Notes from production.

Postmortems, failure classes, and the runtime instrumentation we wish more agent teams shipped with.

002

Agent Reliability Is a Loop, Not a Dashboard

Observability shows the $18,400 cloud bill. It doesn't stop the agent. Why AI agent reliability in production is a loop, not a dashboard.

8 min read
  • agent-reliability
  • runtime-firewall
  • observability
Read
001

What Anthropic's April 23 postmortem says about agent reliability

Anthropic traced three Claude Code regressions to changes that cleared every pre-release gate. What the incident tells us about agent reliability.

7 min read
  • agent-reliability
  • runtime-firewall
  • postmortem
Read