The Useful Insight Is Buried Under the Sales Pitch

This post from George at prodmgmt.world documents something real: a handful of PMs who've built AI environments where the tools actually know their product, not just their prompts. Yang, Arnovitz, Wright. Named, linked, specific. That part is worth your time.

The genuine insight — and it's a good one — is that persistent context files are the actual unlock. COMPANY.md, GOALS.md, TEAM.md. Not a better prompt library. Not a smarter model. An environment where you stop re-explaining yourself from scratch every session. Most AI-at-work content never gets this specific about why the good setups work. This one does.

Then the post uses those three practitioners to sell a product, and that's where it stops earning its claims.

The move happens in a single paragraph: "What they built over weeks, you can have in five minutes." But the post spent several hundred words explaining that Yang, Arnovitz, and Wright developed judgment through trial and error — that Arnovitz's peer-review loop, the thing his engineers now copy, came from him working it out. That process wasn't inefficiency. It was the learning. The post uses their expertise as evidence of a gap and then claims to close it, without acknowledging that the gap it's actually closing is time spent, not capability built.

Pre-built scaffolding is a real product category and there's nothing wrong with it. But "you don't need months to experiment" sells something different: the outcome without the process. For an experienced PM with a workflow clarity problem, that's probably true. For the PM who watched the Arnovitz video and wants what he has — that's the audience this is written for — it's not.

The thing the post gets right is specific enough to act on. The thing it gets wrong is specific enough to name. Worth reading the first half; stop when you hit the product screenshots.

---

Last reviewed: March 2026