This is part of a daily summer series where I write about what and how I'm building at ChatPRD, the #1 AI platform for product managers.
This is going to be a short one, because the final outcome is TBD, but I'm continuing to bang away at this Product Graph aka “semantic oracle”, which is a complicated way of saying I want to take all your random product data scattered across your company and tell you what's actually going on.
(By “you” I mean “an agent that I build and sell to you.”)
As I said before, this is not an easy problem to solve, so I've been throwing tokens and scripts at about 5000 internal data points to see if I can get a solid, repeatable prototype going.
In the meantime, I've found a fun little hack I hope you all find useful.
Using your coding agent to prototype your agentic harnesses
Guess what? Codex can access its own auth to prototype it's own AI feature