AI coding has made CI a bottleneck, so we reworked ours to keep up

	AI coding has made CI a bottleneck, so we reworked ours to keep up(linear.app)
	310 points by julian_digital 22 days ago \| 384 comments
	tl;dr: Linear reworked their CI pipeline to handle AI-accelerated code output, cutting PR wait time from 6+ minutes to just over 5 despite quadrupling their test suite. Key changes included switching to faster third-party runners and tsgo, removing TypeScript dependencies from lint rules, trimming critical-path jobs (sparse checkouts, scoped pnpm installs, prebuilt CI images), and parallelizing Vitest into 8 shards with opt-in non-isolated execution. The Vitest change alone cut monthly CI costs ~17%, and consolidating small checks saved another ~12% in runner-minutes.
	HN Discussion: ↓AI-accelerated coding hasn't produced better products, suggesting the speed gains are illusory ↓CI isn't the real bottleneck; human review and validation is what actually matters ↑Agrees GitHub Actions is slow/unreliable and moving to third-party or self-hosted runners is the right call ↓AI-generated tests are largely useless boilerplate, so faster CI just runs more junk •Adds technical suggestions or questions (Bazel, peering issues) without judging the article