The largest AI labs are signaling, through how they are spending, where the real value in AI is shifting.
OpenAI acquired a 150-person applied AI consulting firm called Tomoro and is folding it into a new deployment-focused venture. Around the same time, Anthropic, with Blackstone, Goldman Sachs, and Hellman & Friedman among its partners, backed a $1.5 billion firm aimed at helping companies adopt AI at scale. Both announcements landed within the same week of early May. Two years ago, these were the companies telling you the model was the product. Today, they are spending billions to build out the layer of work that sits next to the model.
That is a meaningful tell. The people closest to the technology have looked around and concluded that the technology is no longer the constraint. The constraint has moved to the human and process layer that wraps around it.
That layer is what you would call adoption in a tidy slide. In practice it is messier than that. It is the question of which part of your work is actually a good candidate for AI and which part is not. It is the workflow the tool is supposed to fit into, and what that workflow looks like before and after. It is the handoff between a person and a model and back, and who is responsible for what at each step. It is the conversation about which old habit gets retired and which new one gets installed. It is the meeting where someone says, out loud, that this is going to change how the work gets done, and we are going to be patient with each other while we figure it out. None of that is technical. All of it is hard. And none of it scales the way a model does.
For most businesses, access to capable models is no longer the primary constraint. The constraint now is operational. And if you have been quietly wondering whether you are somehow behind on AI, that is the frame that matters. The hard part right now is adoption, and it is the hard part for everyone, including the biggest companies in the world. The $1.5 billion firm and the 150-person consulting acquisition are the most honest version of that point we have seen so far.
There is good news inside that news, especially if you run a small business. You do not need a $1.5 billion firm. You do not need a 150-person engineering team. You need to be honest about which piece of your week is repetitive enough, well-defined enough, and bounded enough that putting AI on it would save you something real. That is a smaller question, and the only person who can answer it for your business is you. Then you try the thing, you measure what it changes, and you do the next one. That is what adoption looks like at your scale. It is the same shape of work the labs are now spending billions to help bigger companies do, only without the procurement reviews, the security audits, the change management committees, and the eight-month implementation timelines that come with enterprise life.
The other piece of good news is something the spending decisions quietly reveal. A small business can adapt to a new tool in days. A large enterprise cannot. That is not because the small business is smarter. It is because the small business does not have to sell the change through fourteen layers of stakeholders, six committees, two legal reviews, and a quarterly steering meeting. You can decide on Tuesday that a piece of your week is going to look different by Friday, and on Friday it can look different. That capacity to actually change how the work gets done, quickly, in response to what you are seeing, is exactly what the labs are now spending billions to build inside their largest customers. You already have it.
This is where value is being created right now, and it is not a model selection problem. It is an execution issue.
The businesses seeing results are not starting with broad transformation efforts. They are picking one specific task that is repetitive, well-defined, and bounded, putting AI on it, measuring impact, and expanding from there. Strategic restraint becomes the moat in this environment, because the place where AI does the most damage is the place you cannot rebuild on demand, and the way through is one careful, well-chosen problem at a time.
Model capability still matters. But outcomes are increasingly determined by how well the surrounding work is designed and managed.