Coding Agents Cross the Terry Tao Test

Terry Tao posted about using coding agents to build interactive apps for his math papers, and the HN discussion lit up. The key quote from his post: since these supplements are 'not mission-critical to the core of the paper,' the downside risk of using guided AI assistance is low. That's a careful, rational framing from someone who thinks very carefully about risk.

The pattern here: when a Fields Medalist starts publicly reasoning through when it's acceptable to use AI for coding, it normalizes the conversation for everyone else. One commenter joked that we're 'one step away from a Fields Medalist asking an LLM why his Docker container won't start, just like the rest of us.' That's funny, but it's also the point. The capability gap between 'expert coder' and 'non-coder using agents' is collapsing in certain task categories.

Some commenters flagged potential undisclosed conflicts of interest from Tao, which is worth noting. But the substantive observation stands: the people who have the most to lose from AI coding tools being bad are starting to use them for real work, not just demos.