我独自运营四个AI产品：第五个花了三天，且由Claude设计

903 reads I Run Four AI Products Alone: The Fifth One Took Three Days - and Claude Designed All of It by Jacob Yang September 3rd, 2026 TLDR Your browser does not support the audio element. byJacob Yang@jacobpark byJacob Yang|@jacobparkFounder & Solo Builder at Indie maker running five AI products solo — Tracetify, DatePhotos AI, UGCfy AI, FragCut. Building in public with Claude as my cofounder. Story's Credibility I'm Jacob. I run DatePhotos AI, UGCfy AI, FragCut, and a couple of smaller things — all solo, all bootstrapped. No team, no funding, no designer. For about a year now, my de facto cofounder has been Claude. Last week, I shipped my fifth product, Tracetify. It took roughly three days. This post is partly about why I built it, but mostly about what a year of building alongside an AI model actually feels like — because something happened during these three days that genuinely surprised me. The ritual I finally got sick of. Every indie hacker I know does some version of this. Before starting anything new, you pick two or three competitors and try to reverse-engineer their early days. Where did the first users come from? Which launch actually worked? Was it SEO, a Reddit thread, one lucky tweet? The information is all public: backlink indexes, Wayback Machine snapshots, old Product Hunt pages, forum archives. But assembling it takes a full afternoon per competitor, and SEO tools don't really help. They show where a product ranks today, not how it got there. It's like judging a chess game by the final board position. I did this ritual before DatePhotos. Again before UGCfy. Again before FragCut. Somewhere around the fourth repetition, it clicked that I was the target user of a product that didn't exist. There must be thousands of people in my exact situation — solo builders who want to study a proven path instead of guessing, and who would rather spend those hours on their actual product. So, Tracetify: paste a competitor's URL, and it pulls from twelve public data sources, rebuilds the launch timeline with dated evidence, and gives a verdict on whether that path is still copyable. Every claim links to a source with a date on it, because I don't trust growth advice without receipts. You can run two traces free without signing up. That's the pitch, and the product section of this post ends here. What a year with Claude actually changed. When I started building with Claude a year ago, it was a very capable pair of hands. I described, it implemented. For UI work, I had to be specific: reference screenshots, spacing values, color codes. Left alone, it produced interfaces that worked and looked like nobody cared. This time, I ran an experiment. One instruction for the entire interface: "you decide." No mockups, no reference sites, no design system. What came back was a coherent aesthetic I would never have dared to spec myself: a terminal-inspired look, zero border radius anywhere, amber accent on near-black, scanline animations sweeping over report cards, a typewriter effect in the hero. Opinionated choices, consistently applied across every page. I corrected an alignment issue twice and moved one call-to-action. That was the extent of my design involvement. I want to be precise about what surprised me, because "AI got better" is not news. Earlier versions could execute taste — mine, or whatever I pointed them at. The current one (Fable 5, as of this writing) seems to have taste of its own. It made tradeoffs — where to be loud, where to shut up, what to leave empty — and those tradeoffs survived contact with real users. Multiple people asked whether I'd hired a designer. The honest answer is weirder: I got out of the way. There's something slightly uncomfortable in that sentence, and I've decided not to resolve the discomfort. A year ago, I edited almost every line it wrote. Now, I review diffs, argue about architecture, and let it own entire layers I used to guard. The division of labor keeps shifting in one direction. For the people in the same boat. If you're a solo builder, your scarcest resource isn't ideas or even skills anymore — it's attention. Both halves of this story are about the same thing: buying attention back. The research that used to cost me afternoons now costs a URL paste. The design work I used to dread now costs a sentence. I don't know where this goes long-term, and anyone who claims they do is selling something. What I know is that shipping five products in a year, alone, without burning out, was not possible for me before. It is now. Tracetify is at tracetify.com if the research problem sounds familiar. And if you've run the "you decide" experiment with your own tools, I'm curious what came back. About Author byJacob Yang|@jacobpark Founder & Solo Builder at Indie maker running five AI products solo — Tracetify, DatePhotos AI, UGCfy AI, FragCut. Building in public with Claude as my cofounder. Read my storiesAbout @jacobpark TOPICS ai-and-ml #indie-hackers#ai#claude#solopreneurship#competitor-analysis#building-in-public#ai-products#bootstrapping