{
  "id": 13223149,
  "title": "The Agent Failure Your Approval Gate Can’t Catch",
  "url": "https://urgent.news/2026/10/09/the-agent-failure-your-approval-gate-cant-catch",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-09T21:40:37.000Z",
  "source": {
    "name": "DevOps.com",
    "slug": "devops-com",
    "url": "https://devops.com/the-agent-failure-your-approval-gate-cant-catch/"
  },
  "original_language": "en",
  "account": "An agent that does nothing and reports success leaves a green checkmark, but this green signal is the one that your incident process trusts. DevOps teams invest heavily in catching failure through exit codes, run status, alert rules, and dashboards, but these are built for software that fails loudly. Andrew Filev's reliability argument in July emphasizes that reliability comes from a system's ability to handle failure, not from a single component. The idea is that approval gates ensure important outputs receive the necessary scrutiny before proceeding. However, in the case of an agent that quietly does nothing, no output reaches the approval gate, so it is approved and no feedback loop is triggered. This failure goes unnoticed, unlike an agent that does the wrong thing, which can trigger alerts and reviews. The issue lies in the fact that absence is not captured by the current workflow. The solution is to verify the actual state of the world rather than relying solely on the agent's self-report. This can be done by declaring the expected effect as a precondition of success, treating implausible speed as a failure signal, and triggering verification based on the expectation rather than the run. By implementing these changes, the system can better handle the failure that never announces itself, ensuring reliability even in the absence of obvious signs of trouble.",
  "summary": "An agent that does the wrong thing leaves evidence — a bad diff, a malformed record, a customer complaint. An agent that does nothing and reports success leaves a green checkmark, and green is the one signal your incident process is built to trust. Worry more about the second failure, because everything a DevOps team […]",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}