Anthropic last week announced a $1.5 billion joint venture. Goldman Sachs. Blackstone. Hellman & Friedman. The goal: embed engineers inside companies to help them actually use AI.
The company that built the model is now sending people to teach you how to use it.
Read that again.
Here's the thing. "Consulting" is not a word anyone in tech says with pride. McKinsey, Deloitte, Accenture — you know the shape of the joke. Ninety-page decks. Three months of workshops. Nothing changes. But Anthropic is walking into that space — not reluctantly. They raised $1.5B for it.
Because they had no other move.
AI companies have spent years saying the same thing: "The model is ready. Go use it." ARR grew. Demos got applause. And 18 months later, enterprise customers were still asking the same question — "How do we actually integrate this into how we work?"
Nobody had a good answer. Now Anthropic does. And getting a seat at that table costs $300 million.
Three weeks ago I sat in on a demo for a clinical software product. It had an AI-powered prescription suggestion feature. It worked. The interface was clean. The doctor opened it, looked at it, closed it.
"Don't you trust it?" I asked.
"I trust it. I just don't know when to use it."
The model was ready. The doctor wasn't. And nobody had defined closing that gap as part of the product.
This is the question every PM should be forcing on themselves: is the product actually being used, or just usable?
They are not the same thing.
"Usable" means the feature exists, it works, it passed QA, it's live. "Being used" means the user has woven it into their daily workflow, trusts the output, and would notice if it disappeared tomorrow.
Most PMs ship the first and measure the second. Except what they're actually looking at — activation rates, feature clicks — doesn't tell you which side of that line you're on.
Activation rate can be high. Adoption can still be zero.
What Anthropic just announced is, if you read between the lines, a public admission: "Our model is powerful. But making it actually useful requires redesigning how businesses operate. And your team doesn't have the people to do that."
OpenAI announced something similar the same week.
This isn't a coincidence. This is an industry finally accepting that the thing it spent years calling the customer's problem is actually its own.
For PMs in vertical SaaS — healthcare, dental, legal, finance — none of this should be surprising. You already know that the distance between "feature shipped" and "feature adopted" is measured in months, not clicks.
Building an AI-assisted anamnesis feature into a clinical product takes two sprints. Getting a clinician to actually trust it — to stop second-guessing every output and start using it as a default step in their workflow — that's a six-month project. It includes training. It includes a lot of "why did it say this?" conversations. It includes redesigning how intake works.
None of that is written on the roadmap.
"We shipped it. If they're not using it, that's on them."
Nobody says it out loud. But look at how most teams prioritize the next sprint and you'll see that sentence written in invisible ink across every roadmap.
Closing the adoption gap: the sprint ticket nobody writes.
Anthropic just invoiced it at $1.5 billion.