Tag
On February 19, 2026, Clerk’s session failover didn’t trigger because Postgres was “technically still online” — degraded enough that queries returning 200 took minutes instead of milliseconds.
GitHub's August 17 outage stretched to nearly eight hours partly because retries amplified the failure they were meant to survive. The fix has existed for a decade. Most client libraries still skip it.
When Coinbase's ingress gateway failed on 14 July 2026, the path engineers needed to reach their own rollback tooling ran through that same gateway. AWS hit an identical loop nine months earlier.
Most engineering metrics measure activity, not outcomes. As AI coding tools inflate the numbers further, it is time to be honest about what actually signals a healthy engineering team.
Most engineering teams measure on-call fairness by rotation count. That metric is easy to generate and tells you almost nothing about who is actually carrying the burden. Here is what it consistently misses.
Most postmortems document everything and change nothing. The failure is in the document design, not the analysis. Here is the structure that works for the three distinct readers every incident review actually has.
Most engineering postmortems are structurally correct and practically useless outside the team that wrote them. The problem is not the template — it is that postmortems are written for the author, not the reader.
Practical guides, product updates, and compliance notes — straight to your inbox. No fluff.
Newsletter is opening soon. We'll switch this on once we've got our first issue ready.