Reliability

30 to 70 PRs a Day: How We Managed to Not Wreck Our Systems

Honeycomb doubled its PR rate with AI and incidents tripled. Its own accounting says the multiplier is not the story; the deploy and ownership practices are.

· honeycomb.io ↗ · 23m read

Illustration of a developer at a laptop beside a code document, a Honeycomb hexagon logo, and a rising bar chart

Why we picked it · the editor's summary

Fong-Jones gives the whole ledger: peak weekday merges from about 30 to about 74 between early 2025 and April 2026, at least 82.6% of new lines attributed to AI by June, the codebase from 0.97M to 2.1M lines in sixteen and a half months, and incidents from a 2024 baseline of 18.5 a quarter to 53 in Q2 2026. Her reading is that incidents tracked change volume linearly, with no catastrophic AI-caused outage, and that the absorbers were already there: continuous delivery, fast CI, observability that ties a line of code to production, and blameless review. The hedges are stated, including that attribution is a floor and that other org changes ran concurrently. Her line is that AI amplifies what you already are. Ours is that the substrate is the investment no roadmap funds until after the incidents triple.

Read on honeycomb.io ↗

Comments

Loading comments…

The Friday deploy, in your inbox

The week’s best writing on shipping software, picked by people who ship.

We’ll email you a confirmation first. Your address is used only to send you this newsletter, and every issue has an unsubscribe link. Privacy policy