Buyer agents and seller agents are AI software that act for the two sides of a media deal. The buyer agent pursues an advertiser's goals; the seller agent represents a media owner's inventory and terms. When they work together, the slow middle of a transaction (finding inventory, checking fit and setting up a deal) could move from email threads to structured exchanges, while activation stays on the buying platforms teams already use.
This guide is for people who build or evaluate these systems. It focuses on roles, responsibilities and the shape of the workflow, not on any specific implementation.
Who does what in an agentic media transaction?
A useful way to picture it is two delegates at a table. Each has a brief from its principal and a set of rules it cannot break.
| Buyer agent | Seller agent | |
|---|---|---|
| Acts for | Advertiser, agency or trading desk | Media owner and its representation partner |
| Starts from | A campaign brief: audience, context, budget, formats, timing | A catalogue of represented inventory and approved terms |
| Main job | Find and compare options, then request the best fit | Describe inventory accurately, match requests, create deals within policy |
| Must protect | The buyer's budget and brand rules | The seller's pricing, inventory rules and access policy |
| Output | A request, a comparison, a decision | Matching inventory, terms and a deal reference |
A buyer agent is only as good as the options it can see. A seller agent is only as useful as the inventory data and policy behind it. Neither is a free-roaming negotiator. Both act inside limits their principals set.
What does the workflow look like, step by step?
Most descriptions of agent-to-agent advertising break into five stages.
1. Brief
The buyer agent receives goals from a person or system: target geographies, devices, content categories, formats, sizes, budget and deal preferences.
2. Discovery
The buyer agent asks one or more seller agents what is available. The seller agent answers from its catalogue, filtered by the request. This is where well-structured inventory data pays off, a topic we cover from the publisher side in inventory discovery and publisher representation.
3. Evaluation and matching
The buyer agent compares options against the brief. The seller agent may help by narrowing results or confirming availability and pricing.
4. Deal creation
When the buyer agent picks an option, it asks the seller agent to create a deal. The seller agent creates it within the seller's rules and returns a Deal ID, the reference that ties bids to the agreed terms.
5. Activation and delivery
The Deal ID is used in a supported buying platform. From here, delivery, pacing, measurement and billing follow the existing programmatic path.
The important design point is the handoff at step five. Agents do not need to reinvent the auction or the ad server. They connect to the parts of the transaction that have historically been manual.
Which standards make agents interoperable?
Without shared standards, every buyer agent would need a custom connection to every seller. Two developments matter here.
Model Context Protocol (MCP)
MCP describes itself as "an open-source standard for connecting AI applications to external systems" (modelcontextprotocol.io). The specification defines three roles: hosts (the AI applications that start connections), clients (connectors inside the host) and servers (services that provide context and capabilities). Servers can offer resources, prompts and tools, where tools are functions for the AI model to execute (MCP specification, 2025-06-18).
For media, a seller agent built as an MCP-compatible service can describe what it offers in a way a compatible buying agent can read and call. The specification is also firm on control: hosts must obtain explicit user consent before invoking any tool, and tool descriptions should be treated as untrusted unless they come from a trusted server (MCP specification). That principle maps neatly onto advertising, where both sides need clear authority before money is committed.
IAB Tech Lab's agentic work
IAB Tech Lab, the industry standards body behind OpenRTB, published an agentic roadmap on January 6, 2026. It describes a path to "secure, interoperable agentic execution across digital advertising" and talks about integrating existing standards such as OpenRTB, AdCOM, OpenDirect, VAST and the Deal API with modern protocols including Model Context Protocol, Agent2Agent and gRPC (IAB Tech Lab). It also lists open-source reference implementations of buyer and seller agents among its plans.
On September 22, 2026, IAB Tech Lab released AAMP 3.0 (Agentic Advertising Management Protocols) for public comment, with a new OpenProposal specification. The release describes "a common way for buyer and seller agents to exchange and evaluate proposals", moving from an RFP to a buy commitment (IAB Tech Lab). Public comment runs until October 22, 2026, so treat it as a specification in progress.
Analysis
The direction of travel is clear: standards bodies are not trying to replace programmatic plumbing, they are wrapping it so agents can use it. For developers, that suggests building agents that speak open protocols at the edges and hand off to established activation paths, rather than inventing private formats that will need rewriting later.
What should builders design for?
If you are building or evaluating a buyer agent, a few principles hold regardless of which standard wins.
- Scope. Give the agent explicit limits on budget, inventory types and deal terms, and keep a human approval step for commitments until you trust the behaviour.
- Provenance. Know which seller agents your agent is talking to and whether they are authorized to represent the inventory they describe. Supply chain files such as ads.txt and sellers.json remain the reference for authorized selling.
- Comparability. Ask for inventory in consistent fields (category, geography, device, format, size, pricing, deal type) so options can be compared fairly.
- Traceability. Keep a record of what was asked, what was offered and which Deal ID resulted, so finance and campaign teams can reconcile later.
Example (illustrative)
An agency's buyer agent receives a brief for a travel campaign: mobile app and CTV, three European markets, video formats, a fixed budget and a preference for private deals. It queries two authorized seller agents. One returns in-app rewarded and interstitial video; the other returns CTV in-stream inventory. The buyer agent ranks them against the brief, a planner approves the shortlist, and the buyer agent asks each seller agent to create a deal. Two Deal IDs come back and the trading team activates them in the agency's buying platform.
What exists today, and what is still being built?
It is easy to read about agentic advertising and assume everything is live. The honest picture is mixed, and that applies to IncrementX too.
Available today: IncrementX Seller Agent is an MCP-compatible interface. Authorized buying agents can discover IncrementX inventory by site or app, IAB category, geography, device and environment, ad format, ad size, inventory type, pricing, deal type and availability. They can evaluate and match it, initiate deal creation and receive Deal IDs used to activate through supported buying platforms. Access is by request.
Being built: wider agent-to-agent buying workflows, where buyer agents and seller agents coordinate across more of the transaction, are being developed with partners. We describe that advertiser-side experience on our Agentic Buying page, and it should be read as future-facing rather than a finished integration.
That split is deliberate. IncrementX is a media representation partner, so the inventory a seller agent exposes is inventory our team has understood, packaged and approved. The agent is designed to carry that representation work to a new kind of buyer, not to replace it.
Where this could go next
Over time, buyer and seller agents could handle more of the proposal stage, compare packages across many sellers and adjust requests as campaigns learn. Standards such as those in IAB Tech Lab's roadmap are designed to make that interoperable. What is unlikely to change is the need for trusted inventory data, clear authority on both sides and a reliable handoff to delivery. For a publisher-side view of the same shift, read what a seller agent is and our look at the future of media representation.