Managing
Preparing a team for AI is less about tool training than about judgment. AI-ready teams have agreed four things: what gets checked before AI-assisted work leaves the team, what must never go into a tool and when AI use is disclosed, which decisions stay human, and how junior people still learn the craft. Agree those first, write them down in a one-page working agreement, then train people on real tasks from their own week rather than on generic prompting tips.
The usual AI rollout goes like this. Licences are bought, someone runs a lunchtime session on prompting, and then one of two things happens. Either almost nobody uses the tools, because nobody is sure what is allowed, or everybody does, quietly, with no shared idea of what gets checked or what must never be pasted in. Both are readiness failures, and neither is fixed by more training on features.
The features are the easy part. They change every few months and people pick them up fast. What does not change is the judgment around them, and that is what a manager can actually prepare a team for.
| Area | The question | Ready looks like | Not ready looks like |
|---|---|---|---|
| Verification | Who checks the confident answer? | AI-assisted numbers are checked against the original source before they reach a client or a decision. | A fluent paragraph is treated as a checked one. |
| Guardrails and disclosure | What never goes in, and who knows AI was used? | Everyone can say, without looking it up, what information must not be pasted into a tool. Substantial AI use is mentioned when work is handed over. | Rules exist in a policy nobody has read. AI use is hidden. |
| What stays human | Which decisions do we keep? | Decisions about people, and difficult messages, are owned by a human who knows the person. | Performance feedback arrives as a polished AI draft. |
| Skill formation | How do juniors still learn? | New joiners do core work unaided before they are shown the shortcut, and reviews ask how they got the answer. | Juniors can produce the output but cannot explain or defend it. |
Notice that none of these is about which tool you use. A team strong in all four areas will adapt to new tools quickly and safely. A team weak in them will make the same mistakes with every new tool.
Involve the team early, while there is still something to decide. Ask each person to list the five tasks that take most of their week. That list is the raw material for everything that follows, and it signals that this is about their work rather than something being done to them.
Sort the team's real tasks into three groups. The point is not to be precise but to make the judgment explicit, so it stops being made silently by each person on each task.
| Group | What belongs here | Examples |
|---|---|---|
| Hand to AI, light check | Low stakes, easy to verify, mistakes are cheap | First draft of meeting notes, reformatting, summarising a long internal document for yourself |
| AI drafts, a human verifies and signs | Useful speed, but errors would reach a client or a decision | Client summaries, analysis behind a recommendation, research for a proposal, code |
| Keep human | Judgment about people, trust, accountability, or the skill you are trying to build | Performance conversations, hiring decisions, apologies, the junior's first attempt at a core task |
Policies are written by legal teams and read by nobody. A working agreement is written by the team, fits on one page and covers six questions:
Once the agreement is drafted, find out how far the team's habits are from it. The free AI judgment check asks twelve questions across verification, guardrails, what stays human and skill formation, and gives you a written report naming the weakest area. Run it with the whole team, separately, and compare the answers.
Generic prompting sessions produce enthusiasm and little change. Replace them with short sessions on the team's own tasks from step two: take a real client summary, produce it with AI, then check it together against the source and count what had to be corrected. People learn more from finding one confident error in their own work than from any list of tips, and it is critical thinking practised where it counts. Our guide to AI training for non-technical teams covers the format in more detail.
This is the step most plans miss. Junior people have always learned by doing the first draft badly and being corrected. If the tool now does the first draft, they can produce acceptable output without ever building the judgment to know when it is wrong. Decide which core tasks juniors still do unaided, and when reviewing their work, ask how did you get here? and not only is this right?
Hold a short retrospective: what did we hand to AI, what errors did we catch, what nearly got through, and what should change in the agreement? Expect to revise it. A working agreement that has never been edited is one nobody is using.
The tools change every few months. The judgment around them is what lasts, and it is the part you can teach.
The useful question is not whether AI can do a task, but whether you want a person accountable for it. Judgment about people, building trust, taking responsibility and the hard conversations that hold a team together stay human, not because the tools cannot produce the words, but because the words are not the point. We set out the full split, task by task, in The New Division of Labour.
Individual habits matter as much as team rules, and our guide on how to use AI responsibly at work sets out five of them. Together they make up what is increasingly called AI literacy. If you want each person to see how wisely they use AI themselves, the AI wisdom check puts fifteen real scenarios to them, from a confident market-size figure due in an hour to a colleague pasting confidential financials into a free chatbot. Psychological safety matters here too: people only admit an AI-assisted mistake early in a team where admitting mistakes is normal, as the glossary entry on psychological safety explains.
Tour De Force runs this as live, experiential training for teams worldwide, online and in person. Talk to us, or play The Weekly Challenge from our gamified learning apps to see the method in ten minutes.
Questions
Start with the judgment rather than the tools. Triage the team's real tasks into what can go to AI, what AI can draft for a human to verify, and what stays human. Write a one-page working agreement on what never goes into a tool, disclosure, verification and sign-off. Then train on real tasks from people's own work, and protect the tasks juniors still do unaided.
An AI-ready team checks confident answers before they reach a client or decision, knows what information must never be pasted into a tool, discloses substantial AI use, keeps decisions about people human, and still builds core skills in junior staff. Tool fluency is part of it, but the easiest part to acquire.
Usually not. Bans tend to push use onto personal accounts with no guardrails at all. A short, clear agreement on approved tools, what information is off limits, and what must be checked is safer and easier to follow.
Ask about specific behaviours rather than attitudes: whether AI-assisted numbers are checked against the source, whether people can name what must never be pasted into a tool, whether AI use is disclosed, and whether juniors still do core work unaided. The free Tour De Force AI judgment check covers these in twelve questions.
Critical thinking to question fluent answers, verification habits, clear communication, judgment about people, and the underlying craft of their own job, which is what lets them tell when the tool is wrong. These are the skills AI raises the value of rather than replaces.
Keep reading
Twelve questions on whether anyone still checks the confident answer, and what your team keeps human.
GuideWhere to start, and why real tasks beat prompting tips.
EssayWhat AI does well, what stays human, and the line between them.
Get in touch
Book a 30-minute discovery call and we'll shape an experiential programme around your goals.
Book a discovery call →