Anthropic's Development Practices: A Customer's Technical Analysis

Technical Issues (Reproducible): Artifacts fail to persist to menu (30+ days) Project file access regression (broken Nov 25—previously working) Context isolation between chat and generation modes Development Practice Failures: No apparent dev/test/staging pipeline: Changes deployed directly to production Users discovering bugs in production File access broke for ALL users simultaneously (suggests no canary deployment) No rollback capability: Issues persist for weeks during "investigation" No feature flags evident Breaking changes can't be quickly reverted Support Response Pattern: Support requests not assigned a unique ticket number 3-7 day response times Generic troubleshooting unrelated to reported issues Three explicit supervisor escalation requests ignored Representative suggested "maybe you have multiple accounts" (I don't) The Contradiction: $5B infrastructure investment announced Basic functionality broken 30+ days No proper testing before deployment Support can't escalate appropriately Question for HN: Is this typical for AI companies at this scale? Or is Anthropic particularly bad at the non-AI aspects of running a SaaS business? ... Note: it only addresses one of the three issues I’ve been asking about, assumes (incorrectly) that the problems I’m experiencing are intermittent, and provides no tangible, practical information. It also has no ticket number since they don’t seem to generate them.