Start with one real task per person, give everyone the same written rules for your codebase, review AI changes like any other, and add a rule after every mistake. The AI stops repeating the same mistakes, because each fix is written down once. We set this up with teams and work alongside them until it sticks.
Start with real work
Pick one task from this week for each person. Real results convince people faster than a demo.
Write the rules once
Keep one shared file with how your codebase works: structure, conventions, what not to touch. Claude Code reads it before every task.
Review AI changes like any other
Every change goes through review and tests, and someone checks what it touches. Speed without checks breaks things.
Add a rule after every mistake
When AI gets something wrong, write the fix down once in the shared file. Everyone's next run starts with it.
Share what works
A short weekly note on what saved time and what went wrong spreads the good habits.
What goes wrong
| What you see | Why | Fix |
|---|---|---|
| Nobody uses it after the demo | No real task | One real task each |
| Same mistakes every week | Rules not written down | One shared rules file |
| AI changes break things | Skipped review | Review and test every change |
| Some love it, some ignore it | No shared habits | A weekly note on what works |
Set it up with us
We'll set up Claude Code with your team, write the shared rules with you and work alongside you until it's part of how you ship.
Questions
How do I get developers to actually use Claude Code?
Give each person a real task from this week's work, not a demo. Results win people over faster than training.
How do I keep AI changes safe?
Review them like any other change, run the tests, and check what each change touches.
How does the team get better at it?
Every time AI gets something wrong, write the rule down once in a shared file. Everyone's next run starts with it.
Originally published at wednesday.is.

