• Sources: primary, discussion
  • Summary: Mufeez Amjad published the work on 2026-09-21 with numbers for each change: PR wait time fell from over 6 minutes to just over 5 and runner time per test roughly halved, while test suites almost quadrupled since January at roughly 2,000 tests added per week. Moving off GitHub Actions to third-party runners made jobs 34 percent faster on average and tsc 52 percent faster, tsgo cut the weekly median tsc check by 73 percent, rewriting type-aware lint rules to pure AST analysis cut full-repository lint time 55 percent and enabled the later Oxlint move, capping fetch depth took the slowest change-detection gate from 94 to 20 seconds, and consolidating seven small checks into two saved roughly 87,000 runner-minutes per month, 11.8 percent of total CI usage. The largest single win was an opt-in Vitest project with isolate: false, worth about 17 percent of monthly spend with the slowest shard falling from a range of 300 to 379 seconds down to about 195, and also the highest correctness risk, so they updated their agent skills to make generated tests respect the opt-in constraints.
  • Why it matters: Sharding only pays once per-shard setup cost is low, and where agents write most of the tests the constraint has to live in the agent instructions rather than in review.