Garry Tan (CEO of Y Combinator) has open-sourced gstack a couple of weeks ago. gstack is an open source setup for Claude Code that turns one AI coder into a small team of specialist roles, like product reviewer, engineering reviewer and designer. It lives in Claude Code and I wanted to have it scrutinize my business plan for the custom storybooks.
I opened Claude Code and ran /plan-ceo-review in the folder with the business plan. It triggered the /office-hours workflow, which guided me through six questions designed to test demand, the status quo, and whether the idea has real potential.
It really questioned my answers. It didn’t accept my initial half-ass answers. It continued asking me things like: be specific, tell me more, or “combination” is too vague for me to work with.
It forced me to think deeper about idea and the execution. One valid concern was the extra friction in my payment flow: first buying a PDF, then paying again to turn it into a physical book. The suggestion was simple: charge the full amount upfront and rely on a strong refund policy. I changed the plan accordingly.
A big surprise for me was that he mentioned a ~$30 CAC (Customer Acquisition Cost) for this niche. Much higher than what I expected. However I think that focusing on the Dutch consumer market it could be significantly lower. Especially with a tighter acquisition execution.
In the end I changed some pretty significant parts of the plan, and then Claude’s ugly usage limits hit again. Annoying, because I was just getting warmed up and this workflow felt genuinely useful.
Key Insight
Usage limits are annoying, but understandable as the costs are probably significantly higher than what they currently charge. But they also train a scarcity mindset: they make me hesitate to use AI fully, even when that would be the best way to work. The hesitation may stay when the limits themselves are gone when AI compute gets exponentially cheaper.