TL;DR: An agent mandate is a formal set of instructions, permissions, and constraints that defines what an AI advertising agent is authorised to do on behalf of a buyer or seller.

The Basics

As AI agents take on more active roles in programmatic advertising, the question of authority becomes critical. Who decides what the agent can do? What limits apply? What happens if the agent encounters a situation outside its instructions?

An agent mandate answers all of these questions. It is the governing document for an agent's behaviour. Before an agent negotiates, bids, or commits spend, it operates within the boundaries set by its mandate.

What a Mandate Contains

A well-formed agent mandate typically specifies four things.

Scope of authority. The mandate defines what actions the agent may take independently versus what requires human approval. A buying agent might be authorised to accept deals below a certain CPM threshold automatically, but required to escalate anything above it.

Budget parameters. The mandate sets daily, weekly, or campaign-level spending limits. It may also specify minimum and maximum unit prices, frequency caps, and audience targeting constraints.

Channel and inventory rules. Mandates often include approved publisher lists, blocked categories, minimum viewability standards, and brand safety requirements. These rules travel with the agent across every negotiation.

Escalation triggers. The mandate defines the conditions under which the agent stops and asks for human input. This is central to responsible agentic advertising. Agents that operate without escalation triggers create risk; agents that escalate on every decision create no value.

Why Mandates Matter Now

Traditional programmatic advertising relies on campaigns, line items, and bid rules set up inside a DSP. A human configures the system; the system executes according to its configuration. The human is not really making real-time decisions. The DSP is.

Agentic advertising changes the dynamic. An AI agent can reason about context, interpret new information, and propose or accept deals in real time. That capability is valuable, but it requires a more explicit governance structure than a DSP campaign ever did.

A mandate is that structure. It is not a technical parameter buried in a config file. It is a human decision, documented and auditable, about what the agent is permitted to do.

In marketplace environments like Alkimi, the mandate also interacts with a shared deal record. The DealSheet that both buyer and seller hold reflects what was agreed, within the limits each party's agent was authorised to accept. The mandate is the upstream document that made the deal possible.

The Earned Autonomy Model

The most effective approach to agent mandates follows what practitioners call the earned autonomy model. Agents begin in an observation mode, surfacing recommendations for human review. As track records build and teams develop confidence in agent behaviour, mandates are progressively extended to cover more autonomous actions.

This is not a binary shift from human control to machine control. It is a deliberate, staged expansion of agent authority grounded in demonstrated performance. Mandates reflect where an organisation is on that journey.

The stages look like this. First, the agent observes and reports. Second, it makes recommendations but takes no action. Third, it drafts proposals for human sign-off. Fourth, it executes human-approved actions. Fifth, within bounded parameters, it acts automatically. Each stage requires a mandate update that reflects the new level of trust.

Mandate Portability

One practical consideration is whether mandates are portable across systems. If an agent operates across multiple marketplaces or exchanges, its mandate should travel with it. Hardcoding mandate logic inside a single platform creates fragmentation and makes auditing difficult.

Standards bodies are working on this. The Agent Communication Protocol (AdCP), published by AgenticAdvertising.org, addresses how agents identify themselves and communicate constraints during negotiations. A portable mandate representation is a logical extension of that work.

What Buyers Should Do Now

For US buyers beginning to think about agentic advertising, the mandate is the right starting point. Before configuring an agent, define what it is allowed to do. Involve legal, compliance, and finance teams in that definition. Document the scope of authority clearly.

Then start narrow. Grant the agent limited authority. Observe its behaviour. Expand the mandate as confidence grows. This approach reduces risk and builds the internal knowledge needed to manage agents responsibly over time.

The mandate is not a technical detail. It is the foundation of accountable agentic advertising.