TL;DR: Brands are using AI agents in advertising in a progression from supervised automation to increasingly autonomous deal execution. The starting point for most brands is AI-assisted campaign optimisation through their agency or DSP. The more advanced use is deploying in-house or agency-managed buy-side agents that negotiate bilateral deals with publisher agents under a mandate the brand has approved. The governance model in both cases is earned autonomy: agents operate within defined parameters, and humans retain approval authority over decisions outside those parameters.

Brand involvement in AI agent use for advertising sits across two distinct tracks. The first is brands adopting AI optimisation tools provided by their agencies or platforms, usually without direct brand engagement in how the AI is configured. The second, less common but growing, is brands taking an active role in defining the mandate under which an agent operates, and retaining approval authority over deal-level decisions that fall outside the mandate parameters.

Both represent real use of AI agents in advertising. The second represents a meaningfully different accountability model.

AI-assisted optimisation: what most brands are doing

For the majority of brands, AI is present in programmatic advertising as an optimisation layer within the agency's or platform's existing workflow. The brand sets campaign objectives, budgets, and audience requirements. The agency configures these in a DSP. The DSP's AI optimisation tools adjust bids, targeting, and pacing to achieve the objectives more efficiently than manual management would.

The brand's experience of this is often invisible: the campaign performs, the reporting shows good outcomes against KPIs, and the AI's role is not surfaced unless the agency specifically discusses it. Many brands are effectively using AI agents in their advertising without being aware of the specific AI decisions being made on their behalf.

This is not a governance problem in itself. It becomes one when the brand's accountability framework does not match the level of AI autonomy their agency or platform is actually exercising. If the agency's AI tool is making campaign decisions that the brand would want to review, and those decisions are happening automatically without any human oversight, that is a gap between the brand's assumed accountability and the actual decision chain.

Mandate-governed agent deployment: the more advanced model

A smaller number of brands are engaging directly with the mandate framework that governs their buy-side agent. In this model, the brand participates in writing or approving the mandate: the document that defines what the agent is authorised to agree, at what CPM ranges, within what inventory scope, with what audience data conditions, and with what financial thresholds for human approval.

This model makes the brand an active participant in the agent's governance, not just a recipient of campaign reporting. The brand approves the mandate before the agent is deployed. The brand sets the approval thresholds for deals above a defined value or outside a defined scope. The brand reviews the deal record log periodically to confirm that the agent's decisions are consistent with campaign strategy.

This level of engagement requires a brand-side understanding of what mandates are and how they work. It is not yet the norm. It is where governance-focused brands are moving as the industry matures.

What the earned autonomy model means for brands

The governing principle for AI agent use in advertising is earned autonomy: agents start with a narrow scope, demonstrate reliable decision-making within that scope, and have their autonomy expanded incrementally as the evidence supports it.

For brands, this means beginning agent deployment with close oversight and a conservative mandate: a small budget allocation, a narrow inventory scope, approval thresholds set low enough that many deals require human confirmation, and weekly review of the agent's decision log. As the agent demonstrates that it is making decisions consistent with the mandate parameters, the approval thresholds can be raised and the scope can be broadened.

A brand that deploys an agent with a broad mandate and high approval thresholds from day one is operating at a level of agent autonomy that is not yet supported by evidence of the agent's decision quality. The earned autonomy model is not about distrust of the technology; it is about building the evidence base that justifies trusting any new system, AI or otherwise, with increasing responsibility.

The brand's practical role

In both the optimisation and the mandate-governed model, the brand's practical role includes three things.

Setting objectives clearly. Agents optimise against goals; vague goals produce unexpected optimisation behaviour. A brand that defines its campaign objective precisely, in terms the agent can evaluate (cost per reach point, viewable impression percentage, inventory category requirements), gives the agent a clearer target and reduces the probability of the agent finding an unintended shortcut.

Reviewing the governance layer. Whether the brand is delegating to an agency-managed agent or running an in-house mandate, the brand should understand what the agent is authorised to do, at what thresholds human approval is required, and who is responsible for reviewing the agent's decisions post-campaign.

Demanding deal record access. For any bilateral deal negotiated by an agent on the brand's behalf, the brand should be able to access the DealSheet: the agreed terms, the parties, the CPM, the inventory scope. Deal record access is the brand's audit right over its own media spend.