What each leader can show at the end
Four things you can check. If one is missing, we are not done.
-
They learn how AI thinks, and how they have to think when they use it. Why it gets things right sometimes and wrong other times, what context to give it and what to check. They write it into a Project in Claude: who they are, which rules they follow and what "done" means.
-
They turn a recurring job into an agent that runs on a schedule. The same job that eats their Monday afternoon today, now named, running on their files and connected to what the company already uses, depending on what is enabled.
-
They build an agent for multi-step work using five steps that repeat: decide, test, design, equip, protect. They decide whether an agent is needed, test it against real examples, give it clear instructions and tools, and set what a person has to approve.
-
They finish with their second brain. What they know about their area (their team, their decisions and why, how they judge a good result) kept where the AI reads it before every job, plus a way to keep it current in twenty minutes a week.
Four weeks: foundations, a scheduled agent, how to build an agent, and a second brain.
01
Aprenden cómo piensa la IA y cómo tienen que pensar ellos cuando la usan.
- They understand why the same prompt gives different results, and what to do about it.
- They change how they ask: the result they want, context, examples, and they review instead of accepting the first answer.
- They save what they would always tell it in a Project, so they stop repeating it in every chat.
Aprenden cómo piensa la IA y cómo tienen que pensar ellos cuando la usan.
- They understand why the same prompt gives different results, and what to do about it.
- They change how they ask: the result they want, context, examples, and they review instead of accepting the first answer.
- They save what they would always tell it in a Project, so they stop repeating it in every chat.
I know how the tool works, and I give it instructions I do not have to repeat every Monday.
Pre-assessment: what they use, how and what for. They bring two real prompts from this week, one that worked and one that did not.
How AI works underneath and what that changes in how you ask. Advanced prompting on their own prompts: context, examples, constraints, format. They turn it into a Project with system instructions and iterate until the result holds up.
They use the Project all week. They note what they had to correct and move it into the instructions.
The same prompt can give two different results. So they check the result, correct it, and write into the Project what a good result looks like, instead of trusting the first answer.
Context, examples, constraints, format. One of their prompts, before and after, on the same job.
What the Project always knows: who it is, which rules it follows, how it sounds. Written once, not in every chat.
Deliverable A Project with system instructions, two of their own prompts rewritten, and a week of use with the corrections noted.
02
They turn a recurring job into an agent that runs on a schedule over their files.
- They see what an agent does that a chat does not: it works on files, in several steps, and comes back with the job done.
- They record every Monday's job as a named agent and call it by name.
- They put it on a schedule and decide what to connect: email, calendar, folders, depending on what is enabled.
They turn a recurring job into an agent that runs on a schedule over their files.
- They see what an agent does that a chat does not: it works on files, in several steps, and comes back with the job done.
- They record every Monday's job as a named agent and call it by name.
- They put it on a schedule and decide what to connect: email, calendar, folders, depending on what is enabled.
I give every Monday's job a name, and it comes out done without me asking.
They pick the job: it repeats every week, it has a "done" you can name, its inputs are safe to share and they own it. They bring the folder with their files and a list of where the inputs live: email, Drive, calendar.
What an agent does that a chat does not: it opens the files, does the steps and leaves the result where it belongs. They turn their Project into a named agent and run it on the folder they brought. They schedule it and decide what to connect, depending on what is enabled.
The agent runs once on its schedule without them there. They review what it delivered and write down the decision on connectors.
The step by step of that job, with their criteria inside, saved as a Skill in Claude. You call it by name; you never explain it again.
The job runs again at a fixed time, say every Monday at 7, and the leader gets the result without opening the tool.
Which systems the agent reads and which it does not. If something is blocked, one page to security: what, why, with which safeguards.
The agent runs again on its own at the time they set, and the leader gets the result. Anything that depends on a connector is taught with what the company has enabled; we do not promise connections security has not approved.
Deliverable A named agent, scheduled and run at least once on real files, plus a written decision on connectors.
03
They learn the five steps to build an agent that handles multi-step work and gets it right every time.
- They decide whether the job needs an agent or whether fixed-rule automation is enough.
- They define a good result with real examples and try the simplest version first.
- They give it clear instructions and tools, and set what it can do alone and what needs a person's approval.
They learn the five steps to build an agent that handles multi-step work and gets it right every time.
- They decide whether the job needs an agent or whether fixed-rule automation is enough.
- They define a good result with real examples and try the simplest version first.
- They give it clear instructions and tools, and set what it can do alone and what needs a person's approval.
Five steps, always in the same order: decide, test, design, equip, protect.
They bring the multi-step job that eats the most time each week, plus five real examples with the correct result already done.
The five steps on their own job: decide whether an agent is needed or fixed rules are enough, test it against their examples, design the steps, equip it with instructions and tools, and set what a person approves. They build it live and watch it run.
They run the whole job through the five steps, dated. They hand in the decision, the examples, the design, the instructions and the limits.
You need an agent when judgment is required, the information arrives messy, or the rules cannot be written out in full. Otherwise fixed automation is cheaper and more predictable.
Five real examples with the correct result. Every version is tested against them; without them you cannot tell whether it improved.
Every job has a named person responsible, and any risky or irreversible action waits for their approval.
The five steps come from the guides by Anthropic (Building effective agents) and OpenAI (A practical guide to building agents), translated into what a leader decides without writing code: whether an agent is worth it, what a good result is, how much freedom to give it, what to equip it with and what a person approves.
Deliverable An agent for multi-step work built with the five steps: the agent-or-fixed-rules decision, five test examples, the step design, the instructions and tools, and the limits with human approval.
04
They build their second brain: what they know about the business, their people and how they decide, kept where the AI reads it before every job.
- They gather into their own folder what today lives in their head, in emails and in loose notes: their team, their decisions and why, how they judge a good result.
- They ask it before a conversation or a decision and get answers with their context: what did I promise X, what did we decide about Y.
- They learn to keep it current: the agents from weeks 2 and 3 write what they did into it, and the leader adds one lesson a week.
They build their second brain: what they know about the business, their people and how they decide, kept where the AI reads it before every job.
- They gather into their own folder what today lives in their head, in emails and in loose notes: their team, their decisions and why, how they judge a good result.
- They ask it before a conversation or a decision and get answers with their context: what did I promise X, what did we decide about Y.
- They learn to keep it current: the agents from weeks 2 and 3 write what they did into it, and the leader adds one lesson a week.
Everything I know about my area sits in one place the AI reads before it works, and it grows every week.
They gather the three weeks together. They list what they re-explain to the AI every time today: their team, their decisions, their promises, their bar. They bring notes from their last three one-to-ones.
They build their second brain: six or seven documents about their area that the AI reads before every job. They ask it three real questions and compare with a chat that has no context. How to keep it current in twenty minutes a week, and what goes in and what does not.
A second dated run of the scheduled agent. They ask their second brain a real question before a meeting and hand in the answer together with the first weekly digest.
Their team, their priorities, their decisions, their quality bar, examples and vocabulary. The AI adds the rest as they use it.
What did I promise X, what did we decide about Y, how do I prepare this conversation. Answers with their context, not generic ones.
The agents write what they do into it and a digest arrives every week. The leader adds one lesson. Twenty minutes a week.
This is each leader's own second brain, with their data and their rules. Sharing it with the team, with its own permissions and rules, is a later step that does not fit in one session.
Deliverable Their second brain with six or seven real documents, three questions already answered with their context, the first weekly digest received, and proof of two dated runs.
How it runs
Live · 1 h 15 min
Just enough theory, then a demo on a real job. After that they practise on their own work, in Claude Cowork. The last 10–15 minutes are questions.
Async work · 1 h 15 min
Before the first session: the pre-assessment, a job chosen against the filter, and the folder of files it runs on today. Between sessions, real dated runs on that same job. Productive work, not filler exercises.
Before and after, on real jobs from their area
We do not teach generic cases. Each leader comes in with one of these, or their own.
Finance
Three days pulling spreadsheets from five areas→the consolidated file already done, waiting on Monday
Hunting by hand for what went off and why→the variances explained, without asking
Two hours of writing the night before→15 minutes of edits on a draft that already sounds like you
Credit and risk
"Let me look it up and get back to you"→the answer with its clause cited, on the spot
Everyone resolves them by their own judgment→the same criteria applied, whoever handles it
Asking someone else for the model and waiting two days→their own agent running the scenarios on their assumptions, same day
Their team
Twelve reviews written from scratch every half-year→twelve drafts with evidence, ready to review
Chasing everyone's progress over chat→the report ready before the meeting, every week
14 files in 14 different formats→one comparable dashboard, on a single page
Regulation and governance
Tracking down which document it was in→the answer with its source cited and what is left to confirm
Loose notes nobody opens again→agreements with an owner and a date, from the meeting itself
If theirs is not on the list, it still fits: the mechanics are the same for any job that repeats every week.
Who it is for
For
Leaders who already open Claude every day but still ask it questions instead of assigning it work, and copy and paste between files to assemble what an agent could hand them finished. They want a recurring job (credit, control desk, regulation, staffing, budget) to come out at the level they demand, repeat without being asked, and be something they can pass to their team. And they want to understand the technology well enough to decide which AI projects to sponsor in their area.
Not for
Beginners: that is what the AI Literacy program is for. Anyone who wants to build software wired into company systems: that is the AI Builder Bootcamp. Anyone who just wants a talk without opening the tool.
How we teach it
Just enough to act, then act. One short idea and a real demo; the rest of the time goes to each person's own work.
Assess
Before the start, a questionnaire for each leader. It is not a test. We ask what they use, how and what for, and which recurring job they would bring. That sets the pre-work and calibrates the room before the first session.
Coach
Each session opens with one short idea and a demo on a real job. The rest is practice on their own work, with us in the room.
Run
Between sessions, the Project and the agents run on real work, dated. Every week leaves one run they can show.
From asking AI questions to assigning it work with Claude Cowork
Today: scattered questions every Monday
- Every Monday they explain to the AI all over again who they are and what they want.
- They copy and paste from one file to another to assemble what an agent could already hand over finished.
- What they achieve cannot be passed to anyone: it lives in their chat history.
After: agents with written instructions, context and rules
- Their instructions and their bar are written once, in a Project and a named agent.
- They build agents with five steps that repeat: decide, test, design, equip, protect.
- Their second brain holds what they know about their area and grows with every job the AI does.
It does not end at the last session
Certificate
A shareable certificate on completing the four weeks and handing in the proof of recurrence.
Opportunity report
A report we send you with what the ten leaders built and what is still running: where it worked, where it did not, and what is worth extending. What to do next, not a survey.
Platform access
Exercises, materials, their Projects and their agents stay on the AdapttoAI platform after the program. The setup does not die in a chat history.