The situation
Three teams, fourteen developers. Some use Claude Code, some Codex, some Copilot. Each has their own prompts, their own rules file, their own idea of what good looks like. Code review has become an argument about style that the agents keep reopening. Mandating one tool caused a revolt, and didn't fix it.
Why the usual fixes don't hold
| The usual fix | Why it doesn't hold |
|---|---|
| One mandated coding tool | People switch back, and the rules still live in each person's head |
| A shared prompt library | It's copied once and drifts; nobody knows which version ran |
| A rules file per repository | Each repository drifts on its own; a new rule takes weeks to reach all of them |
| Stricter code review | Reviewers find the same problem again and again, after the code is written |
Five steps that hold
- Write a standard once, at the right level. Organisation, product, role or skill. "Every API returns errors in the same shape" belongs to the organisation, not to one repository.
- Send it with the work. Each task carries the standards and decisions in force, with their reasons, to whichever agent runs it. Same words for every tool.
- Put it in every repository too. Any coding tool opened in a repository should find the organisation's rules there, kept current by one upgrade, not by hand.
- Run one version of each agent. Every team uses the same agent and skill versions, and a new version is tested before teams get it.
- Let review teach the standards. A change that breaks a standard goes back with a note naming it, and the next attempt reads the note first. When the same note keeps coming back, it becomes a new standard.
A test you can run this week
Give the same small task to three developers using three different tools. Compare the results for error handling, naming and tests. Every difference is a standard nobody wrote down.
With Wakalix
Decisions in force travel with every work packet, the same for Claude Code and Codex. Every repository carries the organisation's rules for any coding tool, and wakalix upgrade brings each checkout current. Agents and skills are versioned and tested, reviews send work back with notes, and upkeep proposes new standards from what reviewers keep saying. See Uniform AI and the walkthrough One standard, every team.