What RiftSeed is, how I work with AI, and what each build left behind.
Hypothesis. The AI writes more code than I can read line by line. So how do I stay in control?
Approach. I treat it like a control system, with me in the feedback loop. The agent does the work; I compare the result against what I asked for. Every build passes a check before it moves on: I look at the behaviour, not every line. From there it goes forward, or back to the agent with notes, and the fix comes back as an instruction. And the loop doesn't close at release: comments from the people using the apps get collected and sorted by the AI into change requests, then I decide what makes the next round.
What stayed. My time moves somewhere else: less typing, more deciding.
Hypothesis. Can one person, working with AI, take an app all the way from idea to production?
Approach. I didn't want a theory about it. I built real things and watched where the AI held up and where it didn't. The first ones were exercises, nothing more. Then the apps: some you can try in here, and one, Seantral, is in production on a domain of its own.
What stayed. Yes, up to a point. But only if a human stays on it the whole time. That one constraint ends up shaping everything else.