{
  "id": 19507342,
  "title": "AI agents run most of my company. Here is what broke.",
  "url": "https://urgent.news/2026/11/11/ai-agents-run-most-of-my-company-here-is-what-broke",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-11T14:09:43.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/heytobo/ai-agents-run-most-of-my-company-here-is-what-broke-5eif"
  },
  "original_language": "en",
  "account": "This story details the experience of a 16-year-old who runs their company using AI agents. The young entrepreneur tracked 453 sessions between October 7 and November 11, with 38.3 hours spent inside tools, and 17,050 tool calls. Nine specific issues arose during this period, each demonstrating a flaw in the AI agents' operations.\n\nFirstly, the agents spent a significant portion of their time waiting, sometimes for up to 68 minutes. This led to a rule that any task taking more than a minute should run in the background, with the agent continuing work elsewhere. A monitoring script failed to alert the user when a site went down for 33 days, demonstrating that a sensor must alert only when a problem changes state, not continuously.\n\nA test of retry functionality showed that an agent could resend messages even when the initial attempt was successful, leading to duplicate communications. Another issue involved a deploy command printing a success message prematurely, while a file was rejected, resulting in lost pushes. The agent had also loaded 17 out of 148 skills, but only 17 of them were used in 453 sessions, illustrating the need to regularly review and remove dormant skills.\n\nEight agents were set up to propose content, costing $5.58 in API calls and nearly exhausting the daily quota. The agents failed to work without a new input and multiplied channel counts quadratically. Parallel sessions led to 37,714 restarts due to missing modules, causing the laptop to overheat and consume excessive power. The user learned that automation can quietly rot, with recurring jobs left unchecked if they produce no results. Lastly, the entrepreneur underestimated the costs of tasks and the existence of pre-existing tools, emphasizing the importance of thorough testing and estimation before deploying AI agents.",
  "summary": "I'm 16. I have no employees: I have Claude Code sessions. They write the code, answer suppliers, publish, and watch the sites while I'm in class. I measured 453 of those sessions and wrote down every failure the day it happened. Here are nine of them, with the numbers and the rule each one left behind. Measured on November 11, 2026, on the transcripts on my laptop: 453 sessions between October 7…",
  "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."
}