was meant to prevent. Since then, he writes the entry before shutting down his computer.
The rule that saves the ritual: if it fails one day, don’t catch up—write only today’s line and move on. A ritual that requires catching up dies in the first busy week.
Wikis and dashboards have a natural enemy: accumulation. Every week adds a little—a fact that’s gotten stale, a dashboard block no one reads—and in two months, the single page has become three. The defense is a five-minute weekly sweep, always the same.
The weekly sweep (5 minutes, Friday or Sunday)
That’s the review that kept attorney Renata’s dashboard from turning into a report: in the first month, she cut two sections that looked useful on paper but were never actually read.
Practice now 0/4 done
Leave with a three-drawer wiki set up and the first real dashboard delivered by your assistant—~12 min.
You just create a folder and chat with the chat: nothing is sent to anyone, and no tools are changed. A poor dashboard on the first try is normal—adjust the request; don’t give up on the page.
Monte meu painel desta manhã, em uma página, três blocos e uma linha final: 1) AGENDA DE HOJE: <cole seus compromissos — ou deixe que o conector de agenda da aula 5 preencha> 2) PRAZOS E VENCIMENTOS: <cole sua lista da semana> 3) PENDÊNCIAS: <as três coisas paradas esperando alguém> Termine com uma linha: "se você só fizer uma coisa hoje, faça <...>", escolhida pelo que tem prazo mais duro. Formato: blocos curtos, mais urgente primeiro, sem parágrafos longos.
Your morning just got a first page—written by the system, in your format, before you ask.
Summary
One track · Lesson 9
By the end of this lesson, you’ll leave with the plan for your AI OS’s next 30 days written and saved in the wiki — one change per week, each tied to a piece you already know how to build.
The biggest risk with a newly built system isn’t that it will break — it’s that you’ll abandon it in the third week, when the novelty wears off. The antidote is a small, dated plan: four weeks, four changes, none taking more than fifteen minutes a day. This lesson wraps up the course by writing that plan with you.
↓ role for studying
Take inventory before planning: a dossier that introduces you in every conversation (Memory). At least one door open at the see step (Hands). A named recipe in the drawer and a work order on the clock (Capability and Cycle). A wiki with three drawers and a morning dashboard. The one-off chat from lesson 1 has become a system with a name and an address.
Attorney Renata made this inventory and noticed the most important thing: none of the pieces required new skills—just a decision and twenty minutes. What seemed like a programmer’s job turned out to be getting organized in writing.
Taking inventory also gently reveals the gaps: whatever you skipped in the exercises is a natural candidate for week 1 of the plan.
New systems fail in two ways: neglect or bloat. The second is trickier—one new connection, three speculative recipes, and five routines every week, and within a month you’re managing the system instead of working. The antidote is a pacing rule: one change per week, chosen through the observe → adjust → expand cycle.
Observe: what bothered you about using it this week? Adjust: fix that before creating anything. Expand: only if nothing needed fixing, add a new piece — the second door, the second recipe.
Accountant Túlio learned the warning firsthand: in the second week, excited, he created four routines at once. By Friday, three were sending useless alerts, and he had stopped reading all of them. He deleted all but the deadline routine and went back to the pace of one a week—and the deadline routine alone has made the system worthwhile ever since.
The 30-day plan doesn’t invent anything new — it schedules what the course has already built, from the simplest to the most delicate.
The plan outline (adapt it to your inventory)
Attorney Renata pasted her plan into the wiki’s first drawer, with a date on every line. Not for bureaucracy—because a plan without a date is a wish, and a wish doesn’t survive a week of hearings.
Practice now 0/3 done
Leave with the four weeks written, dated, and saved in the wiki—~10 min.
It's your document, in your folder: nothing is purchased, scheduled, or sent. If week 2 proves the plan wrong, you can rewrite it in five minutes — it serves you, not the other way around.
The course ends here; the system doesn’t. You leave with the first month’s map in hand—and a method for every month after that.
Summary