A platform PM pushing Claude past busywork into all three levels of product work, with the working templates to prove it. The stat buried inside matters most for this site: 88% of organizations still use AI as a chatbot. That's the middle, measured.

A deployment-side view of the same gap this site is about, written by someone shipping into enterprises rather than theorizing about them. Sits adjacent to the micro-app thesis — worth reading the two against each other.

The essay the entire scaling era keeps proving right: general methods that leverage computation beat human-crafted cleverness, every time, eventually. For operators, the corollary matters — don't over-engineer around today's model limitations, because the limitation is the most temporary part of the system.

Mollick keeps being the best translator between what the labs ship and what organizations can absorb. This one is about how the form of AI tools shapes what people believe they're for — which is the adoption problem stated from the other side.

Shipping the capability is not the same as shipping the behavior. NN/g documents how AI features die quietly when nobody can find them or tell what they're for. Every enterprise rollout plan should read this before the kickoff deck.

Seven years old and still the most practical checklist for designing AI that humans will actually trust. Written before agents, and it holds up better than most things written after them. Guideline 1 — "make clear what the system can do" — is doing a lot of work in 2026.

The single most efficient way I've found to stay current without drowning. Daily, sharp, and focused on what matters for people deploying AI rather than people benchmarking it. This is the firehose filter.

The distinction I keep coming back to: companies choosing "the same with less" versus "way more with a little more." Operators get to pick a lane here, and most of the interesting work lives in the second one. There's a longer post forming on this.