Build old and new apps with coding agents

Tools

custom-built or unnamed agent

Overview

I have been interested in machine-assisted ways to do and teach mathematics from as far back as 1999, when I started coding several applets in Java 1.0, both for my complex analysis and linear alge…

Workflow

The author used an AI coding agent to migrate legacy Java 1.0 mathematical applets to JavaScript and to build new visualization tools for special relativity and the Gilbreath conjecture. The workflow involved providing the agent with old code or conceptual descriptions, iterating through 'vibe coding' sessions, and manually reviewing the output for logical correctness and UI behavior. Human expertise in mathematics and programming was essential to define high-level design decisions and verify that the generated code adhered to specific scientific principles.

Reported result

The agent successfully ported approximately two dozen applets in a matter of hours, restoring functionality and adding graphical upgrades like colorization. It identified two pre-existing bugs in the original code while introducing only one minor drag-event bug in the ported versions. New applications for special relativity and Gilbreath conjecture visualizations were generated, though they remain in alpha or pre-alpha stages with potential rough edges requiring further user feedback.

Human role

The original post does not say how a person stayed involved.

Limitations

The approach relies on the agent's ability to understand standard mathematical concepts and existing code structures; complex, non-standard logic may require more human guidance. LLM-generated code can contain subtle bugs, making this method suitable for secondary visual aids rather than critical mathematical proofs. The author notes that domain expertise and programming experience remain essential for detecting issues and guiding the agent's design choices.

Try it

Act as a coding assistant to help me migrate legacy code or build new visualization tools. I will provide [source files or detailed specifications] for a mathematical or scientific application. First, analyze the input and propose a migration plan or architecture for a new app in [target language, e.g., JavaScript]. Then, generate the code in stages, starting with the core logic engine before building the UI. After each stage, present the code and a brief explanation of how it meets the requirements. I will review the output for logical errors and UI behavior. Do not deploy or publish any code until I explicitly confirm that the sample implementation is correct and meets my standards. If you encounter ambiguous requirements, ask clarifying questions before proceeding.

More cases · Software & development

• Update a code base with Copilot chat and agent mode
• Check code against user intent with review agents and tests
• Build a budgeting app from a single prompt
• Build a mental models framework with a multi-agent dev team
• Build a instant-opening markdown viewer for macOS