30 to 70 PRs a day: How we managed to not wreck our systems
summary
Data and concrete changes decide whether to watch not buzzwords Here is what the session actually claims and shows over eighteen months at Honeycomb a monorepo goes from about 30 to about 74 merges per weekday with two thirds AI attributed PRs The talk documents measurable tooling work not promises like CI time cut from ~12–15m to ~6m on dev branches, feature flags, agent instrumentation, modest incident rise and practices that kept deployments working and auditable
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:40AI adoption challenges
- 8:14AI tools and team changes
- 11:26Managing AI code surge
- 17:27Speeding up CI
- 23:26Capturing design in AI workflows
- 27:06Improving systems with capacity
- 31:01Rethinking code review
- 34:54When to apply and avoid AI tools
- 36:19AI adoption prerequisites
- 38:50Avoiding AI dysfunction
- 40:12Conclusion
related talks
23mFrom Blind Spots to Merged PRs: Continuous Agentic Performance Optimization
43mWriting for Machines: How To Capture Your Project's "Vibe" and Survive AI Slop
32mMerge Strategies at Scale: Developer Happiness Through Better Code-submission Workflows
42m7 Metrics to Accelerate Delivery without Damaging Culture
59mTech leadership, one year after
29mLeadership in AI-assisted engineering
45mPeople and Culture
28mSoftware Factory in Public: 375 PRs in 10 Days