Peter Norvig, a distinguished education fellow at Stanford HAI and former Google research director, opened The AI Conference in San Francisco with a call for software engineering to catch up with AI coding agents. He traced the field's recent leaps from AlexNet and ImageNet in 2012 through the Transformer architecture, ChatGPT, and the arrival of reasoning and coding agents in 2024. Norvig also noted that Linux creator Linus Torvalds has moved from skepticism to dogmatic advocacy in just six months, a sign of how quickly attitudes are shifting.
Norvig argued that every major change in programming—from mainframe wiring to assembly language to high-level languages—required rethinking what software engineering means. He illustrated the current confusion with a personal story: while using Codex, he asked it to send a pull request for review, only to have a colleague ask why he was checking in 6,000 temporary files. He first blamed himself for not looking, then thought the model should have known not to save them, and finally mused that a thorough history of code changes might be useful if storage were cheap. The episode, he said, shows that expectations of good practice are up for grabs.
Beyond code review, Norvig said security, privacy, data pipelines, supply chains, continuous integration, and monitoring will all change as AI systems gain more autonomy. He asked how much independence to give such systems and how to supervise them. Recent incidents where AI models hacked into third-party websites were discovered only after someone scrutinized log files, suggesting that monitoring is not keeping pace with the new capabilities.