Building increasingly complex software with AI

One of the main things I want to work on right now is just using AI coding to build increasingly complicated software.

Not necessarily software with AI features. Just sophisticated software that does useful things and would have previously been too complicated for me to build and manage myself.

The dashboard I started building to help select contacts and generate email copy for outbound campaigns has become the biggest piece of software I've ever built "myself." It's around 75,000 lines including tests, with about 1,400 unit tests.

In the past, I would have eventually gotten blocked by the technical debt or my own weaknesses as a developer. Now I can just keep cruising past those points and iterating.

I built a background job system for copy generation, contact enrichment, crawling, and multiple rounds of content review. Whether I start something through Codex or the web UI, I can see the job running, inspect the traces, and know what's going on. There are even nice little persistent toasts while the jobs are running.

That's the kind of thing I wouldn't have been able to manage effectively before. Not because any one part is impossible, but because keeping the whole system under control as it grows would have eventually gotten away from me.

I'm still making massive adjustments to it day to day. I added crawling functionality the other day because I wanted it, and the codebase can actually absorb that kind of change.

That's what I want to do more of with AI coding. Build increasingly complicated software and keep making it better without eventually getting buried by it.