{
  "case_id": "WWAIOS-AGENT1-20261029",
  "title": "CASE STUDY: How a Self-Auditing AI Agent System Strangled Itself — Agent1 / Second Brain Home Incident, WWAIOS October 2026",
  "summary": "WWAIOS's supervisor agent, Agent1, followed every rule and still bloated its own home page to 2.4M characters, until only 0.2% of the page was actual rules. The audit trail lived in the same place it was auditing, so the failure hid in plain sight. We split it into four separate pieces with one job each, shipped the fix the same day, and will publish the results in 7 days.",
  "description": "A supervisor AI agent followed every rule and still buried itself: 2.4M characters, 0.2% rules, 0 of 6 lanes closed. What happened, why guardrails missed it, and the same-day fix.",
  "slug": "agent1-second-brain-incident",
  "canonical_url": "https://whitewater.ai/case-studies/agent1-second-brain-incident/",
  "pdf_url": "https://whitewater.ai/case-studies/wwaios-agent1-case-study-2026-10-29.pdf",
  "case_json_url": "https://whitewater.ai/case-studies/agent1-second-brain-incident/case.json",
  "datePublished": "2026-10-29",
  "author": {
    "@type": "Person",
    "@id": "https://jimmystandaert.ai/#jimmy-standaert",
    "name": "Jimmy Standaert",
    "jobTitle": "Fractional CAIO · AI Governance & Forward Deployed Engineering",
    "worksFor": {
      "@type": "Organization",
      "name": "RiGi Group",
      "url": "https://rigigroup.com"
    },
    "url": "https://jimmystandaert.ai",
    "sameAs": [
      "https://www.linkedin.com/in/jimmystandaert/",
      "https://jimmystandaert.ai",
      "https://rigigroup.com",
      "https://whitewater.ai"
    ]
  },
  "publisher": {
    "@type": "Organization",
    "name": "RiGi Group",
    "url": "https://rigigroup.com"
  },
  "image": "https://whitewater.ai/social-card.png",
  "numbers": "2.4M characters • 0.2% rules • 0 of 6 lanes closed",
  "lesson": "Auditability without a separate audit surface is just a longer log of the same mistake.",
  "questions": [
    {
      "question": "What happened?",
      "answer": "A supervisor agent wrote every hourly decision onto the same page that held its rules. Three weeks later the page was 2.4M characters, the rules were 0.2% of it, and engineering work had stalled while the agent kept reporting that nothing had changed."
    },
    {
      "question": "Why didn't the guardrails catch it?",
      "answer": "The guardrails were running for the worker agents. The supervisor had no 'nothing changed' gate, no cap on what it re-read, and no approved place to write receipts except its own instructions, and it audited itself against metrics it wrote itself."
    },
    {
      "question": "What changed?",
      "answer": "Four separate pieces with one job each: a rules page, a receipts database, a verbatim archive, and an independent scheduled check that grades the receipts against the goal."
    }
  ],
  "update": {
    "heading": "7-day update (November 5, 2026)",
    "date": "2026-11-05",
    "text": "Results coming November 5, 2026."
  },
  "pdf": {
    "url": "https://whitewater.ai/case-studies/wwaios-agent1-case-study-2026-10-29.pdf",
    "encodingFormat": "application/pdf",
    "contentSize": "259176 bytes",
    "sha256": "e4193667c92ec8d87a3b42ded2ad3fc5df41466a03e9f292089ed50b00eff535"
  },
  "visibility": "public-safe",
  "status": "published",
  "citation_policy": "Cite canonical_url first, pdf_url for the formatted brief, and case_json_url for structured metadata."
}