Strip away the software and the job titles, and the agency has worked the same way for a hundred years. You hire smart people. You sell their hours. You manage clients, build the work by hand, report on what happened, and bill a retainer or a percentage of spend. The tools changed (from media schedules on paper to dashboards in the cloud), but the underlying machine did not. Value is created by humans doing repeatable work, and the business scales by adding more humans.

That machine is about to be rebuilt, because the thing it runs on, skilled human execution, is the exact thing AI has just got good at.

What is actually breaking

It is tempting to file this under "AI hype" and move on. Don't. The pressure on the agency model is not coming from a press release. It is coming from three directions at once.

The work is being commoditised. A large share of what a paid media team does every day (building campaigns across platforms, enforcing naming conventions, pulling reports, reconciling data, spotting fatigue, drafting the weekly deck) is structured, repeatable execution. It is exactly the kind of work that AI does well. When the work can be done by software, the hours that work used to justify start to look expensive.

Clients are noticing. Mid-market and DTC brands are already pulling work in-house or switching to AI-native vendors. The conversation in procurement has shifted from "what does your team cost" to "why am I paying for a team to do something software can do." An agency that bills for execution is now competing, directly, with a tool that does the execution for a fraction of the price.

The margin is in the work that is disappearing. The uncomfortable truth is that agency margin has always been thickest on the routine, leverage-able work: the junior hours, the build, the reporting. That is precisely the layer AI is eating first. The strategic work that is hardest to automate is also the work that was never where the money was.

Why now, and not three years ago

People have predicted the end of the agency before and been wrong. What is different this time is that several things became true at once, and none of them were true in 2022.

  • The models can actually do the job. Frontier models crossed the line from "drafts a human edits" to "runs real production workflows end to end": planning, building, optimising, with reliable tool use and long context.
  • Agents can be orchestrated. Composing specialist agents (a planner, a critic, an executor, a memory layer) held up in research two years ago. Today it ships and survives real workload.
  • Attribution broke, and unified data became survival. Post-cookie tracking, privacy changes and platform black boxes made single-channel attribution structurally unreliable. The teams that cannot unify their data can no longer answer basic questions about ROI.
  • The economics inverted. For the first time, software can deliver in minutes what a team delivered in days, and it gets sharper every campaign, because the data compounds. Humans cannot compound like that.

The fork

Every agency is now standing at the same fork, whether they have noticed it or not. There are two roads.

The first is to compete with AI: to keep selling human hours against tools that do the same execution faster and cheaper, and to defend the retainer on the strength of the relationship alone. Some agencies will survive this way for a while, especially at the top end where the work is genuinely bespoke. Most will be squeezed.

The second is to run on AI: to put the routine execution on software, keep the humans on the judgement, and deliver more for each client while scaling the client base without scaling the headcount. The agency that does this does not lose its people to AI. It points them at the work that actually needs them: strategy, creative direction, the client relationship, the decisions a machine should never make alone.

The agencies that win the next decade will not be the ones with the biggest teams. They will be the ones that built systems around the work that does not need a human in the loop.

What the new agency looks like

Picture the same agency, twelve months from now, running on an agent workforce sitting on a clean, cross-channel data lake. Campaigns are briefed in plain language and built across every platform in minutes, with naming and measurement enforced at creation so the data is never dirty. A team of specialist agents watches every account around the clock, catches fatigue and pacing problems the moment they appear, and surfaces the moves that matter. Reporting writes itself, from one reconciled source of truth, in the client's own context.

The strategists are not gone. They are doing the part of the job they were actually hired for, across more accounts than they could ever have touched by hand. The agency delivers enterprise-grade, cross-channel work at a cost structure that used to be impossible, and the relationship gets stickier, because the data lake compounds in value the longer the client stays.

This is not a thought experiment. It is the difference between an agency that treats AI as a feature it bolts on, and one that treats it as the operating system it runs on.

The honest part

None of this means agencies are finished. It means the model is. The relationships, the taste, the strategic judgement, the trust: those are more valuable than ever, precisely because the execution underneath them is being automated. What is ending is the part of the business that sold hours for repeatable work, because that work no longer needs the hours.

The window to move is open now, and it is the kind of window that closes. The agencies that adapt early will absorb the clients of the agencies that wait. That is usually how these shifts play out: not all at once, but decisively, and in favour of whoever moved first.

Maaten was built for the road that runs on AI: an agent workforce and a cross-channel data lake, in one platform, for agencies and the in-house teams they work with. We built the manual version of this for two decades. Then we automated it. If you would rather get ahead of this shift than be caught by it, that is exactly the conversation we want to have.