Most conversation about agentic advertising focuses on the execution layer: the bid, the buy, the delivery. OpenProposal, the draft IAB Tech Lab specification published as part of AAMP 3.0, starts much earlier. It addresses a problem that predates agentic buying entirely: the bandwidth constraint that limits how many publishers a media team can meaningfully evaluate.

What is OpenProposal?

OpenProposal is a machine-readable specification that defines what is being bought and sold in a format AI agents can assess at scale. It carries a proposal from first draft through to agreed terms, covering its contents, audience, commercial terms, guarantees, execution requirements, transaction eligibility, and settlement references.

OpenProposal is currently a draft. IAB Tech Lab has published it for public comment, open to members and non-members alike, and it will not become a ratified standard until a final version follows.

In IAB Tech Lab's explainer, the specification's eleven headers fall into four categories: Merchandising (identity and lifecycle, agent comprehension, inventory composition, audience reach, creative requirements), Optimisation (targeting envelope), Commitment (commercial terms, guarantees, transaction eligibility), and Reconciliation (execution, settlement reference).

OpenProposal is designed to be transaction-type-agnostic, covering Direct IOs, Programmatic Guaranteed, Preferred Deals, and open auction. It is also channel-agnostic: open web, CTV, mobile, audio, OOH, linear TV, and walled gardens are all in scope.

What problem does it solve?

As Barnaby Edwards, senior director of product marketing at IAB Tech Lab, sets out, the real constraint is time, not the quality of anyone's work. A media team might evaluate proposals from 10 to 20 publishers for a single campaign. Hundreds more may never see the brief, not because the inventory is a poor fit, but because the team does not have the capacity to review it.

The same constraint applies on the sell side: publishers have to decide which RFPs are worth responding to, knowing they cannot respond to every one. The result, Edwards argues, is media plans that lean too heavily on "familiar relationships and last year's choices".

Agents can screen far more proposals than any human team, but only if every proposal describes itself the same way. OpenProposal provides that consistency: a common format that lets buyer agents assess whether a publisher's inventory matches a brief without requiring human translation at each step.

Where does OpenProposal sit in the agentic workflow?

OpenProposal spans everything from the first brief or published offer to the moment both sides commit. Its most visible job is discovery and planning: enabling a buyer agent to move from a brief to an evaluated shortlist at a scale that was not previously practical. The same object then carries the proposal through negotiation to agreed terms.

Execution is where OpenProposal hands over. The specification does not replace existing IAB Tech Lab standards: each line item maps to OpenDirect, the Deals API, or OpenRTB, depending on how it is transacted. OpenProposal also builds on AdCOM and the Content, Audience, and Ad Product taxonomies, so the language a buyer agent reads at discovery stays consistent through negotiation, buying, execution, and reporting.

What happens when a proposal is agreed?

In OpenProposal, an offer becomes a commitment inside the same object. Most of the composition needs no back-and-forth: the buyer agent picks from options the seller has already priced, or sets values within bounds the seller has declared. According to the draft specification, only three of its 71 fields require the seller to respond.

When a proposal reaches agreed status, the buyer's commitment is frozen and the stored record is made self-contained and immutable. The proposal also carries buyer and seller assent records for the binding version, plus an append-only negotiation history showing which party proposed, countered, or accepted each change. Where a deal ID is needed for programmatic execution, it is issued at the point of commitment.

That design matters for accountability: the terms an agent evaluated, the changes each side made, and the version both sides accepted all sit in one record.

Why does a bilateral deal record still matter?

In a world where agents are making decisions at machine speed, the deal record is where accountability lives. If an agent-negotiated campaign delivers against different terms than agreed, the question is: what was actually committed to, by whom, and when?

An agreed OpenProposal answers much of that question. It fixes the terms, records both parties' assent, and shows who changed what along the way. What a shared format cannot settle on its own is how each organisation governs its agents: who is entitled to authorise a commitment, and where that authorisation is kept. A standard can record that a buyer assented. Requiring a person to approve the terms before an agent commits budget is a policy each buyer and seller has to set for itself.

A bilateral deal record makes that policy visible. It sits alongside the agreed proposal rather than replacing it, holds the same terms for buyer and seller, and carries the approval a person gave before execution began. The buyer holds a copy. The seller holds a copy. Neither depends on the other's systems, or on a marketplace, to prove what was agreed.

As OpenProposal moves the industry towards machine-readable discovery and agreement, the harder questions shift from format to governance: who approved the commitment, and who holds the record of it.