Modern Metaprogramming For Young Ladies

Your AI must comprehend the inner model of your software application, that means you need to be modeling it in 3D space with three.js.

You will need many projections of the models, including TODO projection, confusion projection, User Interface projection, user walk through projection.

Whatever user interfaces you have seen in sci-fi movies, this thing needs to be 10 more impressive.

And when you are ready, you will need WebXR, you will need superpowers.

So, at first, you build the nervous system, the data pipelines, the nested user interface tree, you do not have your agent create code.

The moment you create the code, the AI forgets most of the model, any changes you make can only mess things up.

You can use plugin architecture, you can target a custom web desktop, that has a single icon Supermarket where you publish your programs.

Plugin architecture stabilizes things a lot, but if you create a complex application model that you can manage.

Then the code, becomes a formality, because your model will generate very precise application requirements.

And when time comes to make changes, you regenerate, preferably new version of the plugins from the previous code generation.

Here code generation is small potatoes, the main thing is the model that you control.

And the program requirements that you generate in English, so that a large language model can create the code.

The programming, the source code, becomes a formality.

And you can build a system that targets many languages and platforms, creating an optimal application for each target system, including game consoles.

I call this meta programming, but you should take it to the extremes.

Don’t generate the code for your application, have your coding agent built an application modeling platform.

That generates, requirements documents, specifications, SKILLS, Agents, that are then used for generating your application, as the final step.

Do not generate the code of your application, build an application modeling platform that creates build instructions.

And you’ll never have to worry about code, or code rot, or language, target platform, or hosting systems.

You will always be able to regenerate from your model, with new instruction and bug fixes, and new ideas.

Finally, don’t start projects with blank application models, use WebMCP to have your AI use your modeling platform to create application templates.

Pre-create as many as 64 application archetypes, or more, and have your AI improve them often.

This will help you start from template, and decrease development time, especially if you re-use powerful plugins in your end-user apps.