The Holding Companies Are Selling Agents.

At CES this year, the big holding companies made the same announcement in four different ways.

WPP launched Agent Hub inside WPP Open, a platform with more than 75,000 internal users. Omnicom recast Omni as an agentic operating layer. Stagwell launched The Machine. Havas launched AVA. According to a new report from Acalytica, State of Agentic Deployment in Marketing, the agency pitch at the top of the industry is moving from people-hours to curated, governed agents.

That is a big change. It is also not your playbook.

The holding companies have a specific problem. Their model is a pyramid. A few senior people sit on top of a large base of junior staff who do the execution. AI compresses exactly that execution layer. So they are doing the logical thing: turning the knowledge inside that base into software, and selling the software.

The cost of this shows up in headcount. Omnicom and IPG cut about 8,200 roles before their merger. The agents are not an add-on to the pyramid. They are a replacement for its lower floors.

If you run a $1M to $10M agency, you cannot build an agent platform for 75,000 users. You should not try. Your advantage is somewhere else.

Look at what the platforms did this year.

Google, Amazon, and Meta now open their ad systems to outside agents through the Model Context Protocol. In April, Meta released connectors that let Claude or ChatGPT create campaigns directly, paused by default. Later, it opened access to any developer app.

This means your client can run their Meta ads from the same AI assistant they use to write emails. The agent is no longer the scarce part. It is available to everyone, including the people who hire you.

When the tool is available to everyone, the tool stops being the thing clients pay for.

The Acalytica report pulls together the major 2026 surveys, and the numbers tell a clear story.

About half of marketing organizations now use AI agents. Forrester and the 4A's found that roughly half of US agencies use agentic AI for execution. But only 13% of AI-using marketers in Salesforce's study use agents that are genuinely self-directed. And just 6.5% of organizations measure their AI investments against defined benchmarks.

Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027. The reasons are rising costs, unclear value, and weak risk controls.

So the market has many agents and very little accountability for them. That gap is the opportunity.

One number from the Forrester agency data matters most for our audience. 63% of agencies say staff productivity is their main reason for adopting agents. Yet 61% still book AI as a cost of doing business, not as a revenue line.

Think about what that means. Agencies are getting faster, and they are giving the speed away. If you bill by the hour and AI cuts the hours, your revenue goes down while your client gets the same result. Forrester also warns that agencies risk mistaking efficiency for effectiveness.

This is why pricing has to change before the tools do. You price the outcome, not the time. The productivity gain then stays in your business instead of disappearing from the invoice.

The report offers a simple rule for when a workflow can get more autonomy. You must be able to measure its outcome against a baseline. You must be able to reverse its actions. And you must have a named owner for its failures. If any one of these is missing, the workflow stays where it is.

Read that list again. Measurement, reversibility, and ownership. None of these is software. All three are judgment.

This is the work a senior-led agency does well. In an obelisk structure, experienced people stay close to the work. They set the baseline. They decide what the agent can touch. They take responsibility when something goes wrong. A pyramid struggles with this because the people who once did the execution are being replaced, and the senior layer is too far away from the details.

Your clients do not need another agent. They already have access to the same ones you do. What they need is someone who can tell them which agents to trust, what each one costs, and what it actually changed.

You do not need a platform announcement. You need three things.

Know your own agents. List every AI tool and connector your team uses on client work. Decide what each one can read and what it can change.

Set baselines before you promise results. Record the hours, cost, and cycle time of a workflow before an agent touches it. Without a baseline, you cannot prove value, and you cannot defend your fee.

Sell the accountability. Put your name on the outcome. Tell the client which decisions a senior person makes and which ones a machine makes. That clarity is now a product.

The report ends with a useful line of thinking: 2026 was the year marketing deployed agents, and 2027 will be the year it has to account for them. The holding companies are betting on more agents. The better bet for a founder-led agency is to be the one in the room who can account for them.

Want more on building a senior-led agency that grows on its own terms? Subscribe to The Playbook at agencyfocus.xyz.

Source: Acalytica, State of Agentic Deployment in Marketing, October 2026.