Agent retries after a timeout can double your orders; stable operation IDs and state read-backs are the fix
arthaudm · reddit · 2026-09-17
The author highlights an automation failure nobody budgets for: the external system completes the action, the response times out, and the agent retries — leaving you with two tickets, two emails, or two purchases.
The guardrail that matters is not a better retry prompt: assign every intended effect a stable operation ID, read back the external state after an ambiguous response, and only retry when you can prove the first attempt did not land. The authors find this failure mode more useful than most model benchmarks because it tests whether the surrounding system can distinguish failed from unknown.
Related event: Stable operation IDs proposed as fix for agent retry duplication(2 posts)→
More from coding & agent
- Tencent unveils Hyra 2.0, a research agent that builds reusable scientific tools — TencentHunyuan · 2026-11-02
- What broke when I ran a leaderless agent swarm with no orchestrator — 7 failure modes and fixes — LostHumanity67 · 2026-11-02
- Forced yes/no prompts push GPT and Claude to 95%+ confidence on Goldbach's unproved halt test — Nir777 · 2026-11-02
- Automation took over my daily work — so why am I still sitting in the office for 8 hours? — Am1rSori · 2026-11-02
- General Reasoning explains how to train agent swarms with RL and latency pressure — rosstaylor90 · 2026-11-02
- "Coding isn't the bottleneck anymore — usage windows are," says founder Kyle Gawley — kylegawley · 2026-11-02