{
  "id": 3929153,
  "title": "I Built an AI Fleet That Won't Cry Wolf — Here's How Keystone Watches the Grid for Wildfires",
  "url": "https://urgent.news/2026/08/28/i-built-an-ai-fleet-that-wont-cry-wolf-heres-how-keystone-watches-the",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-28T09:00:00.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/innovate/we-built-an-ai-fleet-that-wont-cry-wolf-heres-how-keystone-watches-the-grid-for-wildfires-354a"
  },
  "original_language": "en",
  "account": "In this report, I document the development of Keystone, a submission to the All Things Agentic Hackathon (Google Gemini + Google Cloud) for the Fortified Enterprise Fleet track. The system monitors a critical transmission corridor on PG&E's Caribou-Palermo 115kV line, where a wildfire disaster claimed 85 lives in November 2018. The source of the fire was a single transmission tower with no prior indication of a potential incident.\n\nThe key challenge in building Keystone is to ensure that the AI system does not generate false alarms, which can lead to a loss of trust among utility operators. To achieve this, Keystone employs a multi-step verification process that includes independent rederivation with engineering physics models, sag-vs-temperature models, voltage-class clearance thresholds, and an adversarial critique mechanism. Only after these checks does the system confirm a finding.\n\nOnce a risk is confirmed, Keystone not only logs the situation but also hands it off to three peer organizations through the Agent2Agent (A2A) protocol. These organizations include a mutual-aid utility that can commit line crews, a fire agency that can pre-position a fire engine, and a grid balancing authority that can approve de-energization. Each organization is an independent AI agent with its own Agent Card, and their responses are displayed in real-time on the system's live map.\n\nAn additional feature of Keystone is its integration with utilities' wildfire mitigation plans. The system allows users to adjust a budget slider, which in turn runs a dynamic programming algorithm to optimize the allocation of funding towards specific line assets. The algorithm takes into account factors such as asset length, risk score, and the cost and risk-reduction potential of various interventions, such as undergrounding, covered conductors, or vegetation management.\n\nThroughout the development process, Keystone encountered several challenges. These included issues with IAM roles, Vertex AI session calls, and memory allocation limitations on field engineers' laptops. Each problem was identified and resolved in production, resulting in permanent fixes for the system.\n\nKeystone is now live, providing an autonomous, risk-optimized fleet for wildfire monitoring and mitigation.",
  "summary": "I wrote this post to document how I built Keystone, my submission to the All Things Agentic Hackathon (Google Gemini + Google Cloud), for the Fortified Enterprise Fleet track. A worn hook on a single transmission tower on PG&E's Caribou-Palermo 115kV line. Nobody was looking at it on the morning of November 8, 2018. Eighty-five people didn't survive the fire it started. That's not a hypothetical…",
  "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."
}