AdCP changes the agency trading desk by turning its most repetitive paperwork into structured requests between agents: briefs become typed requests, RFP rounds become seller-authored proposals, and owed reports become obligations the protocol tracks. It does not change who is answerable to the client. Traders keep the judgement calls, set the limits agents work within and approve the commitments that matter.

TL;DR. The Ad Context Protocol (AdCP), published and governed by AgenticAdvertising.org, reached general availability at version 3.2 on 30 September 2026, according to its release overview. Version 3.2 gives buyer agents one task per step of a buy, from discovery through proposals, acceptance and in-flight changes, and keeps every proposal as an immutable snapshot. For agencies, that moves version-chasing, reformatting and report-chasing to agents, while people keep the objectives, the seller choices and the sign-off.

How will AdCP change agency trading desks?

AdCP will change trading desks less by speeding up buying than by making each step of a deal a recorded, typed event. The protocol's own summary of version 3.2 is blunt: "3.1 made agents safe to operate. 3.2 makes the deal between them explicit." A seller authors plans from a brief, the buyer negotiates typed terms, inventory is held, and one exact snapshot is accepted.

That is a change to the desk's paperwork, not its purpose. The AdCP documentation for decision-makers frames adoption as additive, noting that "a buyer agent sits alongside the DSP" rather than replacing existing programmatic relationships (AdCP Decision-Makers module L3).

Finishing a workflow is also not the same as doing it well. In Concourse, a live benchmark of AI media-buying agents run by Alkimi, 55 of 96 registered attempts ended in a completed buy, yet 95 of the 110 contracts signed in those buys promised a viewability floor below 70% (Concourse Bench v1). A protocol can standardise how a deal is agreed; whether the terms protect the client remains a trading judgement.

What does the trading desk look like before and after AdCP?

The table below maps the desk's core tasks onto the AdCP 3.2 tasks that take them over. Task names come from the AdCP 3.2 release overview and the buyer briefs guide, and features the documentation marks experimental are labelled.

Desk task	Before AdCP	After: what the buying agent does	What stays with people
Briefs	A planner writes a brief in a document and each seller interprets it in its own way.	Splits the brief into intent (the brief text) and hard constraints (offer filters and a targeting overlay), so sellers do not guess what is negotiable.	Setting the objective, describing the audience in plain words and naming the sensitivities no field captures.
RFPs	Requests go out by email and responses come back over days in different formats.	Calls request_proposals, and each seller returns non-binding plan snapshots authored from the brief.	Choosing which sellers are asked and which are allowed in the plan.
Proposal comparison	Traders normalise spreadsheets and track versions by hand.	Reads the typed commercial terms on each proposal; refine_proposals records every revision as a new immutable snapshot and decline_proposals records the buyer's feedback.	Judging which terms actually protect the client, and managing the seller relationships behind them.
Approvals	Insertion orders and email sign-offs.	Finalising a proposal holds inventory until an expiry time, and accept_proposal turns the held snapshot into a media buy. Campaign governance checks each action against the campaign plan the buyer sets (experimental).	Approving commitments above agreed thresholds and handling exceptions.
In-flight changes	Traders ask seller reps or ad ops to adjust delivery.	control_media_buy changes delivery inside the accepted terms; anything outside them needs a new snapshot.	Deciding when a change is really a renegotiation.
Reporting	Teams chase late reports and reconcile mismatched files.	Narrows get_media_buy_delivery to the metrics it needs, while Reliable Reporting shows each owed report as waiting, healthy, delayed or needing action (experimental).	Explaining results to the client and raising disputes.

What do agencies hand to agents under AdCP?

Agencies hand agents the work nobody on a desk enjoys: reformatting seller responses, tracking which proposal version is current, re-keying changes and chasing reports. AdCP turns each of those into a structured step, so the agent can do it the same way every time and leave a record. Under Reliable Reporting, for instance, the documentation's aim is that "a missing report shows up as missing" rather than disappearing quietly.

Agency leaders already sort tasks this way. "Every task and everything we do sits on a sliding scale of low to high human oversight," said Rebekah Shalit, head of platforms at Dentsu, in a Digiday report from October 2026. She put repeatable tasks, billing and campaign reporting at one end, and strategic planning, which she said needs "human craft", at the other.

The handover has limits that the documentation states plainly: "The contract ends at acceptance." Accepting a proposal does not standardise invoicing, payment, currency exchange, make-goods or disputes, which sit in a 3.3 commercial-completion workstream. For an hour-by-hour view of a desk running agents today, see A Day on an Agentic Trading Desk.

What stays with people on an AdCP trading desk?

People keep the decisions that carry financial or reputational consequences for the client: the objective, the seller list, the acceptable terms and the final commitment. AdCP's governance model names a human reviewer who "handles exceptions, overrides, and ambiguous policy decisions when required", alongside a buyer agent, a governance agent and a seller agent (governance protocol by role). Campaign governance is still marked experimental in AdCP 3.2, so agencies should check what each seller supports before relying on it.

Large agencies describe the same split. When WPP Media launched its Buyer Agent for Video in June 2026, financial commitments and campaign activation required explicit human sign-off, Digiday reported. Lauren Wetzel, WPP Media's global president of data and technology solutions, summed it up: "It's accountable media buying, humans are at the helm, agents are in the loop."

How much should an agency let an AdCP agent do on its own?

The answer should grow with evidence, not enthusiasm. One way to stage it is an earned autonomy ladder with five rungs, each mapped to AdCP tasks:

Observe. The agent reads offers with list_products and delivery with get_media_buy_delivery, and changes nothing.
Recommend. The agent calls request_proposals and ranks what comes back for a trader.
Draft. The agent revises terms through refine_proposals and prepares the version a person will review.
Human-approved action. A trader approves before the agent finalises a hold or calls accept_proposal.
Bounded automatic action. The agent uses control_media_buy on its own, but only inside terms a person has already accepted.

An agent moves up a rung when its record at the current one justifies it, and drops back when it does not. AdCP makes that judgement easier because every proposal snapshot and accepted term is already on record. The thresholds themselves are a desk decision, covered in more depth in How to Design Approval Thresholds for Agentic Campaigns.

How should an agency start using AdCP on the trading desk?

Start narrow. The AdCP decision-makers module advises that when a surface is experimental, teams should "pilot one surface with one partner to learn cheaply and reversibly", then standardise once the value is proven.

Check capabilities before assuming features. Sellers advertise which compact tasks they serve, and where they do not, buyers use the 3.x tasks get_products, create_media_buy and update_media_buy, which the 3.2 overview says stay available throughout 3.x. Write the desk's approval thresholds and escalation routes before the pilot starts, and read How Does AdCP Governance Check Budgets, Brand Safety and Human Approval before switching governance checks on.

Where do benchmarking, training and a neutral marketplace fit?

Alkimi is neutral on standards: it can work with agents built on AdCP or on IAB Tech Lab's AAMP, with or without a DSP. Its Concourse benchmark tests any buying agent without asking which standard it uses, and its Gym lets buyers and sellers build their own agents, set the rules they work to and spar against counterparty agents before any real money moves. Its Marketplace is neutral and keeps human approval rights at the deal level, with each agreement recorded in a DealSheet that is written at agreement, timestamped, held by both parties and retrievable by either side independently.

For an agency, that is one way to put the earned autonomy ladder into practice: agents take the parts of media buying people do not enjoy, and traders keep the judgement and approve the deals that matter. Teams that want to test a buying agent before it trades can start a scoping conversation through the Concourse contact form.

Frequently asked questions

Will AdCP replace agency traders? No. AdCP standardises how agents discover, propose, accept and report on media, while people keep the objectives, seller choices and approvals that carry risk for the client.

Does an agency need a DSP to buy through AdCP? The AdCP documentation describes a buyer agent as sitting alongside the DSP rather than replacing it. Whether a DSP is involved in a particular buy depends on the seller and on the agency's own set-up.

Which AdCP 3.2 features are still experimental? The AdCP 3.2 overview lists campaign governance, Reliable Reporting, audience activation and the account change feed among its experimental surfaces. Experimental surfaces can change within 3.x, so the documentation advises pinning an exact release before using them.

Does AdCP cover invoicing and make-goods? Not yet. The 3.2 documentation states that accepting a proposal does not standardise invoicing, payment, make-goods or disputes, and a 3.3 workstream covers what comes after acceptance.

Do agencies have to choose between AAMP and AdCP? Not necessarily. AAMP is IAB Tech Lab's Agentic Advertising Management Protocols and AdCP is the Ad Context Protocol from AgenticAdvertising.org, and AAMP, AdCP, A2A and MCP: Which Protocol Does What sets out how they relate.