The media agency's value proposition was built on three things: access to inventory at scale, relationships with media owners, and the expertise to translate client briefs into buying decisions. Two of those three are currently being automated. The third is changing. As agentic advertising shifts the execution layer from human traders to software agents, the agency faces a transition that is less about whether it survives and more about what it rebuilds itself around.
TL;DR. Agents can execute buying decisions faster and at greater scale than human traders. What they cannot do is define what the campaign is for, configure the constraints that bound their own behaviour, approve material commitments on behalf of a client, or take accountability for outcomes. The agency's role is not eliminated by agentic advertising. It is relocated. The trading desk becomes a configuration and governance layer rather than an execution layer. The account team becomes the primary point of accountability for agent mandates. The value the agency creates shifts from "we ran the campaign" to "we defined what the agent was authorised to do and we can prove it did that."
What parts of the agency's current role does an agent replace?
An agent can execute what a human trader currently executes: submitting bids, negotiating deal terms, managing frequency, optimising toward a performance target within defined constraints. Those activities run faster, at greater scale, and without the cognitive load limitations a human brings to a high-volume trading environment.
The IAB Tech Lab's agentic advertising framework requires human approval for any commitment above a defined spend threshold. Below that threshold, the agent acts on its mandate without referral. This design reflects the industry's current view: routine execution is appropriate for agents, material commitments require human sign-off. The boundary between routine and material is the agency's job to define.
That framing already describes the transformation. The agency is not the one executing below the threshold. It is the one who set the threshold.
What does the agency do that the agent cannot?
The agent operates on a mandate: here is what you may agree to, here are the constraints, here is the objective. The mandate has to come from somewhere, and it has to be approved by someone who carries accountability for it. That is the agency.
A client brief cannot go directly from a human marketing director to an agent without a translation step. The brief describes outcomes: reach a specific audience, build awareness of a product, drive a purchase action. The agent needs instructions: these deal types, these CPM ranges, these brand-safety conditions, these frequency caps, an approval threshold at this spend level. The work of translating the first into the second is a distinct competence, and it is now one of the agency's primary functions.
The agency also carries the accountability the agent cannot. When something goes wrong, the client's contract is with the agency, not with the agent. The IAB Tech Lab's requirement for human approval on material spend exists precisely because accountability has to sit somewhere. The agency is where it sits.
How does the measurement and reporting function change?
In a human-executed campaign, the agency's reporting role was to interpret data and make recommendations. In an agentic campaign, the agent generates data continuously: every deal negotiated, every term agreed, every impression delivered. The agency's role shifts from periodic interpretation to ongoing governance of what the agent is logging and whether the log matches what the client authorised.
The shared deal record architecture that both the IAB Tech Lab and AdCP are building toward is not just a technical specification. It is the audit infrastructure the agency needs to demonstrate that its agents acted within mandate. An agency that cannot show a client a deterministic record of every deal its agent agreed, compared against the mandate the client approved, is carrying liability it cannot quantify.
What happens to agency relationships with media owners?
The programmatic era did not eliminate publisher relationships. It changed their function. Publishers with strong direct relationships could negotiate programmatic guaranteed deals, floor prices, and preferred access. Agencies maintained those relationships because they affected the terms their buying platforms could access.
In agentic advertising, the same logic applies with a different mechanism. A publisher's agent and a buyer's agent can negotiate without human involvement. But the parameters of what each agent is authorised to offer or accept are set by the humans on each side. An agency with strong publisher relationships can establish preferred deal frameworks that its agents operate within. The relationship value does not disappear; it moves upstream into the configuration of what the agent negotiates.
What does the transformed agency look like?
It looks like a business that has two distinct functions it did not previously separate. The first is mandate design: defining what agents are authorised to do, on what terms, for which clients, at what approval thresholds. The second is governance and accountability: verifying that agents acted within mandate, presenting that evidence to clients, and carrying the accountability for outcomes.
The execution layer runs on agents. The accountability layer remains human, because accountability cannot be delegated to a system that has no legal standing to carry it. The agency that understands this distinction early builds the governance infrastructure while volumes are low and the tooling is being standardised. The agency that waits until agentic volumes are large enough to demand accountability will find the infrastructure has to be built under pressure.
This article references the IAB Tech Lab's published agentic advertising specifications on human approval thresholds and mandate governance, and the industry architecture context established by the AdCP and AAMP standards bodies.