We gave you a progress report on AAMP a few weeks ago, and now we are ready to unveil AAMP 2.0. This release takes us from concept to reality, as what were reference implementations and frameworks, become full-featured agent SDKs. These Buyer and Seller Agent SDKs can negotiate with each other, or share an agreed rate card, and facilitate transactions, all within the guardrails of existing standards and organization-controlled boundaries. Everything is logged and audited so humans can remain in the loop for efficient approvals and updates. AAMP 2.0 also includes a new Agentic Audiences OpenRTB extension, Prebid module, and open source scorer. This allows immediate adoption of the specification by partners in both agentic and traditional trading.
Below we’ll explain some of the key features, built by the Agentic Task Force, in the Buyer Agent and Seller Agent SDKs, which are available for you to download and contribute to in the AAMP GitHub. Companies are already using these SDKs to support their own Agentic strategies, so the time to get started is today, by building your agents and registering in the Tech Lab Agent Registry for easy discovery.
Buyer Agent SDK 2.0
The Buyer Agent SDK is made up of 3 levels of Agents: Orchestration, Channel Specialists, and Functional Agents that handle the execution tasks. With these levels working together, the Buyer Agent SDK enables a Media Planner to efficiently assemble media plans that match a campaign, audience, and goals, in moments rather than hours.
In version 1.0 of the SDK, this was limited to Direct Sold campaigns only, but in 2.0, a higher level of sophistication has been added, that includes creating and optimizing programmatic deals, and automating campaign setup and management. These new capabilities build against the standard schemas derived from Deals objects, Deals API, as well as OpenDirect, and AdCOM. This ensures that the Buyer Agent is asking Seller Agents for the right controls, and the Agents can communicate using the same digital advertising language.
Its multi-agent architecture leverages specialized roles and skills, such as portfolio managers for budget allocation, channel specialists for domain expertise (CTV, mobile, etc.), and tool agents for execution.
Some of these key new features are outlined below, and everything included can be accessed in detail in the AAMP Buyer Agent GitHub, where you can do pull requests and contribute.
Deals Library
At the core of Buyer Agent 2.0 is a centralized Deals Library that brings structure and portability to the deal lifecycle. Rather than siloing agreements within a single platform, the library allows teams to manage, track, and seamlessly transfer deals across environments. This removes a large number of error-prone steps, enabling human buyers to focus on evaluating changes to strategy, or optimizing against what is already proven to be working.
This portability is paired with a robust execution workflow—moving from quote to approval to booking—powered by a flexible state machine that can be tailored to specific business rules. With native support for APIs such as OpenDirect 2.1, and the Deals API, the Deals Library becomes more than a repository; it acts as an interoperable system of record for programmatic transactions, ensuring consistency, traceability, and operational efficiency.
Discovery and Negotiation
Buyer Agent 2.0 transforms how buyers discover and engage with sellers through a dynamic, multi-seller ecosystem. By integrating with the IAB Tech Lab Agent Registry, the platform enables automated discovery of seller agents, which can then be filtered based on trust signals, capabilities, inventory, and media offerings such as media kits and product catalogs.
Once relevant sellers are identified, the built-in negotiation engine uses any pre-negotiated rate cards that are in place for any seller or gets more information as per buyer policies. It then engages the seller agents in the buying process with the rate card locked in and pricing protection maintained. This allows buyers to optimize outcomes at scale while still maintaining strategic control. The result is a more efficient, data-driven negotiation process that reduces manual overhead without sacrificing sophistication.
Campaign Automation
Campaign Automation is where Buyer Agent 2.0 delivers end-to-end orchestration. The system begins by ingesting campaign briefs, including budget, audience definitions, KPIs, and timelines. From there, it automatically allocates budgets across channels, researches available inventory, and generates actionable recommendations.
Once a plan is approved, the agent can proceed to book deals autonomously, closing the loop from planning to execution. Throughout this process, configurable notifications keep stakeholders informed via channels like Slack, email, or database triggers. Human-in-the-loop approval gates and guardrails ensure that automation remains controlled and reliable, preventing errors and maintaining alignment with business objectives.
Seller Agent SDK 2.0
Seller Agent is designed as a standards-first, intelligent supply-side system that embeds industry protocols directly into its core operations. Rather than treating standards as external constraints, the agent natively incorporates frameworks like OpenDirect 2.1, AdCOM, and IAB taxonomies for content, audiences, and ad products into how it structures inventory, negotiates deals, and communicates with buyers. This ensures interoperability across platforms while enabling consistent, machine-readable transactions between agents.
By aligning with supply chain transparency mechanisms such as sellers.json and OpenRTB schain, Seller Agent not only enhances trust but also makes compliance and data integrity a built-in feature of every interaction. The result is a system where automation, negotiation, and delivery are all grounded in shared industry standards—unlocking scalable, efficient collaboration between buyers and sellers in a fully agentic ecosystem.
The key features in Seller Agent SDK 2.0 are outlined below and can be viewed in detail in the AAMP Seller Agent GitHub, where you can perform pull requests and contribute.
Media Kit Management
Seller Agent introduces a dynamic approach to Media Kit Management, transforming what has traditionally been static PDFs into intelligent, access-controlled storefronts. Media kits can be configured as public or authenticated, with each tier revealing progressively richer detail. Public views provide high-level inventory descriptions, standard packages, and broad pricing ranges, while authenticated buyers gain access to granular inventory data, availability signals, audience segments, and deal-ready packages with more precise pricing.
What sets this system apart is its contextual intelligence. Media kits adapt in real time based on who the buyer is, their relationship to the seller, and the stage of the deal. During active negotiations, kits become even more personalized—generating custom bundles, dynamic pricing curves, and tailored recommendations based on buyer behavior and seller strategy. This tiered and negotiation-aware model ensures that every buyer interaction is both relevant and optimized for yield.
Negotiation and Pricing Agent
At the heart of the Seller Agent is a sophisticated Negotiation and Pricing Agent designed to maximize revenue while maintaining flexibility. Built on a configurable negotiation engine, it supports both automated and human-in-the-loop interactions, enabling seamless engagement with buyer agents across multiple deal stages.
The system supports a range of negotiation strategies and leverages structured workflows—from proposal through negotiation to booking and delivery—ensuring consistency and control. Pricing is not static; instead, it evolves dynamically based on predefined tiers, buyer characteristics, pre-negotiated rate cards and real-time context such as demand, seasonality, and yield targets. This allows sellers to respond intelligently to market conditions while maintaining strategic pricing discipline, and ensuring agreed upon pricing with buyers is respected.
State Engine
The Seller Agent is powered by a robust State Engine that governs the entire lifecycle of deals and orders. This built-in state machine defines clear workflows for negotiation, booking, and delivery, ensuring that each step follows a predictable and auditable path.
Beyond deal-making, the state machine extends into order management, enabling delivery tracking, cross-order reporting, and comprehensive audit logs. This unified approach ensures continuity from initial proposal to final delivery, reducing operational fragmentation and improving transparency. Configurable approval gates allow human oversight at critical points, while an event bus provides real-time notifications across systems, keeping all stakeholders aligned.
Integrations
Seller Agent is designed to operate within a broader ecosystem, with deep integration capabilities across inventory sources, platforms, and industry standards. It can ingest inventory directly from leading ad servers such as Google Ad Manager (GAM) and FreeWheel, as well as from static data files, ensuring flexibility in how supply is onboarded.
On the distribution side, connectors enable seamless synchronization of deals with major supply-side platforms (SSPs) like PubMatic, Index Exchange, and Magnite. Its pluggable architecture allows for easy extension as new partners and systems emerge.
The platform also embraces industry standards, supporting OpenDirect, AdCOM, the Deals API, and multiple IAB taxonomies for content, audiences, and ad products. Features like sellers.json parsing and OpenRTB supply chain (schain) integration enhance transparency and compliance, making the Seller Agent a future-proof component in the programmatic ecosystem.
Agent SDK Production Ready Implementations
Both the Buyer and Seller Agent SDKs are built for real-world deployment at scale, with a production-ready implementation that prioritizes reliability, flexibility, and performance. It supports multiple storage backends—including Redis, Postgres, and SQLite—and exposes APIs across REST, MCP, and A2A interfaces for seamless integration into existing systems.
The SDKs support a wide range of deployment options, from local Docker environments to fully managed cloud infrastructures using AWS CloudFormation, Terraform, and Kubernetes-compatible containers. Features like VPC networking, multi–availability zone support, load balancing, and secrets management ensure enterprise-grade reliability and security.
The system is designed to scale effortlessly, with auto-scaling containers and storage, and compatibility with Kubernetes environments. Production logs and audit trails provide full visibility, while configurable LLM providers allow teams to tailor AI capabilities to their needs. Altogether, Seller Agent delivers a robust, extensible foundation for modern, agent-driven supply-side operations.
Operational transparency is equally critical: comprehensive audit logs, production logging, and an event bus for real-time notifications provide full visibility into system behavior. With support for multiple storage backends (Redis, Postgres, SQLite), auto-scaling infrastructure, and flexible APIs (REST, MCP, A2A), Buyer Agent 2.0 is designed to integrate seamlessly into modern tech stacks while scaling with demand.
Operationalizing Agentic Audiences
The Agentic Audiences specification, donated by LiveRamp, seeks to standardize the usage of vector embeddings in agentic advertising. To utilize this specification at scale, the task force focused on enabling audience signaling within the bidstream via OpenRTB and allowing publishers to implement this within their Prebid environments.
This work is now complete. We have introduced a new community extension in OpenRTB that signals the necessary information to transact on an Agentic Audience. Additionally, a Prebid module conforming to this OpenRTB extension has been created and merged into the Prebid core. Both are available today for partners ready to begin transacting.
To drive full adoption, we have also released an open-source scorer. This tool demonstrates the power of embeddings and their application in programmatic advertising via the new OpenRTB extension. Using this scorer a user can:
- Process an OpenRTB Agentic Audience request.
- Extract the embedding.
- Compare it against campaign targeting embeddings.
- Match and score the relevance of the incoming embedding relative to the targeting embeddings.
What’s Next?
Taken together these agents promise to unlock thousands of hours of manual, often error-prone, tasks that facilitate digital transactions today. Those hours can now be used by experts on strategy and experimentation to create more value for their customers, rather than the tedious work of troubleshooting. That’s the true value of what generative AI can bring to the ecosystem.
And the Agentic Task Force is just getting started. AAMP 3.0 is well underway which builds on some of the features above and extends the framework into more transaction and media types.
Miguel Morales
Director, Agentic Initiative
IAB Tech Lab