Running agents in production: the hard part isn't the LLM, it's durable execution
Responsible-Dare5139 · reddit · 2026-10-21
A Reddit discussion arguing the real difficulty of productionizing agents comes after the model decides correctly — the record update succeeds, the notification times out, the human approval lands after a deployment. A smarter model fixes none of that.
- You need to know which actions actually completed, which results persisted, which requests may still be running, and which approval maps to which version of the action; a “tool failed” log answers none of this.
- Hence the author finds Diagrid Catalyst's focus on durable execution more interesting than yet another prompt-composition abstraction: making run lifecycle a shared infrastructure concern.
- Evaluation should test killing a worker mid-tool-call, duplicate callbacks, cancellation during external requests, and deploys while a run waits — each with an expected business outcome, not just green health checks.
- Bottom line: a good answer and a correctly completed task are separate outcomes; both must be measured before calling an agent production-ready.
More from coding & agent
- AWS Transform's .NET agent now converts Azure SDK code to AWS equivalents — DigitalColmer · 2026-10-21
- Claude Code renders lyric videos frame-by-frame in Chromium, no AI video model needed — aitrendz_xyz · 2026-10-21
- Rule-based Self-Layout system renders a full Mos Eisley 3D map with 5,367 sensibly placed objects — jasonkneen · 2026-10-21
- Redditors ask: what does it really take to keep AI agents reliable in production? — djgoel123 · 2026-10-21
- 'How Do People Code Without Claude?' — 'Easy, We Use Codex' — alexmacgregor__ · 2026-10-21
- Google releases free 1-hour AI engineering course, from first agent to autonomous systems — msharmas · 2026-10-21