Agentic advertising means using AI agents that plan, buy, sell, negotiate or manage advertising on behalf of a business, taking real actions within a mandate a person has set instead of waiting for a person to approve each step.

TL;DR. Agentic advertising is defined by who acts: an AI agent makes and carries out decisions, such as committing budget or agreeing deal terms, inside limits a human wrote. It differs from programmatic advertising, which executes rules a person set in advance, and from AI-assisted advertising, where software recommends and a person acts. Its working parts are buying agents, selling agents, the mandates that bound them, protocols such as IAB Tech Lab's AAMP and AgenticAdvertising.org's AdCP, and a deal record both sides can check.

What is the full definition of agentic advertising?

Agentic advertising is advertising in which AI agents take consequential actions, such as buying media, negotiating terms or moving budget, on behalf of an advertiser, agency or publisher. The agent works inside a mandate: a written set of objectives, limits and approval thresholds that a person defines before the agent runs. The organisation behind the agent stays accountable for what it commits to, which is why a governed agentic system keeps a record of what was agreed, by which agent and on whose authority.

A practical test checks for four properties, each necessary.

Model-driven decisions. An AI model interprets context and chooses what to do, rather than a rule engine matching fixed conditions.

Consequential actions. The system does something that creates a cost, an obligation or a commitment, such as signing a deal or reallocating spend; a report or recommendation does not count.

Named human approval points. Before the agent runs, its owner defines where a person can review, redirect or halt its actions, typically above a value or volume threshold.

A record of what was committed and why. Without an auditable account of an agent's commitments, nobody can be held accountable for them.

A worked example makes the four properties concrete. In the marketplace run by Alkimi, buyer and seller agents negotiate within mandates their owners set, humans keep approval rights at the deal level, and the agreed terms are written to a DealSheet. That record is written at agreement, timestamped, held by both parties and retrievable by either side independently.

What does "agentic" mean in AI?

In AI, "agentic" describes a system that pursues a goal by taking actions, rather than producing a single answer to a prompt. It receives an objective, works out the steps, takes them, observes the results and adjusts, without a person specifying each step. An assistant advises and waits for a person to act, while an agent acts and then checks what its action changed.

Why do people define agentic advertising differently?

The term is used in two overlapping senses, which explains why published definitions disagree. In the first, an agent operates one advertiser's campaigns across the paid media cycle, adjusting audiences, creative, bids and budget against a goal. In the second, agents trade with each other: a buyer's agent and a seller's agent discover inventory, negotiate terms and book deals.

Asked by The Current to define the term, Anthony Katsur, CEO of IAB Tech Lab, said: "Agentic is just systems executing and interacting across advertising workflows. That only works if it's structured. Execution, protocols and trust all have to be defined together."

In the same piece, Bill McLaughlin of PubMatic described the change as moving from "AI as a tool" to "AI as an operator", with software agents acting on behalf of buyers and sellers in real time.

The two senses share one test: an agent makes consequential decisions within a mandate, and somebody needs a reliable record of what it did. Agent-to-agent trading adds a second problem, because two parties with separate agents and separate records must agree on what was settled.

How is agentic advertising different from programmatic advertising?

Programmatic advertising executes human-defined rules at machine speed: a trader sets bid prices, audience segments, frequency caps and blocklists, and the system applies them across auctions. Software buying media is therefore not what separates the two.

The first real difference is decision authority. In programmatic buying a person writes the logic and the system executes it, while in agentic buying the person sets objectives and limits and the agent decides how to meet them.

The second difference is negotiation. An open auction is take-it-or-leave-it, whereas an agentic buyer can counter on placement, format, measurement or flight dates. Programmatic brought autonomy of action; agentic advertising adds autonomy of agreement.

Agentic advertising does not replace the programmatic stack. IAB Tech Lab's AAMP builds on the standards that already power the industry, and its 2.0 release added programmatic guaranteed, preferred deals, private marketplaces and the open marketplace through the Deals API. What changes most is the deal layer: how buyers and sellers find each other, agree terms and record obligations.

How is agentic advertising different from "AI in advertising"?

"AI in advertising" is the broad category: predictive bidding, lookalike audiences, generative creative, forecasting and chat assistants that answer questions about a campaign. Agentic advertising is the narrower part of that category in which the AI takes the action itself.

An AI feature optimises one step inside a platform, an assistant recommends and drafts, and an agent decides and commits within its mandate. What makes a system agentic is the authority it is given, not the model inside it.

What is agentic advertising not?

Three kinds of system are often labelled agentic and are not.

Rule-based automation is not agentic. A bidder that raises bids when viewability passes a threshold, or caps spend per domain, is executing instructions that a person wrote and can read in full.

AI-assisted buying is not agentic buying. If a person reviews the AI's output and then commits manually, the person took the consequential action and the AI was an analysis tool.

Autonomy with no approval rights is not governed agentic advertising. A system that commits budget with no defined human review points has a governance gap. The organisation that deployed it still carries the accountability for what it agrees.

What are the key components of agentic advertising?

Buying agents. A buying agent acts for an advertiser or agency: it reads a brief, finds inventory, negotiates and books deals within its mandate. IAB Tech Lab publishes an open-source reference Buyer Agent SDK as part of AAMP on GitHub.

Selling agents. A selling agent acts for a publisher or media owner: it presents and prices inventory, answers buyer requests and accepts or counters offers within the seller's rules. IAB Tech Lab publishes a matching Seller Agent SDK, and its AAMP page states that "the Agentic ecosystem will start with a Buyer Agent and Seller Agent talking to each other."

Mandates. A mandate is the written instruction that bounds an agent: budget, price ceilings, inventory, audiences, brand safety rules and the thresholds that need a named person's sign-off. A well-written mandate is the main control a business has over what its agent may do.

Harnesses. The harness is the software around the model that supplies the brief, checks each proposed action against the mandate, routes decisions above a threshold to a person and logs what was committed. Harness design matters as much as model choice.

Protocols. Protocols give agents built by different companies a shared language. AAMP, IAB Tech Lab's Agentic Advertising Management Protocols, is an open-source umbrella initiative whose components include agentic-direct for agent communication and orchestration, the buyer and seller agent SDKs, and OpenProposal for the brief and RFP stage. AdCP, the Ad Context Protocol published by AgenticAdvertising.org, is an open protocol that lets agents discover, plan, buy, sell and measure media, with tasks exposed over MCP or A2A according to its documentation.

Identity. Before agents transact, each side needs to know who or what it is dealing with. The IAB Tech Lab Agent Registry, part of AAMP, lets companies list their MCP and A2A agents with categories and capabilities.

Deal records. A deal record captures what the agents agreed, when and on what terms. Separate records drift: in the published Agree. Transact. Verify. simulation of 90,202 deals, separate records disagreed on at least one of eight deal dimensions in 95.3% of transactions, against 0.19% with a shared record. A record both parties hold, plus approval gates both can confirm were applied, is what makes an agent's commitments checkable.

What are examples of agentic advertising?

An agent-negotiated direct deal. In a demonstration published on IAB Tech Lab's AAMP page, COO Shailley Singh used the Tech Lab reference SDKs to create a media plan from a brief, negotiate with the seller, confirm the transaction and push it to Google Ad Manager. The demo used existing standards including OpenDirect and AdCOM.

An agent-to-agent buy with real money. AgenticAdvertising.org reports that on 16 October 2025 what it describes as the first agent-to-agent media buy was executed, with real money and real inventory from LG Ads. AI agents handled creative, targeting and approval, with human oversight.

A sell-side agent answering a brief. AAMP 3.0 extends the framework to the RFP stage through OpenProposal, a draft specification of what is being bought and sold, built so a buying agent can parse and evaluate it at machine speed.

A buy-side agent running a campaign (hypothetical). A hypothetical retailer gives an agent a quarterly budget, a target cost per acquisition and a rule that large shifts of spend between channels need a named trader's approval. The agent moves budget within that limit and queues larger moves for sign-off with its reasoning attached.

How much autonomy should an advertising agent have?

Autonomy is best treated as earned rather than assigned, because the cost of an agent's errors scales with its authority. The five-stage earned autonomy model sequences that authority: observe, recommend, draft, human-approved action and bounded automatic action.

Movement from one stage to the next depends on evidence from the stage before and on explicit authorisation by a person. AgenticAdvertising.org sets out a related principle, which it calls Embedded Human Judgment: humans stay in the loop on decisions with real consequences.

How can a buyer tell whether a product is genuinely agentic?

Three questions, which can be called the Deal Record Test, separate substance from labelling.

Who owns the record of the deal, and can both parties read it independently? If the record lives inside one platform and the counterparty only receives a report, it is one party's account of events.

Where are human approval rights enforced, and which party controls that enforcement? "Human in the loop" can mean a gate neither agent can bypass or an override that one platform controls alone.

What happens when the two agents hold different accounts of what was agreed? In agent-to-agent trading disagreement is the normal case, so the system needs a way to resolve it that does not depend on teams reconciling logs by hand.

A product that cannot answer all three may still be useful AI-assisted advertising, but it is not agentic in the structural sense.

Which related terms should you know?

Each term below has a fuller entry in the agentic advertising glossary.

AI agent. Software driven by an AI model that pursues a goal by deciding on actions, taking them and observing the result.

Agentic marketplace. A venue where buyer and seller agents find each other, negotiate and transact, with identity checks, approval gates and a shared deal record.

Agent2Agent (A2A). An open standard for communication between AI agents built on different frameworks by different vendors.

Model Context Protocol (MCP). An open-source standard for connecting AI applications to external data sources, tools and workflows.

OpenProposal. The core object of AAMP 3.0, a draft specification of what is being bought and sold, whose public comment period ran from 22 September to 22 October 2026.

Agentic Real Time Framework (ARTF). An IAB Tech Lab specification for agent services that run as containers inside a host platform, for uses such as deal activation or pre-impression fraud detection.

Agentic Audiences. An IAB Tech Lab open standard, formerly the User Context Protocol and donated by LiveRamp, for how agents exchange identity, contextual and reinforcement signals as embeddings.

Mandate. The written instruction that defines what an agent may do and when it must ask a person.

DealSheet. A deal record written at agreement, timestamped, held by both parties and retrievable by either side independently.

What does agentic advertising mean for buyers and sellers now?

In practice, agentic advertising moves human judgement from approving each transaction to writing mandates, setting approval thresholds and checking records. Teams that test agents before live spend and insist on a deal record both sides hold can tell a working agent from a relabelled dashboard.

Alkimi, the platform in the worked example above, builds for that shift with three products: Concourse, a live benchmark for AI media-buying agents; a Gym where buyers and sellers build their own agents, set their rules and spar against counterparty agents before any real money moves; and a neutral Marketplace that works within IAB Tech Lab's AAMP framework today and is being built to be AdCP compatible. It filed comments and change requests during the OpenProposal v3.0 public comment period, and the simulation cited above is its own research.

Frequently asked questions

What is agentic advertising in simple terms?

Agentic advertising is advertising in which AI agents buy, sell or manage media for a business, within limits a person has set. The person sets the goal and the rules, and the agent decides how to meet them.

Is agentic advertising the same as programmatic advertising?

No, programmatic advertising executes rules a person wrote in advance at machine speed. Agentic advertising lets an agent choose its actions and negotiate terms within a mandate, usually on top of the same programmatic systems.

Who is accountable when an advertising agent makes a bad deal?

The organisation that deployed the agent remains accountable, because an agent is not a legal entity. That is why mandates, approval thresholds and a reliable deal record matter.

What is the difference between AAMP and AdCP?

AAMP is IAB Tech Lab's umbrella initiative of open-source agentic advertising protocols, built on existing industry standards. AdCP is a separate open protocol published and governed by AgenticAdvertising.org that lets agents discover, plan, buy, sell and measure media.

Does agentic advertising only apply to advertisers?

No, publishers and media owners use selling agents to price and package inventory, answer briefs and respond to offers. When both sides use agents, the deal itself becomes agent-to-agent.

What Does Agentic Advertising Mean? Definition, Components and Examples