In four weeks, your people stop chatting with AI and start delegating complete jobs. The agent works across their files and comes back with finished work they review.
If any of these five is missing, we didn't finish.
An agent that works on their files. It opens the Word, Excel, and folders already on their computer, edits them, and saves the result where the team expects it.
Put the agent on real company systems. Not a demo account. It pulls from and writes into the same tools the team opens every day.
Skills they can rerun for jobs they already do: the steps captured once, so the agent can do the loop again without them in the chair.
Work on a schedule: recurring tasks in Claude tied to real jobs, not one-off demos. Permissions proven on a manual pass first.
Outputs in their real formats: Word, sheets, decks, or the files the team already uses, written somewhere durable a teammate can open. Not stuck in chat.
Three jobs from a single team. Before: paste into chat, copy the answer out. After: an agent that opens their files, hits company systems, and runs on the calendar. Same shape your people leave with.
One client · three jobs · real systems
The agent reads sources on disk and online, then saves a one-page briefing in the shared folder the commercial team already opens before a call.
From scattered sources → brief in the folder
Connected to the helpdesk export, the agent drafts the chase list in Excel, ready for the morning stand-up. No paste of yesterday’s dump into chat.
From raw export → queue that runs
Headcount, attrition, and open roles land as separate workbooks. The agent merges them into the monthly deck template and leaves the file where HR already expects it.
From three sheets → deck on the template
Slides for the theory you need, then a demo on real work. They watch the agent run, then run the same loop on their own files. Last 10–15 minutes are questions.
Reading and exercises, plus setup of standing context and hands-on work on the recurring file they already own. Productive work, not busywork.
People who already use AI fluently in chat and are ready to manage an agent, not just converse. Ops, sales, finance, recruiting, founders: anyone who already builds a recurring file. They leave able to scope a job, give context, and review the output with judgment.
True beginners (that is the Literacy track). People who need to build systems with Claude Code (that is the Builder track). Anyone who only wants a briefing with no hands-on work. This is not a coding bootcamp.
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 run it 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 project work on their real job. We put growing weight on the practical build. A short follow-up after the program keeps it from fading.
Arc: run it, give it a memory, teach it a skill, put it where the work lives. Open a week for what happens in the room and what they take back.
You already know how to chat with AI. This is different. You describe the outcome. It comes back when the work is done. You still own the judgment.
Paid Claude plan and desktop confirmed. First taste: Cowork reads a folder of their real files, writes a summary, then a Word document. Then an AI audit: mine chat history, then (no AI) name the recurring files they already produce.
Camera on Cowork. A well-shaped task delegated from a real folder: plan, questions, execute, a file in the shipped folder. Second session in parallel: draft, don't send. They learn where the agent runs (desktop vs cloud) so a closed laptop does not surprise them later.
They pick the winning candidate. Next week that job becomes the project.
The answer is the conversation. One question, thinking out loud. Stay in chat.
Several steps, ends in a real file, touches several tools. Hand it to the agent.
Customer-facing, money, or a decision you cannot undo. The agent prepares. A person ships.
Artifact A first real file produced from their own folder, plus 2–3 candidate jobs in the shape the project needs.
Chat forgets. Standing context remembers. Improving the brief is how they improve the work.
They draft a short standing brief (who they are, how they write, what the agent may connect, when to use chat vs Cowork).
Camera on us. A messy task comes back mediocre. They watch how we phrase the ask, catch the drift, steer, and judge. Then the same task with the brief in place. They set where finished work should land. First project is created on their real inputs.
They leave standing context the agent can return to, plus a first slice of the recurring file. The project runs through week 4.
Standing context is the brief the agent should know every time, plus a durable place finished work is written (a folder, Drive, shared drive — whatever the team already uses). We do not prescribe one folder for every session. Primary company email stays disconnected.
Artifact Standing brief filled in, a durable place for finished work, and the recurring-file project started.
By now they have delegated something real and hit real friction. This week they capture the loop so it can run again without them in the chair.
Name the project in one sentence. List three click paths worth turning into a skill (prefer one with no API). Bring a plugin recommendation from a short Cowork interview. Do not install yet.
Record one skill live: goal, steps, exceptions, done. Run a try-out on the skill we just built. Show a scheduled task firing, and the agent acting inside a logged-in system when a connector does not exist. Safety sits next to the browser beat. Rest of the room is studio on the project.
They finish the skill from the identified click path. They leave notes on what to connect and what to leave alone.
Skills are the reusable procedures they teach the agent. A scheduled task only counts if it actually runs: grant permissions on a manual pass first, or it stalls unattended. A laptop asleep is a skipped run, not a queue. Drafts before sends until proven.
Artifact One recorded skill on their own tools, checked against a try-out, plus a written call on connectors.
The work has to live in Excel, Word, PowerPoint, and Outlook if that is where the company already lives. The schedule going live is the finish line, not the demo.
Project at about 80%. Draft when to use chat vs Cowork vs Code for their own job. Submit a progress snapshot that feeds the live critique.
The cross-app chain: analysis in Excel, then a short deck in PowerPoint using their template. Two or three volunteer demos with live critique, picked for teaching value. Why primary email stays disconnected. How not to burn the plan. Decision tree edited live from the demos. Judgment stays yours, same as week 1.
Five days: the winning task is actually scheduled. The last reusable file sits where the team already expects it. If it is not scheduled, it is not done.
Cowork when the work pulls from several sources and ends in a deliverable. Claude in Microsoft 365 when they are inside the Office file, editing in place. Most substantial work uses both. Google Workspace cohorts get the same decision, on the suite they already pay for. Live dashboards that refresh from connected apps are taught only on Team and Enterprise, where they can actually be shared.
Artifact A project ready to schedule, a written when-to-use decision for their job, and a reusable output in the format they already ship.
A shareable certificate when they complete the four lessons and the scheduled task is actually live.
A report we send you: the schedules and skills the room shipped, and where a second workflow would copy the same pattern. What to do next, not a survey.
A 1-hour session after a month. What is still landing on the calendar, what needs a nudge, what the next recurring file should be.