Notes On AI Codegen

Much of the implementation of Dartantic was written by AI. That has its upsides and downsides.

The major upside is that I can produce an Agentic framework for Dart quickly and it was needed! (or at least missing and I wanted it : ).

The downside is that not all of the code is what I would have written. I can't tell you how hard that is for me, but it is the dawn of a new age.

On the gripping hand, the AI surprises me with things I'd have never thought of, so it feels like a wash from a quality standpoint.

The parts I do write by hand are the APIs. I want to ensure that Dartantic has a great developer experience, mostly because I use it in my daily work building actual agentic apps. That means a great API that gives you the control you need without being too complex, great coverage of samples that shows off that power, and great docs that are easy to understand and use.

I also pay close attention to the tests to triple-check that I'm testig the behavior I want to see. That's an excellent way to ensure that the AI is heading down the right path.

I figure that if I can get the API, the docs and the tests right, then letting the AI write the bulk of the code is worth the trade-off in turn-around time on features and bug fixes. Plus, I'm building an Agentic framework, so I wanted to eat my own dogfood, as they say. : )

If that doesn't work for you, I understand. Change is challenging, especially change of this magnitude. No hard feelings if a library written with the assistance of AI is not what you want.