What agents make easier across repo boundaries
An agent can propose a consistent change across a dozen polyrepo services about as easily as across a dozen packages in one monorepo, provided it has visibility into all of them — something a human making the same change would find prohibitively slow. This genuinely reduces one of polyrepo's biggest historic costs.
It does not reduce the operational cost of coordinating a dozen separate deployments and release trains, which remains a largely human and process problem regardless of who wrote the diffs.
Where the new risk concentrates instead
Giving an agent visibility and write access across many repositories concentrates a lot of blast radius behind one planning step gone wrong — a single bad plan can touch far more surface area than a human typing changes by hand ever could in the same timeframe. That argues for stronger per-repository review gates, not fewer, even as cross-repo changes get structurally easier to propose.
Agents make cross-repository changes easier to propose. They make a bad cross-repository change far more dangerous to approve carelessly.MNB Software Engineering Practice
Key actions
- Give agents clear visibility into related repos before cross-cutting tasks
- Keep per-repository review gates strong even as proposals get easier
- Treat a single bad cross-repo plan as a high-blast-radius event
- Don't conflate easier proposal with reduced deployment coordination cost