{
  "id": 11033750,
  "title": "Peter Norvig says all aboard for AI coding",
  "url": "https://urgent.news/2026/09/30/peter-norvig-says-all-aboard-for-ai-coding-11033750",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-30T19:05:25.000Z",
  "source": {
    "name": "The Register",
    "slug": "the-register",
    "url": "https://www.theregister.com/ai-and-ml/2026/09/30/peter-norvig-says-all-aboard-for-ai-coding/5300244"
  },
  "original_language": "en",
  "account": "Peter Norvig, a distinguished education fellow at Stanford HAI and former Google research director, delivered his opening keynote at The AI Conference in San Francisco on Wednesday, expressing his thoughts on the future of AI coding. He noted the increasing number of AI conferences in San Francisco this year and the upcoming one scheduled for Thursday. Norvig traced the evolution of AI from AlexNet and ImageNet in 2012, the Transformer architecture in 2017, and the debut of ChatGPT in 2022, to the emergence of reasoning and coding agents in 2024. He also highlighted the growing competency of AI models in performing complex tasks like math, where GPT-4 could only count the number of 'r's in \"strawberry\" two years ago, but in August, 25% of math preprint papers on arXiv acknowledged AI assistance.\n\nNorvig discussed how the attitude of Linux kernel creator Linus Torvalds toward AI has shifted from skepticism to cautious adoption, and now to dogmatic advocacy. He emphasized the need for a significant change in the process of software engineering, as the tech industry has undergone numerous transformations before, such as moving from plugging wires into mainframes to higher-level programming languages. At each stage, people had to adapt their approach to software engineering.\n\nTo demonstrate this point, Norvig shared a recent experience where he used Codex, an AI coding assistant, to write some code and promptly send a pull request for review. However, he overlooked the fact that Codex had generated 6,000 temporary files during the process. His colleague questioned why there were so many temporary files, leading Norvig to explain that he had not noticed them. He then instructed Codex to delete the unnecessary files, but initially, he doubted the model's ability to make such decisions. Norvig then reflected on how AI models should ideally be aware of such user preferences to avoid unnecessary burden on reviewers. He also pondered the potential benefits of having a more comprehensive history of codebase changes, even if it meant storing more temporary files.",
  "summary": "Software engineering practices need to catch up with increasingly capable agents",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "The Register Science",
        "title": "Peter Norvig says all aboard for AI coding",
        "url": "https://urgent.news/2026/09/30/peter-norvig-says-all-aboard-for-ai-coding",
        "published": "2026-09-30T19:05:25.000Z"
      }
    ]
  },
  "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."
}