Terry Tao on Modernizing Educational Apps with Coding Agents

Coding Agents Enable Domain Experts to Build Software

Modern coding agents are significantly lowering the activation energy required for domain experts—such as mathematicians and scientists—to create functional software and interactive visualizations. By using LLM-based agents, individuals who possess deep domain knowledge but lack professional software engineering skills can now build tools that were previously too complex or time-consuming to develop manually.

Reviving Legacy Educational Software

Coding agents are proving highly effective at porting legacy code to modern web standards. A primary example is the modernization of old Java applets, which have been obsolete for years due to browser security and compatibility issues.

Terry Tao has utilized these agents to recreate interactive mathematical objects, such as honeycombs and Besicovitch sets, which he originally coded in Java 1.0 as far back as 1999. By leveraging AI, these legacy educational tools can be brought back to life in a modern, accessible format. This trend is echoed by other developers who have used Claude to port 30-year-old Java applets to JavaScript, effectively reviving software that would have otherwise remained dormant.

The Role of AI in Mathematical Visualization

For mathematicians, the ability to generate interactive dashboards and visualizations is a powerful supplement to theoretical work. While these tools are not always mission-critical to the core of a mathematical paper, they provide immense educational value.

Terry Tao notes that the risk of using LLM-generated code for these supplements is acceptable because they are not central to the formal proof or the core logic of the research. This allows for a rapid iteration cycle where the domain expert provides the guidance and the AI handles the boilerplate and implementation details.

Insights from the Technical Community

Discussion among developers and educators suggests that the impact of coding agents extends beyond simple hobby projects. Key insights include:

• Latent Demand for Software: There is a significant amount of "latent demand" for software in non-software-focused fields. Many experts have ideas for tools that they never built because the coding overhead was too high.
• Lowering the Barrier to Entry: The ability for a Fields Medalist to use AI to overcome coding complexity demonstrates that LLMs are expanding software development to a much larger group of highly intelligent people who were previously excluded by the "coding barrier."
• Modernization vs. Emulation: While tools like CheerpJ (which runs Java bytecode via WebAssembly) provide a way to run legacy applets, the use of AI agents to perform "proper modernization" (rewriting the code in a modern language) is seen as a more sustainable path for accessibility.

"There is infinite latent demand for software, most especially outside the traditionally software-focused spaces. If LLMs stopped improving today it would take us 10 years to catch up to the new software-writing abilities that have become available."

Limitations and Risks

Despite the productivity gains, coding agents are not a replacement for professional engineering. The process still requires significant guidance, and the resulting code may contain UI bugs or complexities that can still overwhelm the user if the project grows too large. The consensus remains that AI is a powerful tool for specific use cases—such as dashboards and educational visualizations—but should not be trusted blindly for mission-critical systems.