Agentic media buying is the real AI play. Ad copy is the distraction.
by Luis Gomes, Founder & Growth Lead
If you are evaluating an agency right now and the AI pitch is "we generate 100 ad copy variations in an afternoon," you are being shown the small play. The big play is happening underneath the platforms — and most of the agencies pitching ad copy can't ship it.
Here is the actual shift, and the test for whether your agency is on the right side of it.
What the small play actually is
AI-generated ad copy is real and useful. We use it. Every operator we know uses it. It produces volume against Meta's Andromeda update — where Pilothouse Digital documented the new rule that creative needs to be at least 70% different or Meta buckets the variations together — and it lets a single creative operator do the work that used to need three.
That is a productivity gain. It moves the same architecture faster.
It is not an architectural shift. The agency still bills the same way. The team still has the same shape. The decisions about what to spend, where to spend it, when to scale a winner, and when to kill a loser still happen the same way — humans looking at dashboards, arguing in meetings, pushing changes by hand. The AI made the inputs cheaper. Nothing about the system changed.
The architectural shift is somewhere else.
What the big play actually is
The big play is closing the loop between bid, creative, audience, and outcome. Letting an agent system see all four signals at once and make routine decisions inside the loop without waiting for a human meeting. Keeping the human accountable for the consequential decisions — the brand calls, the budget reallocation, the kill switch — but moving the routine optimization out of the human's calendar entirely.
Butler/Till published the first credible numbers earlier this year. Their first agentic media-buying tests cut media costs and supply-chain costs by roughly 40%. That is a structural change. You do not get a 40% cost reduction by writing better ad copy. You get it by removing the layers of human coordination that were doing the routing in the old model.
This is the same architectural shift Block described in From Hierarchy to Intelligence. The middle layers that existed to coordinate information start being replaced by AI doing the coordination directly. In media buying specifically, the layers were a media planner, a media buyer, an analyst, and a strategist all passing information up and down a chain. The agentic model collapses those layers and puts a smaller number of accountable operators at the top of a much more direct system.
Why most agencies cannot ship this
The cost structure is the answer.
A traditional performance agency makes money on hours plus a percentage of media spend. The hours pay the team. The percentage of spend pays for the platform expertise. If agentic media buying cuts the hours needed to manage the account by 60%, the agency cuts its own revenue by 60%. So the rational thing for most agencies to do is to keep selling AI as a productivity layer that doesn't threaten the hours line — which is exactly what most agencies are doing.
Common Thread Collective's Taylor Holiday says it directly: "AI context layers are making our operators move 10x faster." That is the augmentation framing — humans still at the center, AI making them more productive. It is the right framing for a 100-person PE-backed agency with a billable headcount to protect. It is not the architectural shift.
Tinuiti, the largest US independent agency at 1,200 people, has built the Bliss Point measurement suite — patented Rapid Media Mix Modeling, incrementality, in-store sales lift — and the AI SEO offering that breaks an offering into five named components. These are real artifacts. They are not the architectural shift either. They are sophisticated tooling layered on top of a 1,200-person services org that is structurally not going to rebuild itself around agents.
This is not a criticism of those agencies. They are running their model well. It is an observation about which architecture you are buying when you sign with them.
What "AI-native paid media" actually looks like
Three things change when an agency rebuilds the paid media function around agents instead of layering AI on top of the existing team.
The agents see the full signal. Bid, creative, audience, conversion event, downstream LTV — all queryable in one place by a system that can act on what it sees. Not a dashboard a human looks at twice a week. A live model that runs the routine optimization between sprints.
The operator's job changes shape. Less time pulling reports, reformatting them for meetings, and pushing manual bid adjustments. More time deciding which segments to build, which offers to test, which customer behaviors to actually go after. The judgment work — the part that matters and the part AI cannot do — gets the operator's full attention because the routine work is no longer eating it.
The pricing decouples from hours. Once the agent system is running the routine optimization, the agency's cost structure stops being "humans times hours times rate." The pricing can be tied to outcomes the work actually produces — efficiency targets, contribution margin, return on ad spend at a specific scale. The shift is gradual. It happens engagement by engagement as the outcomes become measurable. But it is the only pricing model that survives long-term in this category.
The test for your next agency conversation
The next time an agency pitches you on AI in paid media, ask three questions.
What is your AI actually doing inside the account? If the answer is "generating creative variations" or "automating reporting," you are looking at the small play. Useful. Not architectural.
What decisions does your AI make without a human in the loop? If the answer is "none" or "it just suggests changes for our team to approve," you are looking at augmentation, not agentic. The team is still doing the routine work; AI is just helping them do it faster.
How is the engagement priced? If the answer is "monthly retainer plus percentage of media spend," the cost structure is built on the old model. That is not disqualifying — but it is the signal that the agency is selling you the productivity tweak, not the architectural shift.
The architectural shift exists. The published proof points exist. Ad copy is the distraction. Agentic media buying is the real AI play in paid, and the agencies that are going to win this decade are the ones rebuilding the function around it.