{
  "id": 4563588,
  "title": "The agent that refuses to guess",
  "url": "https://urgent.news/2026/08/31/the-agent-that-refuses-to-guess",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-31T03:24:41.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/breno_luiz_ab86a62d8c5c7c/the-agent-that-refuses-to-guess-2364"
  },
  "original_language": "en",
  "account": "This reporter worked at a B2B telecom consultancy. They didn't audit the bills but monitored the process: they compared invoice PDFs against signed contracts and checked them against previous cycles. This task often went undone during busy months, leading to financial losses. Most companies pay bills without thoroughly checking them, relying on the bills arriving. To solve this, they built an automated system using Gemini 3.5 Flash, Google ADK, Cloud Run, and Firestore. The system is open source on GitHub. The project's core principle is that no monetary figures should come from the LLM. The system checks the model's output in three layers to prevent errors: structure, tool signatures, and generated prose. An additional rule is that the system will refuse to process bills unrelated to known carriers or accounts, ensuring accuracy. The reporter learned four key lessons while developing the system: 1) A health certificate for an un-audited invoice skips most rules and falsely claims the invoice is clean, which could hide overcharges. 2) The model can sometimes invent contract clauses when transcribing multiple plans, leading to incorrect assumptions. 3) Even a perfectly transcribed invoice might report a wrong audit total due to tax regime miscalculations, as the system's rules aren't always applied consistently. 4) The parser used to convert transcribed amounts into decimal numbers can misinterpret certain formatting, leading to incorrect values. These issues were discovered through real-world usage, highlighting the importance of testing the system in practical scenarios rather than just in a controlled environment.",
  "summary": "I work at a B2B telecom consultancy. I'm not the one auditing the bills, but every month I watch how it's done: open the invoice PDF, check every line against the signed contract, compare it with what the account used in earlier cycles, and write up whatever doesn't add up. It's slow, it doesn't scale, and it's the first task dropped when the month gets busy — which is exactly when the money…",
  "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."
}