A product manager decides to build systems software

I can't write code. I'm building native Mac audio software anyway, entirely with AI. Here is why that is less reckless than it sounds — and where it might still break.

I want to be clear about what I am, because the whole point of this blog rests on it. I am a product manager. I have written product specs for years. I have not written production software, and I am not going to pretend otherwise. Jiva — a fully-local Mac app that records meetings, transcribes them on the device, and remembers who said what — has been written line by line with an AI.

The obvious question is whether that is reckless. My answer is that it is less reckless than it looks, for one reason, and more fragile than it looks, for another.

It is less reckless because the hard part of building with an AI is not writing the code. The AI writes the code. The hard part is remembering why — why a past decision was made, what was tried and rejected, what a piece of the system is for. Note that a model has no memory between sessions, and honestly neither do I after a few weeks. So the thing that made this possible was not the model, rather a structure I built around it: a vision doc, a roadmap, a live status file, a numbered decision log, and a wiki of component specs. Any cold session — a different model, or me on a Monday morning — picks up full context from those files without me re-explaining anything.

It is more fragile than it looks because writing code an agent produced and shipping code you can responsibly stand behind are two different things. I have found real bugs in my own software only because I happened to rewrite a piece of it generically. The gap between "an agent can write it" and "a non-engineer can safely ship it" is real, and closing it is half of what I want to write about here.

So the honest framing is not "look how easy this is." It is a question I do not yet know the answer to: how far can one person get this way, on genuinely hard software, before that fragility catches up? The category I am building in is owned by companies worth a billion dollars and more. Whether one person with an AI can stand next to what those teams ship is the thing this whole blog is trying to find out.

More soon.