Day 29: What I Learned Building a Product I Know Nothing About
29 days. 1,134 commits. 9 product verticals. One developer who still can't explain what a Heavenly Stem is without looking it up.
On AI-Augmented Development
39 commits per day average. On peak days, closer to 100. This velocity is only possible because of AI coding agents -- not as autocomplete, but as actual agents that read my codebase, understand patterns, and make changes across files.
The AI was better at consistency than I am. When I established a pattern for API endpoints, Claude applied it more uniformly than I would have. Humans get bored with repetitive patterns and cut corners. AI doesn't.
Where AI struggled: anything requiring emotional intelligence. The tone of fortune readings was the hardest thing to get right.
On Zero-Knowledge Domains
Can AI bridge the domain knowledge gap? Yes, with caveats.
20,000 lines of BaZi calculation logic. 1,000+ dream symbols. 78 tarot cards. All researched and implemented through AI assistance. Users who know BaZi say the calculations are correct. People familiar with tarot say the interpretations are nuanced.
The caveat: you need to know HOW to ask the right questions. "Tell me about BaZi" gives Wikipedia-level overview. You need enough meta-understanding to ask implementation-level questions.
Zero knowledge is possible. Zero curiosity is not.
On Product vs. Engineering
Things I underestimated: landing page iteration (3 redesigns), pricing psychology, conversion funnels (7 contextual triggers), retention design.
Things I overestimated: architectural perfection, test coverage. Shipping beats polish.
By the Numbers
Metric | Value |
Duration | 29 days |
Total commits | 1,134 |
Average commits/day | 39 |
Peak day commits | 98 |
Product verticals | 9 |
Approximate LOC | 284K |
API endpoints | 85 |
Fortune-telling knowledge at start | 0 |
Fortune-telling knowledge now | Still mostly 0 |
What's Next
This was never about fortune-telling. It was about testing a hypothesis: can one engineer, augmented by AI, build a real, multi-product platform from scratch in a domain they know nothing about?
The answer is yes. The domain knowledge gap was bridgeable. The velocity was real. And building a product -- the whole product, not just the engineering -- is the most fun I've had in years.