A buyer agent is software, usually driven by a large language model, that buys advertising on behalf of an advertiser or agency. Given a campaign brief, it identifies sellers, reads what they offer, compares prices, negotiates terms and books the deal, pausing for human approval above its spending authority. Its counterpart is the seller agent, which represents a publisher's inventory in the same exchange.
The category exists because direct buying never scaled. Real-time auctions automated the pricing of single impressions, but guaranteed deals, private marketplaces and television commitments still move through requests for proposal, spreadsheets and email. Buyer agents are the latest attempt to automate that negotiation layer without forcing every transaction through an auction.
How a buyer agent works
Everything starts with a brief of budget, flight dates, audience, channels and performance targets. The agent must then find legitimate sellers. Under the Ad Context Protocol (AdCP), publishers list the agents authorised to sell their inventory in a file at /.well-known/adagents.json, much as ads.txt lists authorised exchanges, and must pin signing keys for any agent permitted to create or change bookings, according to the AdCP documentation. IAB Tech Lab, the standards body of the Interactive Advertising Bureau (IAB), runs an Agent Registry that checks each company against its Global Privacy Platform (GPP) and Transparency and Consent Framework (TCF) identifiers.
Discovery comes next. In AdCP the agent sends the brief in plain language through a task called get_products and receives products in a common schema, each carrying pricing options, delivery forecasts and creative requirements. IAB Tech Lab's open-source reference buyer agent, still labelled alpha, browses seller media kits instead. Anonymous callers see only price ranges. A buyer that reveals its seat, agency or advertiser identity unlocks progressively better rates, with discounts of roughly 5%, 10% and 15% in the reference code. A portfolio manager agent splits the budget across branding, connected television, mobile and performance, while channel specialists research the sellers.
Negotiation is deliberately mechanical. In the documented example, the default strategy opens at a $20 CPM (cost per thousand impressions), raises its offer by $2 a round, accepts any seller price at or below $30 and walks away after five rounds.
Booking is where protocols replace conversation. In AdCP a purchase is a create_media_buy call containing packages, budgets and a chosen pricing option; version 3.0 made an idempotency key of 16 to 255 characters mandatory on such calls, so a retried request cannot buy the same inventory twice. Governance runs alongside: check_governance tests each purchase against a plan registered through sync_plans, and anything beyond the buyer's authority waits in an input-required state until a person approves it. IAB Tech Lab's agent books through OpenDirect 2.1 and the Deals API, handles linear television scatter buys priced per rating point, and returns Deal IDs for activation in The Trade Desk, Display & Video 360 (DV360) or Amazon DSP.
That hand-off places the buyer agent upstream of the demand-side platform (DSP) in most deployments: the agent negotiates, and the DSP still bids on each impression. Under AdCP's Trusted Match Protocol, the publisher later asks separately whether a page fits a package and whether a user qualifies, so the agent never sees identity and context together.
Agencies and trading desks run buyer agents, and vendors such as Scope3 build them. So, increasingly, do sell-side firms: Magnite introduced its own buyer agent on April 27, 2026, with Kepler and MiQ testing it against Disney Advertising inventory.
From Kasbah to AdCP
The idea is older than the models. In April 1996 Anthony Chavez and Pattie Maes of the Massachusetts Institute of Technology's Media Lab presented Kasbah, a marketplace where people created agents to buy and sell goods for them. A Kasbah buying agent held a desired price, a highest acceptable price and a deadline, adjusted its offer as time passed and left final approval to its owner: in substance, the threshold strategy IAB Tech Lab's code ships 30 years later. Advertising joined the research agenda in July 2009, when the Trading Agent Competition held its first Ad Auctions tournament, in which agents bid for sponsored search positions on behalf of simulated advertisers.
The rails came next. AOL, Yahoo, Microsoft and Yieldex formed a working group in June 2013 to standardise the automated sale of reserved inventory. The IAB finalised OpenDirect 1.0 in January 2015 and released version 2.0 for comment on January 10, 2018, adding negotiation support.
Large language models revived the concept. Ari Paparo, founder of Marketecture Media, argued on July 21, 2025 that agentic systems could take over the setup, targeting and optimisation work that justifies DSP fees. On October 15, 2025, Scope3, Yahoo, PubMatic, Swivel, Triton Digital and Optable launched AdCP on Anthropic's Model Context Protocol (MCP), with nine core tasks. The first live test followed in December 2025, when Scope3 acted as buyer agent against Magnite's seller agent in SpringServe. Version 3.0, published by the governing trade association AgenticAdvertising.org, arrived in April 2026, and version 2 was fully deprecated on August 1, 2026, according to the protocol's documentation.
IAB Tech Lab chose to extend existing standards instead. It published buyer and seller agent software development kits (SDKs) alongside an agentic roadmap on January 6, 2026, and on February 26, 2026 named the programme AAMP, the Agentic Advertising Management Protocols. AAMP 2.0 added programmatic guaranteed deals and agentic private marketplaces in April 2026. Version 2.3, released on July 30, 2026, introduced a pricing provenance field to stop agents inventing prices and a vendor approval gate linked to the IAB Diligence Platform. In the reference code the gate fails closed, admitting no seller when the approval service is unreachable.
Why marketers are watching
Fees are the first reason. An agent that books deal-based inventory and passes a Deal ID to a DSP competes with the impression-level decisioning that DSPs charge for. Jeff Green, chief executive of The Trade Desk, used an August 6, 2026 earnings call to attack rivals wrapping fixed-price deals in agentic technology and competing on low fees, saying such approaches "look more like ad networks of 2006".
Integration cost is the second. Jochen Schlosser, chief technology officer at Adform, has put custom supply-side platform (SSP) deal integrations at 250 to 350 days of work, against 50 on a shared standard. Speed is where vendors have evidence, though thin. Magnite reported roughly 70% less setup time in its first agentic campaign in Europe, the Middle East and Africa, run with Amnet France. Deployments now cross channels, including an out-of-home campaign in which Draft Digital's buy-side agent traded with Broadsign's sell-side agent, announced on May 27, 2026.
Where it breaks down
Price evidence is mixed. DataBeat's June 2026 report found conventional buyers clearing at a $6.95 CPM against $6.13 for agentic buyers, a 13.4% gap, with agentic demand entering 86% fewer auctions. Lower prices can signal efficiency or cheaper, less contested inventory. The data measures price paid, not results.
Running costs push the other way, because every tool call consumes tokens. Draft Digital, a Dutch agency, collapsed twelve MCP calls into a single buyer agent to contain that burn, without publishing a benchmark for the saving.
Loyalty is the sharper dispute. A buyer agent is meant to serve the buyer, yet several come from sell-side firms. In the Amnet France test, Magnite's buyer agent transacted with Magnite's seller agent through Magnite's own coordination layer, the structure behind Green's charge that such products favour publishers. Paparo questioned in November 2025whether agents could overcome sellers' resistance to price transparency, scale economics and data fragmentation, the problems that stalled programmatic direct.
The standards also compete. Shailley Singh, chief operating officer of IAB Tech Lab, counted 13 functions on which AAMP and AdCP overlap on August 20, 2026, from buyer and seller agents to human approval before spend. They book through different objects, so an agent built for one does not automatically speak the other. Even the vocabulary splits. AdCP says sales agent where IAB Tech Lab says seller agent, and AdCP's earlier documentation described the buying side as an orchestrator acting for a principal.
Not the same as
Demand-side platform. A DSP bids on individual impressions in real time; most buyer agents work upstream of it. Assistants such as Amazon's Ads Agent, announced on November 11, 2025, automate work inside one platform rather than trading with outside sellers.
Buyer. In OpenRTB, the buyer is the account billed for an impression, distinct from the bidder that submits it. The agent acts for that buyer.
Seller agent. The counterpart that exposes a publisher's inventory, packages and prices. Prebid.org maintains an open-source AdCP sales agent.
Shopping agent. In agentic commerce, software buys goods for consumers, as with the agent Meta placed inside WhatsApp. In property, a buyer's agent is a licensed broker representing a home purchaser.
Recent developments
IAB Tech Lab published AAMP 3.0 on September 22, 2026 with a draft specification called OpenProposal, describing how sellers present packages so buyer agents can discover, compare, qualify and negotiate them against a brief. Each line item then executes through OpenDirect 2.1, the Deals API or OpenRTB 2.6, depending on its transaction_mechanism field. Comments close on October 22, 2026.
On September 16 Magnite extended its Orchestration layer so buyer agents can reach local linear television, Samsung home screens and pause ads, and iHeartMedia audio, disclosing neither volumes nor fees. Michael Barrett, its chief executive, has put the most optimistic 2027 forecasts for agentic spend at $600m to $700m. AdCP stood at version 3.1, with 3.2 in release candidate, as of September 2026. Schlosser's assessment, aired on September 23, was that neither protocol yet runs at scale.
Timeline
- April 1996: Anthony Chavez and Pattie Maes present Kasbah, an agent marketplace for buying and selling goods
- July 2009: The Trading Agent Competition holds its first Ad Auctions tournament
- June 2013: AOL, Yahoo, Microsoft and Yieldex form the OpenDirect working group
- January 2015: The IAB finalises OpenDirect 1.0
- January 10, 2018: IAB Tech Lab releases OpenDirect 2.0 for public comment
- July 21, 2025: Ari Paparo argues agentic AI threatens DSP business models
- October 15, 2025: Six companies launch the Ad Context Protocol with nine core tasks
- November 11, 2025: Amazon announces Ads Agent at unBoxed
- December 2025: Scope3 acts as buyer agent in Magnite's first seller agent test
- January 6, 2026: IAB Tech Lab publishes its agentic roadmap with buyer and seller agent SDKs
- February 26, 2026: IAB Tech Lab names its agentic programme AAMP
- March 11, 2026: The IAB Tech Lab Agent Registry reaches ten entries
- April 2026: AAMP 2.0 adds programmatic guaranteed and agentic private marketplace transactions; AgenticAdvertising.org releases AdCP 3.0
- April 27, 2026: Magnite introduces a buyer agent tested by Kepler and MiQ
- May 27, 2026: Broadsign and Draft Digital announce an agent-to-agent out-of-home campaign
- June 22, 2026: DataBeat reports a 13.4% CPM gap between conventional and agentic buyers
- July 30, 2026: AAMP 2.3 adds pricing provenance and a vendor approval gate
- August 1, 2026: AdCP version 2 is fully deprecated
- August 6, 2026: Jeff Green compares agentic fixed-price products to 2006 ad networks
- August 20, 2026: IAB Tech Lab counts 13 functions shared by AAMP and AdCP
- September 8, 2026: Magnite discloses its first agentic campaign in Europe, the Middle East and Africa, with Amnet France
- September 16, 2026: Magnite extends Orchestration to linear television, home screens, pause ads and audio
- September 22, 2026: IAB Tech Lab publishes AAMP 3.0 and the OpenProposal draft
- October 22, 2026: The AAMP 3.0 public comment period is scheduled to close
Related PPC Land coverage
- IAB Tech Lab's agent registry hits 10 with Amazon and new deployment types - How agent entries are validated against privacy framework identifiers before approval.
- Magnite expands AI buyer agent and SpringServe mediation with Disney and MiQ - The April 2026 buyer agent launch and its first test partners.
- Explaining programmatic guaranteed deals - The OpenDirect history behind the deals buyer agents now book.
- Agentic AI threatens traditional DSP business models says industry veteran - Ari Paparo's July 2025 case that agents could absorb DSP functions.
- Ad Context Protocol (AdCP) launches for advertising automation - The October 2025 launch, founding members and nine core tasks.
- Magnite builds seller agent into SpringServe for AI-driven ad buying - The first agent-to-agent test, with Scope3 as buyer agent.
- IAB Tech Lab names its agentic ad initiative AAMP to end market confusion - The three-pillar architecture and the three-level buyer and seller agent hierarchy.
- AAMP 2.3 blocks AI agents from inventing ad prices, IAB Tech Lab says - Pricing provenance, the vendor approval gate and the buyer agent's deployment surface.
- Trade Desk faces rivals wrapping guaranteed deals in agentic AI at 1% fees - Jeff Green's critique of agentic fixed-price buying and the fee gap behind it.
- Neither of ad tech's two AI agent protocols runs at scale, Adform CTO says - Jochen Schlosser's integration cost estimates and his view of protocol adoption.
- Magnite's first EMEA agentic test cuts CTV setup time 70% - A single-campaign speed claim from Magnite's work with Amnet France.
- Broadsign and Draft Digital's AI agents run first fully agentic OOH campaign - A buy-side agent and a sell-side agent completing an out-of-home purchase.
- Programmatic buyers gain 13.4% CPM edge over AI agents, DataBeat finds - Clearing prices, fill rates and auction participation for agentic and conventional demand.
- Draft Digital cuts 12 MCP calls to one buyer agent to stop token burn - Consolidating buying decisions into one agent to contain inference costs.
- AI buying agents gain Samsung, iHeartMedia and DIRECTV ad space via Magnite - The September 2026 Orchestration expansion and the question of sell-side neutrality.
- Industry expert questions AdCP media buying protocol for ad automation - Ari Paparo's doubts about agents overcoming the problems of programmatic direct.
- AAMP and AdCP overlap on 13 functions, IAB Tech Lab COO says - Shailley Singh's count of shared functions between the two frameworks.
- Amazon launches AI agent for automated campaign management - An agent that automates work inside a single DSP rather than trading with sellers.
- Explaining buyer - The billed account in OpenRTB, seat identification and the difference from the bidder.
- Meta puts an AI agent that buys things inside WhatsApp, US only - A consumer shopping agent, the commerce meaning of the term.
- IAB Tech Lab gives industry until October 22 to weigh in on AAMP 3.0 - The OpenProposal draft and the lifecycle it defines for buyer and seller agents.
Summary
Who. Advertisers, agencies and trading desks that deploy buyer agents, vendors such as Scope3 and Magnite that build them, publishers and SSPs that run the seller agents they trade with, and two standards efforts: AgenticAdvertising.org, which governs AdCP, and IAB Tech Lab, which maintains AAMP and its reference buyer agent.
What. Software that takes a campaign brief, discovers authorised sellers, compares products, negotiates price within set limits and books media, returning Deal IDs or media buys that a DSP or ad server then delivers. Its mechanics run through tasks such as get_products and create_media_buy in AdCP, and through OpenDirect 2.1, the Deals API and tiered identity pricing in AAMP.
When. The concept dates to academic agent marketplaces in 1996 and advertising agent competitions from 2009. Commercial buyer agents arrived with AdCP on October 15, 2025 and a first live test in December 2025, followed by AAMP in February 2026 and draft proposal standards in September 2026.
Where. Upstream of the auction in programmatic guaranteed deals, private marketplaces, connected television, out-of-home, audio and linear television, with deployments concentrated so far in the United States and Europe.
Why. Direct media buying depends on manual proposals and custom integrations that software negotiators could compress. The evidence so far shows faster setup but lower clearing prices of uncertain meaning, while disputes continue over fees, token costs, competing standards and whether an agent built by a seller can act in the buyer's interest.
Discussion