How do you actually test agents before production? Devs tackle failure-path testing
Last-Bluejay-4443 · reddit · 2026-10-22
Reddit user Last-Bluejay-4443 asks how builders test agents beyond the happy path: tool calls timing out after the action actually succeeded, agents retrying things they shouldn't, and workflows stalling mid-way with no recovery. Standard evals and prompt tests don't cover real-world failure modes, so the author asks whether people build custom failure simulations, use off-the-shelf tools, or just discover issues in production — a key concern when shipping agentic workflows for clients.
More from coding & agent
- Microsoft Research's PDD Kit formally proves AI-generated code meets your intent — omojumiller · 2026-10-22
- d3-parsets v3.0.0 rewritten for D3 v7, visualizing how people use Claude — Elijah_Meeks · 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