TL;DR: Brand-side deployment of AI agents in advertising is at an early and bounded stage. Brands are starting with narrow, measurable use cases: retargeting and lower-funnel performance buying, where the mandate scope is limited and the measurement signal confirms whether the agent is within intent. Deal negotiation with direct publishers is largely aspirational for most brand-direct buyers because the required infrastructure is not yet widely available on the sell side. What early deployers consistently find is that mandate design is harder than anticipated: agents operate on the literal terms of the instruction, not the intent behind it. The WPP Research simulation of 90,202 agent-to-agent transactions found that deal records diverged in 95.3% of cases under separate record-keeping; that figure dropped to 0.19% when both agents referenced a shared record.
Brand-side deployment of AI agents in advertising is proceeding cautiously and with intent, rather than at the pace that conference coverage suggests. The most useful data is not projected future volume but what brands are learning in current deployments: about what agents do reliably, about where human oversight remains essential, and about the specific competency that agentic advertising requires from brand marketing and procurement teams.
What use cases are brands starting with?
The earliest brand-side agent deployments are concentrated in retargeting and lower-funnel performance buying. These use cases share two characteristics that make them suitable for initial deployment: the mandate scope is narrow and the measurement signal is immediate.
In retargeting, the agent operates against a defined first-party audience, users who have visited the brand's properties, abandoned a checkout, or engaged with previous advertising, and a clear objective: drive conversion or return visit. The creative asset is pre-approved. The placement constraints are set. The agent's task is to find the right moment and price to place the advertisement, not to make judgements about audience suitability or contextual fit that require the brand's intent to be interpreted correctly.
In lower-funnel performance buying more broadly (app installs, lead generation, direct response), the measurement signal is fast and attributable. If the agent is performing within intent, conversion data confirms it quickly. If the agent has found a gap in the mandate, conversion data shows the anomaly. The feedback loop is short enough that errors are detectable before they compound.
Prospecting, upper-funnel brand campaigns, and direct publisher deal negotiation are less common in early brand deployments. These use cases require the agent to operate in broader mandate territory, where the measurement signal is slower and the consequences of a mandate gap are harder to detect before they become material.
What is the current scale of brand-side agentic deployment?
Volume data for brand-side agentic advertising is limited, but the clearest available signal comes from publisher infrastructure. Andrew Mole of pubX told ADOTAT in August 2026 that live agentic media spend was running at approximately $3,000 a day from one named operator. That figure represents live commercial deployment as of the date of publication: present, but narrow.
The more instructive data is from simulation. WPP Research modelled 90,202 agent-to-agent transactions to assess what happens to deal records when buy-side and sell-side agents operate under separate record-keeping versus a shared record. Under separate books, the two parties' deal records diverged in 95.3% of transactions. Under a shared record that both agents referenced, the divergence rate fell to 0.19%.
That simulation result matters for brands deploying agents at any scale because deal record divergence is the source of measurement reconciliation failures: the point at which a brand's verification tooling attempts to confirm delivery against a deal record that disagrees with the publisher's version of the same transaction. Brands building agent buying programmes need to treat shared record access as a platform selection criterion, not a feature to evaluate later.
What are early deployers learning about mandate design?
The consistent finding from early brand-side deployments is that mandate design is more demanding than it appeared before deployment. Brands entering agentic buying with the expectation that a general campaign brief translates naturally into a working agent mandate find that the translation requires explicit specificity that campaign management did not.
An agent does not interpret a campaign brief. It operates on the literal content of the mandate it was given. A mandate that specifies a CPM cap without also specifying a daily spend limit produces an agent that may submit multiple sub-threshold deals in a single day, with the aggregate commitment exceeding the brand's intent. A mandate that specifies brand-safe content categories without also specifying format constraints produces an agent that may win technically brand-safe placements in formats the brand did not intend to run.
These are not agent failures. They are mandate gaps: places where the literal instruction was consistent with the agent's behaviour but inconsistent with the brand's intent. The discipline of mandate design is the discipline of closing those gaps before deployment, not after.
The IAB Tech Lab's agentic advertising specifications require that mandates include explicit approval thresholds, defined data authorisations, and clear scope on what the agent may and may not commit to. Brands using those specifications as a design checklist before initial deployment report fewer mandate-gap incidents than brands writing mandates from first principles.
What does the infrastructure gap mean for brand-direct agentic buying?
Direct publisher deal negotiation, where a brand's agent negotiates deal terms directly with a publisher's agent, requires sell-side infrastructure that is not yet widely available. The publisher needs to have deployed a sell-side agent with the capability to negotiate deal terms in a machine-readable format, and both parties need access to a shared record infrastructure that holds the authoritative version of what was agreed.
The protocols enabling this are being standardised. The Ad Context Protocol, being developed by Brian O'Kelley and colleagues, and the Agent Advertising Marketplace Protocol (AAMP), developed through the IAB Tech Lab, both address the machine-readable deal negotiation layer. Neither is yet widely deployed on the sell side. Brands designing agentic buying strategies now are doing so against an expectation that this infrastructure will be available for mainstream use within a few years, with the precise timeline dependent on publisher adoption.
What brands can do now is build internal capability in the areas that will transfer directly to direct-deal deployments: mandate design, threshold governance, and data authorisation frameworks. These competencies apply equally to performance buying through existing DSP infrastructure and to direct agent-to-agent deal negotiation. Building them at low volume and low risk creates the institutional readiness that larger-scale deployments will require.
*This article references a WPP Research simulation of 90,202 agent-to-agent transactions, with findings on deal record divergence rates under separate versus shared record-keeping; volume data from Andrew Mole of pubX, published in an ADOTAT investigation in August 2026; the IAB Tech Lab's published agentic advertising working group specifications; and information about the Ad Context Protocol from Brian O'Kelley and the Agent Advertising Marketplace Protocol from the IAB Tech Lab.*