Handling agents that crash mid-task: the hard part is the unknown outcome across the network
Responsible-Dare5139 · reddit · 2026-10-16
A discussion of the hardest agent crash case: an external API created a record, but the worker died before saving the response, so durable state says the call never completed. Blind restart risks inconsistency; resuming at that tool call risks duplicates. The author's approach: a stable operation ID that survives retries, idempotency keys where supported, result lookup for uncertain outcomes, and handing the run to human review when the third-party API offers neither dedup nor reconciliation. Durable runtimes (e.g. Diagrid Catalyst's Dapr-based approach) handle progress persistence, but the interesting boundary is the contract between the tool and the external service, not whether the framework has a resume method.
More from coding & agent
- Open-source MCP Lets Claude Generate Image & Video at Provider Price — Demo Cost $0.10 — Master_Relative5687 · 2026-10-17
- Spreadsheet Agent Allowlists Are Not Acceptance Tests — Define What Must Stay Unchanged — Competitive-Mud7832 · 2026-10-17
- Thorsten Ball questions pull requests in the AI coding era: "Why are you still using them?" — kieranklaassen · 2026-10-17
- Zed CEO: OpenAI's Dots is a low-effort model that abandons tasks midway — zeeg · 2026-10-17
- Agent Arena: OpenAI and Anthropic Lead on User Enjoyment as Labs Converge on Reliability — arena · 2026-10-17
- Google AI Studio Build gains Max and Plan modes as Gemini 4 Argon surfaces in UI — testingcatalog · 2026-10-17