The short answer.
Train your team on AI in two layers: a short set of ground rules every person learns once, and training on the specific workflows each role runs, taught by the person who owns each workflow, on the team’s own past cases. The ground rules cover what goes into ChatGPT, Claude, Copilot or Gemini, how output gets checked, and where the team’s workflows live. The role training is where people actually learn, because the workflow repeats the lesson every time the work comes round.
Training sits in the third step of a plan: plan it, develop the workflows, deploy them. Is it OK to let employees figure out AI themselves? covers what the company owes people beyond freedom to explore. This piece covers the training itself.
Why general AI training fades.
General AI training fades because nothing in the trainee’s week asks for it again. A session on how to write a good prompt, or a tour of what an assistant can do, gives people a set of examples from somebody else’s job. On Monday the person returns to their own queue, the examples do not match it, and the session becomes a memory.
Training on a workflow holds because the workflow arrives with the work. A scheduler who learns the weekly schedule-change workflow runs it the next week, and the week after, on real cases. Every run, and every check against the team’s standard, repeats the lesson. The question to ask of any training is simple: on which day next week will this person use it, and on what piece of work? If nobody can name the day and the work, the training will fade.
What to teach everyone, once.
Everyone in the company learns the same five ground rules once, in one session, from the person who owns the AI plan. The ground rules are the same for every role, so teach them once and keep them on one page.
- Which assistant the company uses.The company’s own ChatGPT, Claude, Copilot or Gemini account, and how to sign in to it. Work goes there, not into a personal account.
- What may go in, and what may not.The kinds of records the company allows in its assistant, and the kinds it keeps out. Write both lists in plain words.
- A workflow prepares; a person decides.Every output is a draft until the named person checks it. Nothing leaves the team, reaches a client or changes a record on the strength of the assistant alone.
- Where the team’s workflows live.One shared place, with one current version of each workflow. People run that version, not a copy saved last month.
- How to report a miss.When a workflow gets something wrong, the person who caught it writes down the case and tells the workflow’s owner. A reported miss is how the workflow improves.
That is the whole of the general training. Exploring on top is welcome; the ground rules are not optional.
What to teach by role.
Each role is trained on the two or three workflows that role runs. The role’s workflows come straight from the plan: How to plan AI by role sets out what each role should be able to do with AI. Role training teaches four things for each workflow.
- What it starts from.The export, the queue or the file the workflow reads, and how to get it.
- How to run it.The steps, in the company’s own assistant, on a real case.
- What good output looks like.Two or three past examples the team agrees are right, side by side with the workflow’s draft.
- What the person still decides.The exact point where the person reads, changes and approves, and what makes them stop and ask.
Teach on the team’s own past work, never on sample data. Take last month’s real cases, run the workflow on them, and compare the drafts with what the team actually sent. People learn the standard by seeing where the draft differs from it.
The workflow’s owner trains, not an outside trainer.
The person who owns a workflow trains the people who run it. The owner knows the standard and reads the record of misses, so the owner is the one person who can answer “is this draft right?” for that piece of work.
The owner trains in three moves: shows the workflow on a past case, watches each person run it on a second past case, then reviews each person’s first live runs before the output goes anywhere. After that, the ordinary review of the work is the training. When a new person joins the role, the same three moves apply; How new hires learn the AI workflows their role runs covers that case in full.
A role training sheet to copy.
Copy these headings onto one page for each role. The workflow’s owner fills it in before training starts, and it stays with the role.
- Role and team.The role, named for the work, and the team it sits in.
- Workflows this role runs.Two or three, by name, each with its owner.
- What each workflow starts from.The system, export or file, and who can get it.
- Past cases to train on.Three real cases from recent weeks, with the output the team actually sent.
- What good looks like.The points the owner checks in every draft.
- What the person decides.The approval point, and the cases that always go to the owner.
- First live runs reviewed by.One name.
- Ground rules session attended.Yes, or the date it is booked.
- Signed off to run alone.The owner’s name and the date.
One home health agency trains its office.
Take a 60-person home health agency whose office runs four teams: intake, scheduling, billing and caregiver recruiting. The office manager owns the AI plan. The training covers office and scheduling work only.
Illustrative example.
- Everyone, once.The office manager runs the ground rules session for all office staff together. Its page states which records stay out of the company’s assistant, names the shared folder where the workflows live, and says that every draft is checked by the named person before it is sent.
- Scheduling.The scheduling lead owns two workflows: drafting the week’s schedule changes from the visit calendar, and drafting the messages to caregivers about open shifts. She trains both schedulers on last month’s schedule changes, compares each draft with the schedule they actually published, and reviews their first live weeks.
- Intake.The intake coordinator owns the referral summary workflow, which turns an incoming referral into a one-page summary for the office. He trains the two intake staff on past referrals and keeps the record of misses.
- Caregiver recruiting.The recruiting lead owns the applicant follow-up workflow and trains the recruiting assistant on past applicants’ messages.
- Billing.Billing has no workflow yet, so billing staff attend the ground rules session and nothing more.
Nobody in the office sat through a general course on prompts. The open-shift messages began from a free workflow template in the Workflows Directory, adapted to the agency’s own wording before anyone trained on it.
What is different in a company of under 200 people.
In a company of under 200 people, the trainer is a colleague, not a department. A large company assigns a bought course to everyone. You have a lead for each team who already knows the work.
- One session covers the ground rules.The whole company, or the whole office, can learn the ground rules together, from the person who owns the plan.
- The owner often runs the workflow too.Training and review become the same conversation.
- Train as each workflow is ready.Role training follows the plan’s order. A team with no workflow yet learns the ground rules and waits, rather than sitting through training it will not use.
Start from the plan, and the training follows from it. FullyPowered Planner is where the plan lives: you set out your teams, their roles and the work between them, choose what to build first, and your colleagues add their own teams, so every role’s training sheet starts from the workflows on the plan. How organizations plan their use of AI covers the rest of the method.