What breaks when AI agents run in production? Developer explores an SRE-style control layer
Fantastic-Sleep-3352 · reddit · 2026-09-05
A developer is exploring building an SRE-like control layer beneath AI agents that monitors trajectory and state, then decides whether to retry, replan, switch tools, downgrade models, restore state, escalate to a human, or stop — verifying task success before completion. He lists 12 validation questions for teams running agents in production, targeting pain points like runaway loops, token cost spikes, untraceable failures, missing replay/audit, and unclear human-approval thresholds, arguing this unglamorous reliability layer may matter more than another agent framework.
Related event: Devs Debate Whether Production AI Agents Need an 'Agent SRE' Layer(3 posts)→
More from coding & agent
- Grady Booch revives Weinberg's 1971 quip: software so buggy one woodpecker could destroy civilization — Grady_Booch · 2026-10-23
- One Agent Replaces Half an SDR Team: Full Lead-Research-to-Email Workflow — tibo_maker · 2026-10-23
- Why don't coding agent harnesses ship with their own sandbox? Only amp does — Aryvyo · 2026-10-23
- Dev builds a VST synth with Codex and MCP, lets the AI jam in his studio — Yamapama · 2026-10-23
- AgentsInTheCloud update: use your Claude subscription with built-in coding agents — lucasmeijer · 2026-10-23
- Phone agents vs. secure screens: banking apps treat helpful agents like malware — ghostmalone02 · 2026-10-23