The labs want your clients now

Something shifted in the last few weeks. The large AI labs, the ones that used to position themselves as pure platform companies, are now moving into territory that studios and agencies have always occupied. Not quietly, either. The announcements are deliberate. The messaging is confident. And the services they are describing, strategy, implementation, bespoke deployment, sound a lot like what we do.

The honest question is whether this is a genuine threat or just another wave of lab overconfidence. My read: it is both, and conflating the two will get you into trouble.

What is actually happening

OpenAI has been building out what it calls an applied research and deployment function. Anthropic has quietly staffed a professional services arm. Google DeepMind is embedding teams inside enterprise accounts. These are not sales teams. They are operators. They sit with clients, scope problems, design workflows, and ship things.

This is a structural shift. For the first years of the current AI wave, the labs competed on model quality and API access. Now they are competing on outcomes. That is a different game, and it pulls them directly into the room where agencies have always made their living.

The pitch to enterprise buyers is straightforward. why go through an intermediary when you can work with the people who built the model? It is not a subtle argument. For certain buyers, particularly large enterprises with the budget to pay lab rates and the appetite for perceived credibility, it will land.

Where the labs are strong and where they are not

Let us be precise about this instead of hand-waving.

The labs are strong at depth on their own models. If you are deploying GPT-4o in a complex context-window architecture, nobody understands that better than the team that trained it. They are also strong at credibility with certain procurement committees. A logo on a contract matters in some organisations.

What they are not strong at is breadth of context. A studio that has shipped twenty products across five sectors has pattern-matching that no internal applied team at a lab can replicate quickly. The labs are also expensive. Their professional services rates, where they have disclosed them, are not competitive with a well-run boutique. And they are slow to turn. Enterprise services divisions inside research organisations move on research-organisation timelines.

The more important limit is cultural. Doing good agency work requires a specific kind of relationship with a client. It requires the willingness to tell someone their brief is wrong, their timeline is delusional, or that the feature they want most is the one that will sink the product. Labs do not have a culture of doing that. Their culture is publication, not provocation.

The real risk is not competition, it is platform dependency

Here is where I think most studios are looking in the wrong direction.

The threat that keeps me up at night is not that OpenAI will poach our clients. It is that the work we do becomes inseparable from a single vendor's infrastructure, and then that vendor changes its pricing, its terms, or its priorities.

We have seen this pattern before. Agencies built practices on Salesforce. They built practices on Adobe. The platform wins, extracts margin, and the agency either gets acquired or gets squeezed. AI is running the same playbook faster.

The studios that are most exposed are the ones whose entire value proposition is "we implement this specific model for your use case." That is not a studio. That is a reseller with a nicer website. The labs will commoditise that position in eighteen months.

What is harder to commoditise is judgment. Knowing when not to use AI. Knowing when a simpler system, a better brief, or a clearer process would do more than another model call. That requires someone who has failed enough times to know what failure looks like early.

What to do about it

Three things worth doing now, none of them involve panicking.

Stay model-agnostic on infrastructure. If your architecture can swap the underlying model without a rewrite, you are insulated from a lot of vendor risk. This is good engineering anyway. Abstract your integrations. Do not build your differentiation into a wrapper around one provider's API.

Go deeper on domain. The labs will go wide. They will offer services across every vertical because they have to justify the team size. A boutique studio can go narrow, which means going deep. Depth in one sector, one type of problem, or one stage of the product lifecycle is defensible. Width is not.

Make the relationship the product. This sounds soft, but it is not. The highest-value thing a studio can do is become the organisation that a founder or CPO calls before they write a brief. That positioning is not built with a capability deck. It is built over years of honest conversations. No lab is going to out-relationship you if you have been in the trenches with a client through three product cycles.

The labs moving into services is real. It will reshape the market. Some studios will not survive the adjustment. But the ones that treat this as a forcing function to get sharper, not broader, will come out with stronger positions than they had before.

A closing thought

Every few years someone announces that the agency model is finished. The accountancies would eat the creative shops. Then the consultancies. Then the product studios. Then the in-house teams.

The agency model is still here. Not unchanged, but here. Because good work requires people who are close enough to care about the outcome and far enough away to see the problem clearly. No amount of lab funding changes that dynamic.

The question is not whether your studio survives the labs moving into services. The question is whether your studio was ever selling something that only a studio can sell. If the answer is yes, you have work to do, but you have a foundation. If the answer is no, that is the real problem, and it predates this week's announcements.