The tools are in. Will the work change?
A note on the distance between a workforce that has been trained on AI, and a function that actually runs on it.
Most enterprise AI rollouts run the same script, and it’s a sensible one. Licences get bought. A vendor runs the training. People learn the features: the prompt that drafts the email, the one that summarises the meeting, the one that cleans up the deck. Usage gets measured. A few champions get named.
By month three, the adoption dashboard looks healthy.
Then.
Somewhere between month three and month six, you notice the number you actually care about hasn’t moved. Adoption is up. The work is the same. The cycle that took thirty-odd months still takes thirty-odd months. People are using AI.
The work didn’t change.
The cause isn’t the training. It’s what the training was designed to do. A workshop delivers a skill, how to operate the tool. It was never built to do the two things that actually move a function: hold a habit past the quarter, and rewire the workflow so the time saved compounds into something the business can see.
Personal productivity and functional transformation are not the same project. They aren’t even the same design.
Two things become clear when you’ve watched enough of these programmes from the inside.
Industrial · Manufacturing
What made AI stick on the floor wasn’t the classroom. It was that people learned on their own live work — instrumented, timed before and after — so the hours saved were a number on a real task, not a self-reported guess on a survey. The skill carried production stakes from day one. Nothing was practised on a toy example.
There’s a layer the standard rollout doesn’t name.
Training covers the skill. Governance covers the rules. Champions cover the evangelism. But the part in between, the structured practice that turns a skill into a habit, and a habit into a workflow that is genuinely faster, is just assumed to happen on its own.
It doesn’t happen on its own. It has to be built.
Call it the practice layer. It’s the whole difference between a workforce that can use AI and a function that runs on it.
KEY TAKEAWAY: So we stopped leaving that layer to chance and started building it on purpose. Four moves, run in sequence. None of them a slideshow.
01
Start at the work, not the tool.
We sit with the people closest to the job to find the repetitive tasks that quietly eat the week — then build a prompt library in their own terminology, on their own workflows.
02
Measure the work, not the logins.
The same real tasks, timed before and after, quality scored blind. The metric is the work itself — and it can’t be gamed by opening the app.
03
Give the champions a structure.
A named role with time and a remit, so adoption has an owner after we leave. Governance sized to accelerate, not to gate.
04
The AI Gym.
Short, timed, hands-on challenges drawn from real jobs — done at their own desks, debriefed, repeated until the new way feels normal. Skill becomes habit through reps, the way a gym builds muscle.
What changes when the practice layer is in place is structural, not motivational. The time saved stops being a line on a dashboard and starts showing up inside the workflow. Individual productivity compounds into functional output. The pipeline of new use cases keeps generating itself, because the people doing the work are now the ones who see them.
Month six looks better than month three.
If you’re standing at the start of a rollout — licences in hand, training about to begin — the question worth sitting with isn’t whether people will use the tools. They will, for a while. It’s whether the work itself will be different a year from now.
That’s a design decision. It’s the one most programmes skip.
We’re always happy to compare notes — no agenda attached. The most useful conversations are usually with people running this same problem on their own ground.
