turned on the remote cache for our turborepo at 11pm out of pure spite. CI went from 14 minutes to under 4. going to bed.
// unpaid, unprompted, mostly typed at night
11m 40s2m 06s
Median CI time across 212 public repos that switched in Q1 2026.
turned on the remote cache for our turborepo at 11pm out of pure spite. CI went from 14 minutes to under 4. going to bed.
honest review: the setup docs assume pnpm. we're on yarn 1 like cavemen, so it took 25 min instead of 5. still the best afternoon of the quarter.
PR checks now finish before I'm done writing the PR description. That's not a speed bump, that changes how the team works.
we had one engineer whose whole job was babysitting the Jenkins box. he works on the product now and is happier about it than any of us.
ferrule why just told me a single .env.example was busting the cache for 40 packages. Two years. Two YEARS.
CI bill, February: $4,212. March, with Ferrule: $1,380. Our CFO replied with a thumbs up, which has never once happened.
Migrated 3 repos off self-hosted runners. Deleted 1,100 lines of bash. Nobody has asked where they went.
the free tier is 10 GB of cache, which sounded small until I checked and we were using 3.2 of it
Our flaky-test quarantine list shrank after we moved to hermetic runners. I did not touch a single test. Slightly offended, honestly.
the cache hit chart is the first CI graph I have ever opened on purpose
small thing, but replaying a failed CI job on my laptop, same env, same cache, same failure, is actual witchcraft
first time I saw a 600 MB docker layer come back as a remote hit I laughed out loud in a quiet office