If AI Handles the Execution, Who Sets the Strategy?

💡 Summary: Key Takeaways

• AI can reduce an agency's repeatable execution work, but it does not replace that work with strategic capability. Problem definition, judgment, measurement design, and accountability require a different investment in people and organisation.
• Execution capacity and decision capacity are not the same. An agency may run defined work reliably at scale without being equally strong at deciding what the client should do and why.
• Adding “strategy” or “outcomes” to a contract does not close the capability gap. Clients need to know who will create that value, what authority they will have, and how the result will be measured.

If AI lowers the value of an agency's repeatable execution work and cuts the specialist hours it requires, what is left for the agency? The client may be able to bring much of that work in-house using a few AI agents and very little human input. With a $20 subscription to a tool such as Codex or Claude Code, the client may already have enough tooling to handle much of the execution. Someone still has to set the strategy—assuming anyone involved has the capability to do it.

In some of the agencies and projects I have worked with, the team's core strength was running defined work reliably. The organisations I encountered often had substantial execution capacity and more limited strategic capacity. Even a thin layer of strategy could deepen the client's dependence when the client did not want to build the execution capability in-house. Problem definition, strategic trade-offs, measurement design, and accountability for outcomes were concentrated in far fewer roles.

The work agencies showcased, their public LinkedIn posts, and employee profiles offered similar clues about where capacity had accumulated. Execution and coordination were visible across large teams. Judgment, measurement, and responsibility for results appeared to sit with a much smaller group.

These observations are not a study of the agency market, and they say nothing about the ability of any individual employee. They come from a limited set of working relationships and public signals. The relevant question is organisational: what kind of work was the agency designed to scale?

AI is now compressing precisely the repeatable work that many of those structures were built to multiply. Yet the capacity it releases does not turn into strategy by itself. An agency must answer more than how it will produce the same output with fewer human hours. It must also show which client problem it can define more clearly, which decision it can improve, and which outcomes it can reasonably accept responsibility for.

Reducing execution does not create strategy

In Pay Less, Grow More: Agencies in an Agentic AI-Era, Avinash Kaushik argues that marketing agencies need to shift their value away from campaign setup, bid adjustments, asset variations, and reporting towards strategy, advanced analytics, governance, senior judgment, and business outcomes.

His strongest point is that work already being handled by advertising platforms will become increasingly difficult to sell as agency labour. He also gives precise estimates for potential fee reductions and new areas of work, but does not provide a method or representative sample behind them. Those figures are better read as an experienced practitioner's estimates than as market evidence.

The harder part is not adding the new work to a scope of services. It is having the capacity to do it.

An organisation that has spent years scaling repeatable work through large teams does not become a strategy and measurement partner when that work contracts. A service description can change in weeks. Experience, decision rights, working practices, and accountability cannot.

Execution capacity is not decision capacity

Both capacities can exist in the same engagement. They still perform different functions:

Execution is not trivial work. It demands quality, continuity, context, and error control. An agency can create real value by providing an operation that a client does not want to build internally.

But running the work well does not determine whether it is the right work. Producing more creative variations does not decide which message fits the brand and customer. Preparing a report faster does not explain what caused the change in performance. Automating a campaign does not prove that the client chose the right commercial objective.

AI makes this distinction harder to ignore. As production and coordination require fewer human hours, an agency will be judged less by how much work it moves through the system and more by the quality of the decisions it improves.

An AI licence is not workforce redesign

Giving an agency team new tools can accelerate repeatable work. The same people may prepare an initial research brief, content variations, campaign structures, or reporting summaries in less time.

That gain matters. It does not teach the team the client's business model, explain why a market is changing, or establish which measure can be trusted. The ability to choose between conflicting signals, reject a weak request, and take responsibility under uncertainty does not arrive with a software licence.

This is not a reason to divide employees into “good” and “bad.” The more useful question is what the organisation has expected and enabled them to do. Someone who has spent years being assessed on the speed and accuracy of defined work does not acquire the authority to challenge a client's objective overnight.

At least four parts of the operating model need to change:

1. Hiring: Look beyond channel and tool experience to problem solving, measurement, and commercial judgment.
2. Role design: Do not confine strategy to a few senior people in pitch meetings. Connect it to the decisions made throughout the work.
3. Authority: Let the team challenge a weak brief or the wrong objective. Accountability without decision rights is largely symbolic.
4. Development: Give less experienced employees a role in defining problems, designing experiments, and reviewing outcomes—not only in producing the work.

Without those changes, reducing headcount may lower cost. It does not demonstrate that the agency has improved the client's decisions.

Strategy cannot remain a thin senior layer

An agency's most senior people may frame the problem well in a sales meeting. The test is whether that judgment remains available during the work itself.

If strategy depends on a small group who appear mainly in the pitch, day-to-day work continues under the old model. An execution team without access to client context, data, or decision rights may produce faster, but it is unlikely to make better decisions. The senior team cannot make or oversee every consequential decision on every account either.

Job titles alone therefore reveal little about how capability is distributed. Better questions are:

• Who defines the problem?
• Who sets the success criteria?
• Who connects data from advertising and analytics platforms to commercial outcomes?
• Which decisions remain with the client, and which belong to the agency?
• When results differ from expectations, who is responsible for explaining what happened and deciding what to change?

LinkedIn posts are also signals, not evidence. An agency may publish constantly about new tools and the volume of work it produces while saying little about trade-offs, measurement limits, or how it evaluates results. That tells us something about what receives attention, but it cannot establish the agency's true capability. For that, a client needs to see the project team, its working practices, and examples of real decisions.

Pricing and capability are different questions

In Who Captures the Productivity Gains from AI?, I examined how the economic benefit is divided between clients, employees, employers, and service providers when the same work requires fewer human hours. That article focused on how AI changes the basis on which professional work is priced.

This question comes earlier: does the agency possess the new value it wants to price, and who in the organisation can produce it?

A fixed project fee, retainer, or outcome-based agreement does not create that capability. Value-based pricing means little unless the agency can make a distinct contribution to the client's decisions.

Accountability for outcomes also needs a clear boundary. An agency cannot reasonably own a business result if it lacks authority over the levers that shape it, or if the change cannot be measured with sufficient confidence. Renaming the price does not solve the problems of control and attribution.

Not every agency needs to become a strategic partner

Some agencies can continue to create value as clearly defined execution specialists. Complex production, data preparation, quality assurance, channel operations, or deep expertise in a particular field may provide capabilities that a client has no reason to build in-house.

That is a legitimate choice. The agency can show why it performs that work more reliably, quickly, or with less risk. Alternatively, it can invest in decision capacity. Keeping the same staffing model while renaming the work “strategy,” “transformation,” or “outcomes” does neither.

What I would ask before hiring an agency

I would look past the tool list and ask:

1. Which decision will you improve? Can the contribution beyond producing the work be described in concrete terms?
2. Who will prepare that decision? Will the senior people in the pitch remain involved throughout the engagement?
3. How will we measure success? Are advertising-platform metrics separated from commercial outcomes, and is the baseline clear?
4. Which parts of the outcome can you influence? Are pricing, product, sales, and operational decisions outside the agency's control made explicit?
5. Which work will AI reduce? Will the released capacity go into producing more material, or into better decisions and measurement?
6. Can you show why you chose not to do something? Strategy is visible in the options rejected as well as the work approved.

The answers may reveal more than the size of the team or the number of AI tools it uses.

Kaushik's direction is plausible: agency value may shift from repeatable execution towards judgment, measurement, and governance. But automation does not make that transition inevitable. It is a separate organisational decision.

AI can reduce the human hours required for execution. It does not give an agency strategic capability. Building that capability means changing who defines the problem, who has the authority to decide, how success is measured, and who accepts responsibility for the outcomes they can control.