70%-95% failure rates: devs say AI agents still too unreliable for real workflows
Luke17man · reddit · 2026-10-22
Reddit user Luke17man vents that AI agents feel like a dice roll on multi-step tasks, often silently breaking at step three with drift and silent failures. Cited stats: production failure rates for autonomous multi-step tasks range from 70% to 95%, and roughly 62% of knowledge workers consider agents untrustworthy due to compounding errors in long-horizon workflows. The thread resonates with builders questioning whether anyone actually runs autonomous agents in production.
More from coding & agent
- Microsoft Research's PDD Kit formally proves AI-generated code meets your intent — omojumiller · 2026-10-22
- SF Systems Meetup: Agents-Only Version Control and Postgres Rewritten in Rust — ShadajL · 2026-10-22
- BitKindle: AI bootstraps a full C compiler from a 3,584-byte machine-code seed — dotey · 2026-10-22
- dsh-prompt-layer plugin injects your rules into DSH's system prompt, overriding AGENTS.md — teortaxesTex · 2026-10-22
- LLM writing still garbage in 2026, dev shows the blunt prompt he uses — arthurcolle · 2026-10-22
- "Patch my PHP vulnerabilities" — Claude also refactors 37 files and breaks the app — petrusenko_max · 2026-10-22