Terry Tao Revives 1999 Java Applets with AI Coding Agents
12 Jul 2026
A 1999 time capsule, rebuilt with AI
Fields Medalist Terry Tao has used AI coding agents to bring a set of 25-year-old teaching tools back to life. According to a blog post, Tao began coding interactive applets in Java 1.0 back in 1999 for his complex analysis and linear algebra courses—including a honeycomb applet built with collaborator Allen Knutson. Those applets had long since aged out of usability as browsers dropped Java plugin support.
Now, with the help of modern AI coding assistance, Tao is migrating his old web pages and blog data to a more maintainable repository, and porting the original applets from Java to JavaScript so they run in today's browsers again—complete with graphical upgrades.
The numbers
- ~24 applets were ported from Java to JavaScript.
- Only one minor bug was found during the entire porting process (its specifics weren't detailed).
- The original code dates back to 1999.
Beyond restoration, Tao also used AI-assisted coding to build entirely new interactive tools: a visualization for special relativity in Minkowski space, and a visualization tied to the Gilbreath conjecture, created to accompany a paper and blog post. He's indicated plans to add similar interactive visualizations as supplements to future papers.
Why founders should care
This is a small, single-source anecdote—not a benchmark study—but it's a suggestive data point for anyone building or evaluating AI coding tools:
- Legacy migration may get cheaper. Porting roughly two dozen applets with only one bug surfacing hints that AI coding agents could plausibly reduce the effort and risk typically associated with modernizing old codebases, though the report gives no data on time or hours saved, so this should be treated as directional rather than conclusive.
- One tool, two use cases. The same AI-assisted workflow that handled maintenance (porting old code) was also used to build new features (the relativity and Gilbreath visualizations). That dual capability—maintenance plus greenfield development—may be relevant to founders assessing how far current coding agents can stretch across a product lifecycle.
- Content-plus-tooling pairing could spread. Tao's stated intention to pair future papers with interactive visualizations suggests a possible pattern: technical writers and educators increasingly treating AI-built interactive tools as a standard companion to written content. If this pattern holds beyond one example, it could open a niche for devtools aimed at authors and educators.
What's still unclear
Several important details are missing from the source material. It's not specified which AI coding agent(s) were used, how much time or effort the migration and new tool-building actually took, what the one bug found during porting actually was, or whether the modernized repository and new visualization tools are publicly accessible. Founders looking to draw firm conclusions about AI-assisted legacy migration should treat this as an encouraging anecdote rather than validated evidence—the risk of subtle bugs during migration (as the one incident shows) still warrants careful verification when relying on AI agents for code ports.