Agentic buying got its infrastructure layer.

Agent-to-agent media buying stopped being a demo in the first quarter of 2026. Omnicom told its Q1 earnings call that it had executed live media buys for real clients through an agent-to-agent framework: software agents connecting directly to publisher APIs, checking inventory, and running automated negotiations, with the stated aim of compressing the supply chain between advertiser and publisher (Digiday, April 2026; MediaPost, April 29, 2026). A holding company telling investors it is already routing budget this way changes the question for everyone else from whether to how, and with what controls.

The plumbing underneath that disclosure is the Ad Context Protocol, an open standard for agentic advertising launched in October 2025 and built on the Model Context Protocol, with founding participants including Scope3, Yahoo, and PubMatic (ExchangeWire and trade coverage, April 2026). AdCP gives buying and selling agents a common language for discovering inventory, activating audiences, and transacting. The activity ran ahead of the governance: agencies were already testing AI planning agents and edging toward buying tools at the end of 2025, most without a shared standard for what an agent is allowed to decide (Digiday, December 15, 2025).

The IAB Tech Lab unveiled its Agentic Roadmap for Digital Advertising on January 6, 2026, and on February 26 named the whole program AAMP, the Agentic Advertising Management Protocols, to end market confusion about which piece was which (IAB Tech Lab, February 26, 2026). The framework stacks three layers: agentic foundations at the base, the protocols in the middle, and trust and transparency on top. IAB Europe publicly endorsed the roadmap in July, which matters because it points toward one global framework rather than regional forks (IAB Europe, July 2026).

AAMP 2.3, released July 30, is the version that makes the trust layer concrete. It integrates the IAB Diligence Platform and SafeGuard Privacy compliance gates directly into the Buyer Agent, adds deterministic negotiation guardrails with server-side trust verification, and requires human approval for decisions that move value (IAB Tech Lab, July 30, 2026). It also widens the rails the agents run on: support for enterprise deployment on Amazon Bedrock AgentCore and Databricks, buying integration with Meta, reporting through Google Ad Manager, full OpenDirect compliance, and an Agentic Audiences specification ready for transactions (IAB Tech Lab, July 30, 2026).

The useful reading of AAMP 2.3 is contractual. Before this release, an agentic vendor’s privacy posture and pricing behavior were policy claims: statements in a deck, checked after the fact if at all. Embedding the diligence platform and the privacy gates into the Buyer Agent itself, and making negotiation guardrails deterministic with server-side verification, turns those claims into infrastructure a buyer can test against (IAB Tech Lab, July 30, 2026). The requirement for human approval on value-moving decisions draws the line most agencies were looking for: agents can research, assemble, and propose at machine speed, and a person signs what spends the money.

That maps directly onto the discipline buyers already apply to the rest of the chain. The same review that asks a supply path to justify every hop can now ask an agentic vendor which AAMP release it implements, where its guardrails run, and what its approval record shows. Vendors that cannot answer are making the older kind of claim.

For an independent agency, the practical question is which decisions to delegate and what record to keep of the ones delegated. Planning research, inventory discovery, and audience assembly are where agents already earn their keep; allocation and price are where the human approval requirement in AAMP 2.3 says the industry itself wants a person in the loop (IAB Tech Lab, July 30, 2026). A written line between the two, per client, is the artifact that survives a client review.

Delegation also raises the bar on proof. When an agent executed the buy, the post-campaign conversation depends even more on measurement that stands apart from the system being measured, and on an operating partner that can walk the client back through what the agent did and why. The standards now exist to make that walk-through possible. The agencies that adopt the record-keeping before the tooling forces it will have the easier 2027.