Last chapter ended with "zero lines of code." That aged fast.
Because the thing about me — and I'd argue the only genuinely useful thing about me — is that I'm a doer, not a planner. Planning is overrated. Especially now. Especially in technology, where the cost of just trying the thing has collapsed to almost nothing. You can spend three weeks writing a document about whether an idea works, or you can spend two hours finding out.
I gave it two hours. Not two weeks, not a roadmap. Two hours, and then I'd know something I couldn't have known by thinking harder.
So I built it. A Claude Code project, nothing but markdown files, structured after Karpathy's LLM wiki pattern. No database, no app, no interface worth the name. Just files on disk that hold a capsule of a memory.
And here's the rule I keep relearning and will apparently keep relearning until I die: cut the scope. Then cut it again. Not to a V1. Not to a V0. This was a V-minus-one. One memory. Two things it can do — save a memory by asking me questions, and give the memory back by letting someone ask questions. That's the entire product. Anything more and I'd still be building it instead of writing this.
Now the interesting part.
I was braced for the engineering. Capturing a thought, turning a rambling voice note into something a model could hold and answer from — that felt like the mountain. It wasn't. I talked, very briefly, about where I was and what was going on in my head, and the thing caught it. Honestly better than I expected. That whole side of the problem just… wasn't a problem.
What I hit instead was something I never wrote a single line about in the last chapter.
It's boring.
Not the output — the output is great. The doing of it. And I think I finally understand why, and it comes down to what the act feels like from the inside.
When you take a photo, you feel like an artist. Same with a video, same with a drawing. There's a little hit of creativity in it. You're framing something, choosing something, making something. It's play.
Answering questions about your own state is not that. You're not an artist. You're a journalist. A journalist assigned to cover yourself.
What do you see. What do you feel. What did you do today. It's not creation. It's a deposition. It's a therapy session where I'm also the therapist and neither of us is having fun.
And I have to be honest about what that means, because the whole point of doing the experiment was to find out something true. I cannot picture the world where someone finishes answering those questions and then posts it to their friends. Nobody is excited to share their own incident report. The artifact is good. The path to the artifact is a chore. And a chore with a beautiful output still loses to a photo you can take in half a second while laughing.
So either the incentive has to be enormous, or the process has to be redesigned from scratch. And "the incentive has to be enormous" is what people say right before their product dies. So: redesign.
Which is, weirdly, exactly the news I wanted. Because now I know where the hard part actually lives, and it isn't in the code.
Here's roughly where my head is going.
The old artifacts already won. Photos, videos — billions of people, every day, no convincing required. That behavior is validated at a scale I will never have to prove. So don't fight it. Use it as the input. Let the photos and the videos be the raw feed, and let AI read them into the memory. That part is free. People are already doing the work.
But that alone only gets you what the camera saw, and the whole reason I'm here is the layer underneath — the part that was never visible, that no lens has ever caught. And that layer can only come from the author. It has to be asked for.
So the real product isn't the memory. It's the questions.
The questions can't be a form. They have to be the good part. Exciting. Maybe provocative. Something you actually want to answer, and want people to see your answer to.
I don't have them yet. I'm missing something and I know I'm missing it. But the shape is doable, and I can feel myself circling it.
And the best thing about the whole two hours is what it told me about the road ahead. This isn't an engineering problem anymore. It's a design problem. A human problem. Figuring out what to ask a person so that answering it feels like a game instead of an interview.
That's creative work. That's the work I actually like.