What it is
An AI system that hunts for products I should build: scans X, Reddit, Product Hunt and Hacker News, tracks a curated list of builders, runs signals through a pipeline (signals → streams → research → ideas → products), grounded in my profile, not generic startup advice. It started as an internal tool (OpportunityOS) and became saashunting.com.
Why I built it
This is the obsession under everything else I do: don’t just build products, build the machine that finds the problems, then the machine that builds the solutions. If agents are the future of software, the meta-layer is where the leverage lives.
What happened, honestly
The v1 aggregation-and-judging engine worked well enough to spawn real products: PermitFlag and ShopTrust came directly from its signals. Then I tried a more ambitious version: model evaluation functions from products with verified revenue, run harvested ideas through them. That one never shipped; the problem turned out to be much harder than I could grasp, and I don’t think current models are good at it yet. The public product today is a mess: it aggregates but doesn’t yet judge well. Rebuild or ditch; I’ll rebuild, eventually.
What I learned
The machine found ideas but not pain. I never gave it the right constraints to distinguish daily suffering from interesting signals. Idea discovery was never my bottleneck anyway; validation and distribution are. The next version has to model those.