The 3Es: Educate, Experiment, Establish

Tools do not become habits by being available

Many teams already have ChatGPT, Claude, Copilot, or some combination of all three. The subscriptions exist. The behaviour does not. That is not a motivation problem. It is a sequence problem.

The 3Es are a deliberately boring way to solve it: Educate, Experiment, Establish. One job at a time.

Educate: give the tool a job

Education is not a long course. It is 60 to 90 minutes with one tool, one real task, and a clear explanation of its limits. A team member learns what a good instruction looks like, what information must not be shared, and how to check an output before using it.

Choose a task that happens every week. A task that exists only in a training exercise will not survive contact with a busy calendar.

Experiment: test it in the real workflow

Use the new approach for two weeks. Keep the existing way of working available so people are not forced into a brittle process. Track the time involved, the quality of the output, and the moments where a person had to take over.

This is where assumptions meet reality. A prompt that looked excellent in a workshop may fail when the input is incomplete. Good. Now you know what needs to change.

Establish: make the useful bit ordinary

If the test works, document it in the place where the team already works. Give it a name, a short set of instructions, a named owner, and a five-minute weekly review. The workflow does not need a grand launch. It needs a home.

If the test does not work, stop. A discarded experiment is cheaper than a permanent tool nobody trusts.

Why the order matters

Teams often try to establish first. They buy a tool, announce access, and ask people to use it. Then the tool becomes another tab and the business concludes that AI did not work.

Educate first. Experiment next. Establish only what has earned a place. This keeps the focus on better work, not busier software.

Next step: Pick one repeated task and schedule a two-week experiment, with one owner and one review date.