Artificial intelligence is changing programmatic advertising by moving software beyond bidding on single impressions and into planning, negotiating and recording whole deals, with people setting the limits. Auctions and optimisation still run, but in some markets a layer of AI agents now sits on top of them. The ten developments below are observed and sourced, and none of them is universal yet.

TL;DR. The clearest AI trends in programmatic are agents negotiating deals, open standards such as AAMP and AdCP that let those agents work together, and checks on which agent is acting for whom. Benchmarks, shared deal records and human approval thresholds are growing alongside them, because an agent that completes a deal has not necessarily done a good one. Curation, creative, CTV and measurement are changing too, with people still signing off on the steps that commit money.

1. Are AI agents negotiating deals rather than bidding on impressions?

Some buyers and sellers now use AI agents to agree deal terms, not only to bid in auctions. AgenticAdvertising.org, which publishes the Ad Context Protocol, says the first agent-to-agent media buy ran on 16 October 2025, using real money, inventory from LG Ads and AI agents working with human oversight.

PubMatic launched AgenticOS in January 2026 so that agents can plan, transact and optimise across Programmatic Guaranteed and Private Marketplace deals. The company reported that early agentic campaign tests cut campaign setup time by 87% and issue resolution by 70%.

The shift matters because a negotiated deal creates commitments that outlast a single impression. Price, guarantees and make-goods all have to be agreed, held and checked later.

2. Why are shared deal records replacing separate books?

When agents agree terms, both sides need to read the same record of what was agreed. IAB Tech Lab argues that for agentic direct-sold advertising, campaign state "must be machine-readable, deterministic, and canonical", held in a structured data model "that both buyer and seller agents read from the same source of record".

The standards are moving the same way. IAB Tech Lab's OpenProposal overview lists reconciliation and a settlement reference among the specification's headers, and AdCP defines performance standards, measurement terms and cancellation policy as "structured, machine-readable fields".

A marketplace such as Alkimi records the agreed terms in a DealSheet: written at the moment of agreement, timestamped, held by both parties and retrievable by either side independently. The practical effect is that neither side has to rely on the other's copy if a dispute arises.

3. Which industry standards are being written for advertising agents?

Two open efforts lead this work. IAB Tech Lab's Agentic Advertising Management Protocols (AAMP) reached version 3.0 on 22 September 2026, adding OpenProposal, a specification for how sellers describe products and respond to buyer briefs in an agent-readable format.

OpenProposal is in public comment until 22 October 2026. Mike Treon, head of CTV strategy at PMG, said in the launch announcement: "What's new in AAMP 3.0 is that the reasoning travels with the proposal."

The Ad Context Protocol (AdCP), published by AgenticAdvertising.org, is the second effort. The organisation describes AdCP as the open protocol that lets AI agents "discover, plan, buy, sell, and measure media across every channel", and lists at least 154 member companies.

4. How do publishers and buyers know which agent they are dealing with?

Agent identity and authorisation are becoming a layer of their own. IAB Tech Lab's Agent Registry lets companies list their agents, describe each agent's purpose and maturity, and ties every registrant to its GPP ID for validation.

The registry's published roadmap includes a way to verify that a seller agent really represents a property, described as "like ads.txt for agents". In the AAMP 3.0 announcement, Michael Palmer of WPP Media said the release standardises "agent authentication and authorization protocols".

AdCP relies on signed requests. Its request signing guide says signing stays optional through AdCP 3.2 and becomes mandatory in AdCP 4.0 for all spend-committing operations.

5. Why are AI buying agents being benchmarked before deployment?

Testing agents before they handle real budgets is becoming a step of its own, because finishing a deal and protecting the buyer are different things. In Concourse Bench v1, 55 of 96 registered attempts by AI buying agents ended in a completed buy, yet 95 of the 110 contracts signed in those buys promised a viewability floor below 70%.

Sandboxes serve a similar purpose. Jonas Jaanimagi of IAB Australia said in the AAMP 3.0 announcement that its role in a seller-side pilot is to help publishers "test discovery, proposal and approval workflows in a sandbox before any live spend is involved".

The lesson for buyers is to test against the outcomes that matter to them, such as contract protections, rather than completion alone.

6. How much autonomy should an advertising agent have?

Approval limits are being designed into agent systems rather than added afterwards. IAB Tech Lab says the AAMP agent SDKs continue to harden deterministic controls over pricing, ad packages, orders and media buy details. AdCP's Embedded Human Judgment principles state that "responsibility cannot be delegated to software" and call for explicit decision boundaries, escalation triggers and identifiable human owners.

Live work reflects the same idea. In the NBCUniversal and RPA initiative reported by AdExchanger, agents refine proposals through automated feedback loops, and the campaign executes automatically only "once humans on both sides of the deal sign off".

One way to stage this is earned autonomy, in which an agent moves through five levels as it proves reliable: Observe, Recommend, Draft, Human-approved action and Bounded automatic action.

7. How are curation and supply path optimisation changing with AI?

AI is pushing some buyers towards supply that is curated before the auction rather than optimised after it. Experian describes dynamic curation as aligning predictive audiences "with contexts where engagement is strongest" and says it moves programmatic "away from broad open exchange buying and toward more curated, intentional activation".

Agents also take on the setup work that curation creates. NBCUniversal's chief product officer, Ryan McConville, told AdExchanger that a DSP can automate buying a curated deal ID, but setting up the deal ID, choosing the inventory and handling the logistics has been manual work that agentic tools are good at automating.

For supply path optimisation, the question widens from which route an impression takes to which agent and which counterparty assembled the deal.

8. How is generative AI changing ad creative, and who signs it off?

Generative AI is now a routine part of ad production. Research by IAB and Sonata Insights found that 83% of ad executives say their company has deployed AI in the creative process, up from 60% in the 2024 study.

The same research found a perception gap: 82% of executives believe Gen Z and Millennial consumers feel positive about AI-generated ads, against 45% of those consumers who say they do. While 89% of advertisers who have used generative AI to create ads disclose it at least sometimes, fewer than half always do.

Scale therefore raises the stakes for human review and disclosure. IAB's AI Transparency and Disclosure Framework takes a risk-based approach, requiring disclosure when AI materially affects authenticity, identity or representation in ways that could mislead consumers.

9. How is AI changing CTV and TV buying?

Television, including live sport, is one of the places where agent-to-agent buying has reached real inventory. NBCUniversal, RPA, FreeWheel and Newton Research announced agentic buying across linear and streaming, with a first execution featuring a brand's Q1 2026 investment in placements including live football playoff games.

Linear still accounts for about 80% of premium TV ad impressions and has remained mainly manual, according to McConville in AdExchanger. FreeWheel's David Dworin told the same outlet that agents can approve creative in seconds, which matters in live sport, where a late approval can mean an ad does not run.

McConville also said that simpler execution could make NBCUniversal's inventory accessible to smaller advertisers that lack the resources for large-scale programmatic buys.

10. How is measurement changing when agents buy the media?

Measurement terms are moving into the deal itself. The AAMP 3.0 lifecycle ends with a report stage, so that results can be "evaluated against the original objectives and commitments of the plan".

AdCP's accountability documentation lets buyer and seller agents negotiate performance standards, the vendor that counts the billing metric and the available make-good remedies before budget is committed.

Incrementality is the harder question, because a promised floor in a contract says nothing about what a campaign actually caused. Advertisers still need their own tests, such as holdout groups, to judge whether agent-bought media drove results that would not otherwise have happened.

What do these trends mean for teams deploying agents?

The common thread is accountability: which agent acted, who approved the commitment and which record both sides trust. Alkimi, an enterprise platform for agentic advertising, is built around those questions with three products.

Concourse is a live benchmarking product for AI media-buying agents and published the Bench v1 figures cited above. The 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.

The Marketplace is neutral, works within IAB Tech Lab's AAMP framework today and is being built to be AdCP compatible. It uses AAMP identity checks, keeps human approval rights at the deal level and records each agreement in a DealSheet. The company is also taking part in the OpenProposal v3.0 public comment period.

Frequently asked questions

How is artificial intelligence changing programmatic advertising?

AI is extending programmatic from bidding on impressions to planning, negotiating and recording deals through software agents. People still set the objectives, the limits and the approvals that commit money.

Will AI agents replace programmatic auctions?

The deployments described here add agents on top of existing programmatic systems rather than removing them. NBCUniversal's chief product officer has described agents as sitting on top of programmatic, not replacing it.

What standards cover AI agents in advertising?

The main open efforts are IAB Tech Lab's AAMP, which now includes OpenProposal, and the Ad Context Protocol (AdCP) published by AgenticAdvertising.org. Both define how buyer and seller agents discover inventory, exchange terms and transact.

Do humans still approve AI-negotiated media deals?

In the deployments cited here, yes: people set the mandate and sign off on consequential commitments. How much an agent may do alone depends on the approval thresholds each organisation sets.

AI in Programmatic Advertising: 10 Trends to Watch