The Normalization of Inexplicable Failures
Points and comments are a snapshot, not live.
AI tools like Jev normalize inexplicable failures where users accept "stupid thing sucks."
A blog post uses a TV show scene to illustrate how characters accept inexplicable failures. It criticizes AI model Jev, which returns typed values with probability estimates, arguing that confidence scores are used in a cargo cult manner without proper calibration or cost modeling. The author fears that LLM-driven development accelerates a trend where both users and builders shrug off failures instead of investigating root causes, normalizing inexplicability in software.
What commenters are saying
Commenters broadly agree that inexplicable failures are not new, citing examples like GitHub 5xx errors, AWS outages, and the age-old advice "turn it off and on again." Some argue this is a choice, not inevitability, while others note that nondeterministic AI systems make contracts and guarantees impossible. One commenter describes experiences with electric car software faults where dealerships shrug, and another notes that warranties often exclude software failures.