If you ask most owners whether their company uses AI, they'll describe a project — something being evaluated, something on the roadmap, something a vendor pitched last month.
If you ask their employees, you'll often get a different answer. Somebody in accounting has been using a chat assistant to draft collection emails. Someone in operations uses one to clean up meeting notes. A salesperson rewrites proposals with it. A manager asked it to summarize a contract last week and didn't mention it to anyone.
None of this is sinister. It's what capable people do when a useful tool shows up: they try it on their own work. But it means that for many businesses, AI adoption isn't a future decision. It's already happening, informally, with no guidance about what's safe, what's good, or what's a waste of time.
The risks of the unofficial version
Informal adoption has three problems, and they get worse the longer nobody addresses them.
Data goes where it shouldn't. A client's financials pasted into a consumer tool. A patient's details in a prompt. Nobody meant any harm; nobody told them not to. For some businesses — the ones we wrote about in When the Data Can't Leave the Building — this is a real compliance exposure.
Quality is invisible. Some people are getting excellent results. Others are getting confident nonsense and sending it to clients. Without any shared standard, you can't tell which is which until something goes wrong.
The good ideas stay private. The person who figured out a great way to use AI on a weekly report keeps it to themselves, maybe because they're not sure it's allowed. The rest of the team keeps doing it the slow way.
Why "ban it" doesn't work
The reflex response is a policy memo: no AI tools, full stop. It's understandable, and it almost never works. The tools are on everyone's phone. Banning them mostly guarantees that the usage continues, just less visibly — which is worse on every one of the three problems above.
The better move is the one that works for any tool your team adopts: make the right way the easy way.
What teaching them actually looks like
Good AI coaching isn't a one-hour webinar about the future of work. It's practical, specific to your business, and aimed at the tasks people already do.
A short list of what's allowed where. Which tools are approved, what data can go into them, and what never can. One page. Plain language. Real examples from your business.
Hands-on sessions built on real work. Not generic prompts — your team's actual emails, reports, proposals, and documents, and how to get good results on them. People learn fastest on their own work.
A standard for checking output. AI drafts; people verify. What that means in practice: the claims, numbers, names, and anything going to a client get checked by the person sending it, every time. It's the same boundary we build into every AI employee.
A way to share what works. A shared set of prompts and examples, and a regular moment where people show each other what's saving them time. The quiet expert in accounting becomes the person who teaches everyone else.
A path from personal use to real automation. Some of what people discover will be worth more than a better prompt. When the same task keeps showing up, that's a candidate for a proper build — scoped, measured, and supervised.
Why this is the right first project
For a lot of businesses, coaching is the cheapest AI project they can run and the one with the fastest payoff. No integration. No build. Just better use of tools people already have, on work they already do.
It also makes every later project better. A team that understands what AI is good at — and what it's bad at — gives better input on what to automate, trusts the result more, and adopts it faster. We've argued in Your Team Won't Adopt What They Didn't Design that adoption is where most automation succeeds or fails. Coaching is how you build the team that designs well.
And it keeps the focus where we think it belongs: on making good people more effective, not on replacing them. The best results come from the people who know the work best, now with a tool that takes the tedious half off their plate. That's the argument of AI Won't Replace Your Best Person, applied one person at a time.
Where to start
Ask your team — without judgment — how they're using AI today. You'll learn more in one honest conversation than in a month of vendor demos. Note what's working, what's risky, and what keeps coming up.
Then decide what you want to formalize: the safe-use list, the training, and the handful of recurring tasks that might deserve a real build. Our AI Coaching work is built for the second: hands-on, on your team's actual workflows, not generic webinars. For the third, the AI Automation Scorecard will tell you which of those tasks are worth automating and which are better left as a well-taught habit.
Your team has already started. The only question is whether anyone is going to help them do it well.