AI agents are most likely to absorb the layers of the ad stack where work is a decision made against a written specification and repeated at volume: buying, routine deal negotiation and much of the legwork in media planning. The layers most likely to stay put are the ones whose value depends on being separate from the parties they check, chiefly verification and measurement, along with the layers that carry legal or financial accountability, such as billing. Ad serving is already machine work, so agents are more likely to configure it than replace it.

TL;DR. A useful way to judge any layer is the spec, volume, independence test: agents absorb work that follows a written spec and repeats at volume, and they do not absorb work whose value comes from independence or accountability. On that test, buying and routine negotiation move first, planning moves partly, and verification, measurement and billing stay with separate systems and people. Ad serving changes how it is configured rather than who runs it. None of this comes with a reliable timetable, so each layer below is paired with an observable signal to watch rather than a date.

How should the ad stack be divided when asking what agents absorb?

The ad stack is easiest to analyse as seven layers: planning, buying, deal negotiation, ad serving, verification, measurement and billing. "Absorb" means an agent makes the layer's routine decisions within a mandate. "Stays put" means the layer remains a separate function, run by an independent system or by people, even if agents call it.

The spec, volume, independence test asks three questions of each layer. Is the work a decision against a written specification? Does it repeat often enough that consistency matters more than judgement? Does its value depend on independence from buyer and seller? Yes, yes, no marks a candidate for absorption. Yes on independence keeps a layer separate, however automatable its tasks are.

Will AI agents take over media planning?

Partly. Planning contains two kinds of work. The first is discovery and comparison: finding inventory that fits a brief, collecting proposals and putting them side by side. That work follows a spec and repeats with every campaign. IAB Tech Lab's OpenProposal specification, part of AAMP, standardises how publishers define products and packages in RFP responses, which lets a buyer agent compare offers from different sellers without translating each one.

The second kind is choosing the objective, budget and audience strategy: judgement that stays with the people who answer for the result. The signal to watch is how briefs reach sellers: if structured, machine-readable requests replace decks and emails as the normal format, the comparison half of planning has moved.

Will AI agents take over media buying?

Buying is the layer most likely to be absorbed, because much of it already has been: a demand-side platform such as The Trade Desk prices bid requests automatically. An agent adds a longer horizon, managing one budget across many sellers and deal types within a mandate. According to IAB Tech Lab, AAMP's support for the Deals API covers programmatic guaranteed, preferred deals, private marketplaces, the open marketplace and linear GRPs, so an agent is not limited to one way of buying.

People keep the mandate and approval of spend above a threshold. The signal to watch is how trading teams are organised: when roles shift from placing buys to writing and auditing mandates, the execution layer has moved.

Will AI agents negotiate deals on their own?

For routine terms, yes, within limits. Price, volume, flight dates and a viewability commitment are terms an agent can propose, counter and accept against a mandate. Non-standard terms, such as exclusivity or unusual data use, belong with a person, because the mandate is unlikely to have anticipated them.

Completing a negotiation and getting good terms are different results. In Concourse Bench v1, a benchmark of AI buying agents negotiating with software sellers, 95 of the 110 contracts signed promised a viewability floor below 70%. Closing the deal did not mean securing a strong commitment. The signal to watch is whether mandates start to specify minimum acceptable terms, not only maximum prices, and escalate weak terms as well as high spend.

Will AI agents replace ad serving?

Unlikely, because ad serving already decides and delivers in milliseconds without people. What agents can take over is the configuration around it: trafficking a deal, assigning creative, setting frequency caps and checking that what was booked matches what is live.

AAMP includes the Agentic RTB Framework, a foundation for agent services that run inside a host platform and that the orchestrating platform can call directly. That points to a model in which ad servers and exchanges host agent services rather than being replaced by them. The signal to watch is whether serving platforms expose trafficking and set-up as agent-callable services.

Will verification and measurement be absorbed by agents?

The tasks may be automated; the functions should stay separate. Verification checks whether an impression was viewable, brand-safe and seen by a person, and its value is that neither buyer nor seller produced the answer. A buying agent that grades its own delivery is marking its own homework.

Measurement follows the same logic. An agent can optimise against measurement outputs, but if it also defines success, nobody outside the deal can check the claim. Accredited standards such as the MRC viewable impression guidelines of 2014 exist so that a number means the same thing to both sides. The signal to watch is whether verification and measurement providers offer their results as services agents can call, while the definitions and accreditation stay with independent bodies.

What happens to billing when agents negotiate the deals?

Billing stays with finance teams and contracts, because someone has to be legally answerable for what is paid. What changes is the input: invoices must be reconciled against agreed terms that may exist only in each agent's logs. The AdCP documentation states that payments, legal agreements, settlement, disputes, credits and makegoods remain account or platform integrations unless a specific AdCP contract covers them, which leaves that record to the marketplace or the contract between the parties.

The likely friction moves from "what was delivered?" to "what was agreed?". The signal to watch is whether invoices start to reference a single deal record that buyer and seller both hold.

How should buyers and sellers prepare for this split?

The practical response follows the test. For layers agents absorb, invest in the mandate: what the agent may agree, which terms are non-negotiable and when it must escalate. For layers that stay put, let agents call them without owning them. For billing, settle where the agreed terms live before the first agent-negotiated deal runs.

Alkimi takes that last point as its starting place. It is a neutral agentic advertising marketplace working within IAB Tech Lab's AAMP framework, and the terms buyer and seller agents agree are written to a DealSheet that both parties hold, with people keeping approval rights over each deal.

Frequently asked questions

Which part of the ad stack are AI agents most likely to take over first?

Buying and routine deal negotiation, because both involve decisions against a written mandate that repeat at volume. Much of buying is already automated through DSPs, so agents extend existing automation rather than starting from nothing.

Will AI agents replace ad verification companies?

The case for replacement is weak. Verification is valuable because it is independent of the buyer and seller. Agents can call verification services and act on the results, but a buying agent checking its own delivery is not independent.

Who stays accountable when agents handle buying and negotiation?

The organisation that deployed the agent. An agent is not a legal entity, so the mandate, thresholds and record of agreed terms remain the principal's responsibility.

How can a business tell whether a layer is shifting to agents?

Watch observable signals rather than forecasts: structured briefs replacing decks, trading roles moving to mandate writing, and invoices referencing a shared deal record.

Which Layers of the Ad Stack Will AI Agents Absorb, and Which Stay Put?