AI labs want to sell services. Agencies need to decide what that makes them.

Something shifted in the last few months that I do not think the agency world has fully processed yet.

The major AI labs are not just selling API access anymore. They are hiring implementation consultants. They are building partner programs that look a lot like the old Salesforce or SAP ecosystem playbooks. They are moving, deliberately and with real headcount, into the space that agencies have historically occupied: helping organizations figure out what to do with the technology and then building it.

That is a reasonable business decision for them. It is also a structural pressure on every agency that has spent the last two years repositioning itself as AI-native.

The question I keep coming back to is a simple one. If the lab that makes the model also sells the implementation, what exactly is the agency for?

This is not new, but the scale is

Tech vendors have always tried to capture services revenue. Oracle did it. SAP did it. Microsoft built a multi-billion dollar professional services operation sitting alongside a partner ecosystem it also competed with directly. The pattern is old.

What is different now is the speed and the margin logic. AI labs are burning cash on compute at a scale that requires them to find high-margin revenue wherever they can. Services, especially enterprise implementation, carry margins that raw API consumption often does not. So the move downstream into consulting and delivery is not accidental. It is structural to how these businesses need to work.

OpenAI has been building out an enterprise go-to-market team that includes solutions architects and what the industry would recognize as pre-sales consultants. Anthropic has formalized partner tiers that reward agencies for volume while also maintaining its own direct enterprise relationships. Google, through its Cloud and Workspace organizations, has been doing this for years and is simply accelerating.

The labs are not hiding this. It is in their job postings and their pricing pages. The question is whether agencies are paying attention to the signal inside the noise.

Three kinds of agency, and only one of them is fine

When I look at how agencies are responding to this, I see roughly three postures. They are not all equally durable.

The integration shop. This agency leads with technical capability: we connect your systems to the model, we fine-tune, we deploy. The problem is that this work is becoming faster and cheaper with every platform release. What took a team of four engineers six weeks in early 2023 can now be done by one engineer in a week using the lab's own tooling. The labs are also building no-code and low-code surfaces specifically to remove the need for this layer. Integration as a primary offer has a compression problem.

The AI strategy consultancy. This agency sells thinking. workshops, roadmaps, maturity assessments, frameworks with proprietary names. The risk here is that the labs are now funding their own thought leadership, publishing their own research on enterprise adoption, and embedding their own strategists into major accounts as part of enterprise agreements. When the vendor can give you strategy for free as part of a six-figure software deal, standalone strategy becomes a harder sell.

The product studio. This agency builds things the client will own: products, interfaces, workflows, internal tools. The relationship to any specific model is secondary. The primary offer is design and product judgment, and the AI capability is in service of that. This is the position that is structurally more defensible, because the labs cannot easily replicate it without becoming something they are not, which is a product studio.

None of these categories are pure. Most agencies are a blend. But the direction of travel matters.

What the labs cannot do

It is worth being precise about where the labs' services arms will actually struggle, because vague reassurance is not useful here.

The labs are good at depth on their own technology. They are not good at the organizational and political work of figuring out where AI actually belongs inside a specific company with a specific culture and a specific set of legacy constraints. That work is slow, contextual, and requires someone willing to sit in rooms where the answer is not always more AI. Labs have an obvious incentive problem there.

They are also not good at interface design and product thinking. The models are impressive. The default UIs that ship with them are often not. There is still real craft required to take a capable model and make it into something a non-technical person would choose to use every day. That craft lives in design and product, not in AI research.

And they are not good at client relationships that require genuine independence. An enterprise that wants an honest assessment of whether GPT-4o or Claude 3.5 Sonnet is the right choice for their use case is not going to get that from either OpenAI or Anthropic's professional services team. They might get it from an agency that has no commercial stake in the answer.

These are real gaps. They are also the places where an agency should be building its actual position.

The services layer question

There is a version of the agency future where agencies become resellers with a design skin. They pick a preferred lab partner, get certified, sit inside that partner's ecosystem, and take a margin on implementations. This works as a business for a while. It is also, structionally, the position of a subcontractor.

There is another version where the agency retains real creative and strategic ownership of the work. The models are inputs. The client gets something they could not have gotten from the lab directly. The agency has a point of view that is not just about the technology but about how people actually work and what good products feel like.

The first version is easier to sell right now. The second is harder to explain but more durable.

The pressure from the labs is real, and it will increase. But pressure creates clarity if you let it. The agencies that will be in good shape in three years are not the ones that tried to stay neutral or hedge every direction. They are the ones that made a clear decision about what they own and then built everything around that.

For us at Studio Hyra, that answer has always been the same. We build products and design systems that happen to use AI. The model is a material, like code or type. The work is the thing we are accountable for. That position does not depend on any single lab's roadmap, and it does not compete with their services arm.

That is a deliberate choice. I would recommend making yours explicitly, before the market makes it for you.

A few practical checks

If you run an agency and you are trying to work out where you actually stand, these are the questions worth sitting with.

Is your primary offer something the lab could replicate by hiring two more enterprise account managers? If the honest answer is yes, the offer needs to change.

Do you have a position on which model to use for a given problem, and is that position genuinely independent of your partner agreements? If not, clients will eventually notice.

Are you building things the client owns and can operate without you, or are you creating dependency on your own tooling and processes? Ownership is what justifies the relationship long term.

And finally. when the lab's own services team pitches against you, what is the one thing you can say that they cannot? If you do not have a clean answer to that, the work is not done yet.