The usual way to introduce AI coding tools is a launch: a slide deck, a demo where everything works first time, and an email with licence keys. A month later, half the team uses the tool for everything, half refuses to touch it, and nobody has agreed how its output should be checked.
There is a better first step, and it fits in an hour.
Start with something going wrong
Demos show AI at its best. Your team needs to see it at its most convincing and still wrong, because that is what they will meet in real pull requests.
So make the first session hands-on. Give people, in pairs, a feature an assistant wrote with every test passing, and a user report saying it is broken. Ask them to find out why before they read the code. Then give them three AI-written patches and ask them to accept, fix or reject each one, with evidence.
People come out of that hour with a healthier attitude than any deck produces: the tool is useful, and its output needs checking, and here is how to check it.
Agree three rules in the same hour
Use the last fifteen minutes to agree three things, and write them down:
- What never goes into an AI tool here. Secrets, customer data, anything your contracts restrict.
- Who owns an AI-written change. The simplest answer works: whoever opens the pull request.
- Which changes get extra review. Name the parts of your own codebase that are high risk.
Do not try to write a full policy in the room. Three agreed rules beat a long document nobody read.
Then pilot, don't launch
After the session, let two to four volunteers use the tool on real work for a few weeks. They label AI-assisted pull requests and meet briefly each week to share what worked and what broke. Then the whole team adopts it, with rules shaped by people who tried them.
Collect a few numbers before you start, such as time to merge and changes fixed again soon after merging, so you can tell later whether anything improved.
What to avoid
- Mandating usage. Measure delivery, not how much each person uses the tool.
- Skipping the sceptics. They usually make the best reviewers. Put them in the pilot.
- Treating it as a one-off. Look at the rules again after a month. The tools change quickly, and so will your team's habits.
Run it yourself
We wrote the hour up so you do not have to plan it. The free one-hour team session is a facilitator guide with a short script, two exercises on a ready-made practice repository, an answer key, and a guided start on your team's rules. For the pilot and the numbers, see measuring AI impact.