The test suite was the incident
A poor test suite change itself caused a production outage. The author argues test suites should be treated as critical infrastructure, with changes reviewed under the same rigor as production code to prevent such incidents.
Background
Christopher Meiklejohn is a computer scientist known for work on distributed systems and eventually-consistent databases. This piece, dated June 2026, critiques how AI development teams treat passing a test suite as a delivery milestone rather than a basic reliability check. The core argument: a test suite that passes is not evidence a system works — it's merely evidence that known failure modes haven't been triggered. The title inverts the usual "the test suite caught the incident" narrative. Meiklejohn draws on analogies from formal verification and real-world engineering (e.g., aerospace, where tests are designed to fail until the system is proven safe) to argue that the AI industry's reliance on narrow, developer-written unit tests creates a false sense of safety, especially in agentic systems that act autonomously. The piece is notable for its timing — appearing amid growing unease about AI agent reliability in production systems.