ChatPRD

ChatPRD's product graph was easy to build but too close to competitors. Claire Vo built ChatPRD's product graph, which pulls everything a company knows about customers into a semantic graph and an auto wiki for agents, almost in one shot. Customers called it interesting, but she felt it matched competitors feature for feature and belonged in the trash.

She doubted the ROI, wondered whether the web interface should be agent first, and wanted something surprising that competitors could not imagine. Her diagnosis is the parity trap: competitors talk to the same customers and "they're all arriving at the same obvious conclusion and building the same obvious product."

Lenny's Podcast▶ 10:51

Keep a feature flagged off until you have conviction it is right. Vo kept the working product graph behind a feature flag, rebuilding it repeatedly, because customers would build on it once they saw it and shipping the wrong thing would burn their trust.

She frames each ship as a probe (a small bet), an experiment (a durable test of a conviction) or a promise (something customers can build on), and the product graph sat in the promise category. Quote: "code is abundant customer trust is not."

Lenny's Podcast▶ 19:26

ChatPRD's AI factory tripled output without proving any of it mattered. Vo's PRs are up 3x, she rearchitected ChatPRD 70 times "because I could", and customer fixes run automatically from intake through a bot to a fix, with issue tracking replaced by just shipping PRs. She calls the result "code to nowhere".

Her test for teams: ask "whose revenue is going up proportionate with the amount of PRs." She suggests measuring big swings taken per month instead of PRs or revenue per headcount, with the evidence that would prove or kill a conviction defined up front.

Lenny's Podcast▶ 4:52