Terence Tao Uses AI Agents to Revive Dead Java Applets and Build New Ones

Mathematician Terence Tao describes migrating his decades-old web materials to a maintainable repository with the help of modern AI coding agents. The centerpiece experiment was porting roughly two dozen Java applets he wrote starting in 1999 — visual aids for complex analysis, linear algebra, and objects like honeycombs and Besicovitch sets — which had stopped working once browsers dropped support for that era of Java. The agent rewrote them in JavaScript in a matter of hours, restored full functionality, and even added upgrades such as colorizing a previously monochrome applet.

Code quality was a near wash. Across the port, Tao found only one minor bug (a stray drag-event behavior), while the agent surfaced two pre-existing bugs in his original 1999 code that he hadn’t known about. Encouraged, he went beyond porting: an AI agent helped him finally build a special-relativity visualizer he had abandoned in 1999 as too complex — essentially “Inkscape, but in Minkowski space” — plus a new interactive tool for the Gilbreath conjecture to accompany a recent paper.

Tao’s framing is pragmatic about risk. LLM-generated code is prone to subtle bugs, so he confines these agents to non-critical work: secondary visual aids and paper supplements rather than components of the mathematical arguments themselves. Under that constraint, he judges the downside acceptable and plans to keep shipping AI-assisted interactive visualizations alongside future papers, releasing the new apps as “alpha” builds and soliciting bug reports.