AI coding agents straining CI/CD infrastructure

Linear published a blog post about reworking their CI pipeline specifically because agentic coding tools were generating so many more commits and build triggers than their old system could handle. The HN thread picked this up fast, with engineers sharing their own Bazel setups and pointing to vitest load-balancing as a partial fix.

This is a new infrastructure problem category. The bottleneck isn't model quality or agent capability anymore, it's the downstream systems that have to absorb the output. CI pipelines, test suites, and review queues were designed around human commit rates. Agents don't pace themselves.

Several commenters noted this is pushing teams toward more aggressive caching strategies and custom runner configurations. Bazel keeps coming up as the serious solution for large projects, but it has a steep setup cost that smaller teams can't easily justify.