The lede
AI coding has made CI a bottleneck, so we reworked ours to keep up
Mufeez Amjad, Linear · made the Sunday
Linear's test suite nearly quadrupled since January and is still taking on roughly two thousand tests a week, which is what agent-speed code generation does to the one part of the pipeline that did not get faster. The write-up is unusually specific about where the minutes actually went: third-party runners with better CPUs bought 34% (tsc alone dropped 52%), the change-detection gating job came down from a 26-second median to 8, preinstalled dependencies and a narrowed monorepo install cut per-shard setup by 44%, and batching saved about 87,000 runner-minutes a month. The headline number is almost comic — six minutes of PR wait became just over five — until you notice the counterfactual, which is that the same suite would now take eleven. This is the shape of the next two years of infrastructure work for anyone whose engineers ship with agents.