TechLeadConf 2026London
One Team, Many AIs
Keeping engineering teams aligned as AI workflows pull them apart
Every developer on a team now works with their own AI, in their own way. Code gets written by an AI that has the whole conversation, then reviewed by a different AI that only sees the diff. Each step looks reasonable, yet nobody really decided what got shipped.
This talk follows one pull request through two AIs to show the new kind of drift that creates, why it matters more now that AI reviews our code, and what tech leads can do about it without limiting experimentation: align what every AI knows, not how each person works.
Key ideas
One context for every AI
Rules live in the repo, with the reason behind them, where every AI that writes or reviews can read them.
The whole team owns the context
Product, engineering and security each own the rules they know best, with owners and changes through PRs.
Humans decide where AI can review
AI review is enough for low-risk changes. Payments, auth and personal data always get a human code owner.
Review decisions, not diffs
Humans can’t read every line, but they can read every decision. PRs should say what was decided and why.
