{
  "id": 655,
  "url": "https://arxiv.org/abs/2606.24453v1",
  "title": "Bayesian control for coding agents",
  "summary": "Modern coding agents pair LLM generators with various tools, including cheap diagnostics and expensive verifiers. The tool-use decisions are typically governed by orchestrators that often use fixed rules and ignore uncertainty. We formulate orchestration as cost-sensitive sequential hypothesis testing: a Bayesian controller maintains a belief over candidate correctness and dynamically decides whether to gather more evidence, refine the candidate, verify it, or stop. Across six generators and nin",
  "authors": "Theodore Papamarkou, Vladislav Smirnov, Viktor Mazanov, Artem Vazhentsev, Preslav Nakov, Timothy Baldwin et al.",
  "category": "research",
  "topics": "healthcare,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-23T11:41:32.000Z",
  "fetched_at": "2026-07-14T14:14:41.551Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/655",
  "original_url": "https://arxiv.org/abs/2606.24453v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}