In four weeks, your people go from generic chatbot answers to a role assistant they use on real work. The method stays in the company when we leave.
If any of these four is missing, we didn’t finish.
Prompts that hold up on any model. Same structure, consistent results, not a lucky chat.
A role assistant they use every day. Projects in Claude and ChatGPT, Gems in Gemini. The rules live in the setup, not in someone else’s head.
Judgment on when to trust AI. What to share, how to check the answer, when a person still has to sign off.
Reports, presentations, and weekly packs produced with AI, on their actual files. Not a sandbox.
Three examples from a single cohort. An operations team that already had chat AI open, with no shared method. Same approach your people would use on their own work.
One client · three assistants · real processes
Credit spent the week chasing the same overdue accounts by hand. They built an assistant that drafts emails by firmness, WhatsApp messages, and call scripts from their own rules. The chat starts from the playbook, not a blank page.
From ad-hoc chase → graded pack
Service tracked orders client by client and only heard about delays when someone called. Their assistant flags late or at-risk deliveries from the week’s list. A person still decides; the miss stops being a surprise.
From reactive inbox → flagged risk
Visits and sell-out files were rebuilt into slides every week. They put the rules in an assistant and attached this week’s file. Direction gets a short board by rep without someone starting from a blank deck.
From manual board → weekly summary
Slides for the theory you need, then demo on real work. They watch the loop. They practice on their own job between sessions. Last 10–15 minutes are questions.
Reading and exercises, plus building their assistant and feeding it a safe document. Practical work is the weight we grow over time. Productive work, not busywork.
The whole team. Beginners, operators, coordinators, analysts, office managers, and anyone new to AI. No prior knowledge. No code. People who will open the tools and practice on work they already do.
People who already hit the ceiling of chat and need Builder. Anyone who only wants a briefing with no hands-on work. Software engineers who already ship with agents.
Learn enough to act, then act. A tight briefing on the idea, then studio time where the agent does real work on their stack.
Before kickoff, a pre-program assessment for every participant. We gauge knowledge, attitude (who is ready vs who will push back and why), map what they actually do day to day, and map where the opportunities sit. This is how we personalize the room before anyone joins a live session.
Live sessions mix slides for theory with demos on real work. They watch the loop, then practice on their own work with us in the room. Length and count follow the format above.
Async between sessions: reading and exercises, plus setup and hands-on work on their real job. We put growing weight on the practical build.
Four weeks. Four things your people can show you. Open a week for what happens in the room and what they take back.
The model is the same for everyone. The prompt changes the result.
Short reading. Four words they will use all month: probabilistic, context, prompt, iterate.
Why answers vary. What AI is good at and bad at, honest both ways. Then the 5 pieces on a live task. They watch one weak answer get iterated in front of them. Last 10–15 minutes are questions.
They write a 5-part prompt for a task from their own function and run it. Next week they stop rewriting the rules every time.
What to do, and who the model is supposed to be while it does it.
What it needs to know, and what it must not invent or skip.
How the answer should look, so it can drop into real work.
Artifact A 5-part prompt for a real task in their job.
The chat forgets. The setup remembers.
Reading on the prompt you do not write every time. They bring one repetitive task from their own work.
What a system prompt is. Which levels they control. We build a reusable Project live in two tools from the same briefing. The chat is the task of the day. The config holds role, context, limits, format.
They build their own Project or Gem for that repetitive task. If their plan has no Project access, personal instructions as a fallback.
We teach Projects (Claude and ChatGPT) and Gems (Gemini). We do not teach Custom GPTs. Those need a paid ChatGPT plan. The principle is the same. The button changes.
Artifact A role assistant for a repetitive task they already do.
Your assistant knows how to work. Today it learns what to read.
They finish personal instructions and their first assistant. Short reading on documents versus instructions.
Attach a file versus put a document in the knowledge base. A new chat still knows the policy. Then the same data becomes an HTML report and slides. Then privacy: public, internal, confidential, personal. If in doubt, do not upload. Ask first.
They feed their assistant one safe document. They turn off model training on personal accounts.
Policy and rules that do not change this week. Upload once. The assistant keeps them.
This week’s spreadsheet, this client’s file. Attach to the chat. Do not park it forever.
Do not upload. Ask first. Privacy is part of the work, not a slide at the end.
AI reduces effort, not verification. The report is a draft until a person checks it.
Artifact An assistant that consults a real (safe) document, plus a first AI-made report they can show.
The last week is not more theory. It is their job, named, and a way to verify.
They upload a first document. They walk through training opt-out. They read the three roles. They send three real tasks from their job.
Recap by example. Deep research inside the strategist role, on a free-tier path. Capstone: they classify their own task and fill a prompt template. Verification: ask for the source, check what you can check, give it to another AI to attack.
One concrete action that week on the task they picked. Next work starts from the assistant and the playbook, not a blank chat.
Draft the email, fill the template, do the recurring pack. You still own the send.
Start the deck, the catalog, the first page. You still edit before it leaves.
Research, compare, recommend. You still decide. Check sources before you act.
We do not teach Claude Code, Cowork, or connectors here. That is the next track. This week they leave able to use AI on their job with judgment.
Artifact Their own task classified (assistant / creator / strategist), plus a prompt they can run on Monday.
A shareable certificate when they complete the four lessons and the capstone task is in.
A report we send you: use cases surfaced during the program, scored for effort and impact. What to do next, not a survey.
Exercises, materials, and their assistants stay in the AdapttoAI platform after the program. The setup does not die in a chat history.