Three announcements in one week:
OpenAI launched a $4 billion subsidiary called the OpenAI Deployment Company, backed by TPG, Bain Capital, and Brookfield, to “help organizations build and deploy AI systems they can rely on every day.” They bought a 150-person consulting firm called Tomoro to staff it.
Anthropic and PwC announced a partnership putting Claude in front of hundreds of thousands of PwC professionals, with PwC spinning up an entirely new Office of the CFO business unit anchored on Anthropic’s technology.
And Anthropic released Claude for Small Business, embedding Claude inside the tools small companies already pay for: QuickBooks, HubSpot, PayPal, Canva, Docusign, Microsoft 365.
Read those together. The two largest AI labs in the world spent a week telling us, in three different ways, that the bottleneck on AI value is not the model. It’s everything that has to happen between the model and a decision someone actually makes.
That’s not a frontier-research problem. It’s a translation problem. And it’s the problem most growing businesses are quietly losing right now.
What the vendors stopped pretending
The original pitch was simple: the model gets smarter, you give it to your team, productivity goes up. Buy the seats and the value shows up.
It didn’t. Not at the rate that was promised, and not in the places that mattered.
So the labs reorganized around the gap. OpenAI didn’t launch a research lab this month. They launched a consultancy. Anthropic didn’t ship a new model. They shipped distribution into the tools where small business owners already live, and a partnership with a Big Four firm whose entire job is helping enterprises implement things.
What the vendors stopped pretending is that capability equals adoption. A model that can draft a strategy memo doesn’t produce a strategy. A workflow that can summarize a P&L doesn’t change what a CFO decides. The model is doing its part. The part that wasn’t getting done was the part nobody was calling AI in the first place: deciding what the AI was supposed to be doing, in operational terms, against a goal someone owned.
That’s the translation layer. It’s where strategy becomes execution, where capability becomes habit, where a tool becomes a decision.
And it’s been understaffed in almost every AI rollout I’ve seen.
Why this matters for a $5M company
If you run a business in the two-to-fifty-million-dollar range, the new vendor strategy reads two ways.
Read one: the AI labs now have professional services arms that will help you deploy. Convenient.
Read two: the AI labs now have professional services arms that will help you deploy. Pay attention to that.
Both reads are true. The convenience is real. So is the dependency it creates.
If you hire the OpenAI Deployment Company to install AI into your workflows, you get speed. You also get a wiring job done by someone whose long-term incentive is keeping you on their stack. The same dynamic that made you cautious about cloud lock-in applies here, with sharper teeth, because AI deployments touch how your people think, not just where your data lives.
The companies that compound value from AI over the next five years won’t be the ones who outsourced the translation layer to the vendor. They’ll be the ones who treated it as a capability they had to build inside the business.
Building that capability isn’t an IT project. It’s a leadership question. Who, on your team, owns the gap between “Claude can do this” and “here is the decision we make differently because Claude does this”? If the answer is a vendor, you’ve rented a capability. If the answer is your own people, you’ve built one.
The diagnostic question I’d actually ask first
Before you sign anything with OpenAI’s new arm, a consultancy, or anyone else, answer one question.
What decision in your business is currently bad, and why?
Not “where could AI help.” That question lets vendors do your thinking for you. Start with the decision. A pricing call you’re getting wrong. A sales conversation that keeps stalling at the same point. A hiring loop that produces the wrong shortlist. A marketing budget that’s being spent against a strategy nobody updated.
Once you’ve named the bad decision and the reason it’s bad, you know what good would look like. From there, the AI question becomes specific: does this tool help me reach that better decision faster, more consistently, or with less wasted motion? If the answer is yes, buy. If the answer is “it could,” you don’t have a deployment problem yet. You have a diagnosis problem.
The vendors won’t ask you that question. It’s not their job. It’s yours.
What I’d tell a CEO this week
I’d tell them the same thing the announcements are telling them, with the marketing stripped off.
The hard part of AI is not buying it. It’s owning the translation between what it can do and what your business decides. The labs just spent four billion dollars and three press releases confirming that.
You can outsource that work or build it. Either choice is defensible. Defaulting to “the vendor will handle it” is not a choice. It’s how growing companies end up two years in, with three AI seats per person and no clearer view of the decisions they were trying to improve.
Name the decisions. Own the translation. Decide whether AI capability is something you’re going to build inside your business or rent from someone whose interests aren’t yours.
That’s where the next five years of compounding actually live.
Aaron Douglas is the founder of Auspicious LLC, a fractional CMO practice. He’s the author of AI Empowered and host of the AI Empowered podcast.