One of the World's Best Mathematicians Just Revived His 1999 Software — With AI, in Hours

Terence Tao, Fields Medalist and one of the most respected mathematicians alive, spent the past few days doing something quietly remarkable: he asked an AI coding agent to bring his old software back from the dead. In 1999, Tao wrote a couple dozen interactive math visualizations — small teaching tools called applets — in an early version of Java. Browsers stopped supporting that technology years ago, and the applets went dark. The agent ported all of them to modern JavaScript in a matter of hours, and they now run again, some with graphical upgrades.

His quality report is the most valuable part, because Tao is exactly the kind of person who checks. Across roughly two dozen ported applets, he found just one minor bug — a drag interaction misbehaving at the edge of a box. Meanwhile, the agent found two bugs in his original 1999 code that he’d never noticed. His verdict: “a net wash as far as code quality was concerned.”

Then it got better. Encouraged, Tao dug out an idea he’d abandoned in 1999: a visualization tool for special relativity — he describes it as “Inkscape, but in Minkowski space,” a drawing program where the geometry follows Einstein instead of Euclid. Back then, the code complexity defeated him. This time, “a couple hours of vibe coding” — conversationally directing an AI agent rather than writing code by hand — produced a working app matching his 27-year-old vision. A day later he added a third tool, visualizing the Gilbreath conjecture to accompany a new paper, in a few more hours.

What’s behind this? Note what Tao is not claiming. He isn’t saying AI does mathematics for him, and he’s careful about risk: these are visual teaching aids, not components of a proof, so an undetected bug is annoying rather than catastrophic. That’s a thoughtful frame anyone can borrow — the question isn’t “can I trust AI code?” but “what happens if this particular code is wrong?” For low-stakes tools, the calculus has clearly tipped.

What this means for you: That old project you shelved because the technology moved on, or the tool you never built because the learning curve was too steep — the barrier just dropped, dramatically. If a task is self-contained and easy to verify by using the result, coding agents are now genuinely practical for non-programmers. Tao’s whole workflow was describing what he wanted in plain language. That part, anyone can do.

Sources

• Terence Tao: Old and new apps, via modern coding agents
• Tao’s restored applet collection