Is it possible to use AI to generate high-quality Codename One code for our apps?

[jsfan3]

I'm curious to hear about other developers' experiences using AI tools to generate high-quality Codename One code.

Since Codename One is Java-based but uses its own APIs, I’ve noticed that generative AI can produce confusing or incorrect results, especially when the AI mixes standard Java approaches with CN1-specific patterns.

That said, in my experience, even a "general-purpose" AI like ChatGPT has been extremely helpful. With the right prompts and guidance, it can produce useful code, suggest ideas, and significantly speed up development.

For instance, I built this CN1Lib with substantial help from AI: https://github.com/jsfan3/nihonjindes-cn1lib-tts/tree/main.

What have you tried with AI tools and Codename One? How have you dealt with incorrect or "hallucinated" API usage? What techniques and strategies work best?

[shai-almog]

I use LLM agents extensively. With newer agents like codex, claude code, opencode etc. you can just point at the local Codename One maven project and it would build/test the project for you. It will add unit tests and verify that everything works correctly.

One of the motivations for Adding Java 17 support was to make this process easier for the cloud based hosted agents that typically only have Java 21 installed.

When I started this process I built an initial version of an MCP: https://github.com/shai-almog/cn1-mcp
I'm now not so sure if it's worth investing in it since modern agents have gotten so good it's no longer necessary.