Agentic Ad Buying vs. the Agency Model
AI agents are starting to run strategy-to-execution end to end, and a standards fight over how to "agentify" OpenRTB and VAST is underway. Here's what survives the agency reckoning.
The agency business is built on selling hours, and AI agents are beginning to do the work those hours used to bill for. In 2026, AI agents are starting to take advertising workflows — strategy through execution — end to end, while a standards fight plays out over how to "agentify" the protocols that run programmatic advertising. The combination is squeezing holding-company headcount and forcing a question the industry has avoided: when a buyer's agent can negotiate directly with a seller's agent, what is the agency actually for?
The standards fight: revolution vs. evolution
Before agents can transact at scale, the industry needs shared protocols — and there is open disagreement about whether to build new ones or upgrade what exists. Two camps have formed:
Build from scratch — the Ad Context Protocol (AdCP). Launched in late 2025 by a consortium including Scope3, Yahoo, PubMatic, Swivel, Triton, and Optable, AdCP is built on Anthropic's Model Context Protocol and now governed by the independent AgenticAdvertising.org. It defines core agent tasks — discovery, comparison, campaign activation — and operates asynchronously, so responses can take seconds or days, allowing human approval while agents negotiate.
Build on what works — the IAB Tech Lab. The Tech Lab's stance is that agentic execution is already part of digital advertising and the smart path is to "agentify" established standards like OpenRTB, AdCOM, OpenDirect, and VAST. On February 26, 2026, it named its umbrella initiative the Agentic Advertising Management Protocols (AAMP), comprising the containerized Agentic Real-Time Framework (ARTF) and the Agentic Audiences Framework (formerly the User Context Protocol, donated by LiveRamp on November 3, 2025).
IAB Tech Lab CEO Anthony Katsur has been pointed about the philosophical divide: "When you remodel the kitchen, you don't burn down the house." He has also tempered expectations, predicting "several false starts" and warning that practical adoption "will require years of market experimentation, standardization and alignment across platforms, agencies and publishers." The honest state of play in mid-2026: real infrastructure, early live deployments, and loud disagreement about which standard wins.
The capability gap: what agents do, and don't, do yet
Agents are genuinely shipping. Digiday has reported on a Claude-powered campaign agent built on The Trade Desk that can ingest a media plan, reformat it into a compatible template, and build the campaign. Brands and their agency partners have started using agents to plan, execute, and optimize ad purchases. But the gap between "runs parts of it" and "runs the parts that matter" is the real story. As one 2026 field guide put it, agents can handle budget allocation and negative-keyword mining in recommend-only mode — but not yet the strategic judgment that requires understanding the business.
The headcount reckoning is already underway
The pressure on agency staffing is not theoretical — it is in the financial filings:
WPP cut headcount from 108,044 in December 2024 to 98,655 in December 2025. Its "Elevate28" turnaround targets £500 million in annual savings by 2028 and explicitly aims to become "a simpler, lower-cost, AI-enabled business." More than 75,000 employees — over 90% of client-facing staff — now use the WPP Open AI platform.
Omnicom, now the largest holding company after absorbing IPG, doubled its cost-savings target to $1.5 billion and confirmed more than 4,000 role cuts, largely in duplicative functions. CFO commentary made the logic blunt: labor costs lead the balance sheet.
Publicis is the counter-case — it grew headcount roughly 5% in 2025 and raised salaries about 7%, betting that data, technology, and people make it the "most valuable partner" rather than the biggest.
The most quoted warning comes from Sir Martin Sorrell, WPP's founder, now running S4 Capital, who said at SXSW Sydney in October 2025: "If there are 250,000 people globally in media planning and buying, there won't be in two or three years' time." (That remark predates our 2026 window and is included as the framing thesis.) His argument: media buying becomes fully algorithmic — Performance Max, Advantage+, and agentic systems handle targeting, allocation, and bidding automatically. Forrester calls the broader shift the "workforce inversion": the labor-arbitrage model, where expensive seniors oversaw armies of cheap juniors, breaks when AI does the junior work.
The demand side is also walking out
Even if agencies solve the AI problem internally, clients are bringing work in-house. The Association of National Advertisers reports 82% of members now have in-house agencies (up from 58% in 2013). Gartner's CMO survey found 39% of CMOs plan to cut agency spend, and 22% explicitly said AI reduced their need for external agencies. The pincer is real: agents automate the execution layer while clients absorb the strategy layer.
What survives
Not everything agencies do is automatable, and the survivors are visible in the pivots already happening:
Proprietary data and identity. WPP Open, Publicis's data acquisitions (AdgeAI, Lotame), and Omnicom's enterprise AI all bet that the durable asset is first-party data and the technology to activate it — not headcount.
Strategy and creative judgment. Agents widen the query net and run the buy; they cannot yet set the strategy or build the brand story. The differentiator moves to creative and signal quality.
Orchestration across a fragmenting ecosystem. As Publicis's Arthur Sadoun argues, the role is helping clients navigate and connect an ecosystem of platforms, data, and agents — a coordination job that survives even as execution automates.
Accountability and standards literacy. Someone has to decide which protocol to bet on, audit agent performance against human benchmarks, and own the measurement. That is a higher-value function than the manual buying it replaces.
The agency model that survives 2026 is not a slightly smaller version of today's. It is a fundamentally different operator: lighter on labor, heavier on technology and data, and valued for judgment and orchestration rather than the hours it bills. The agents are coming for the execution layer. The agencies that treat that as inevitable — and move their value up the stack before the squeeze forces it — are the ones still standing.
Bibliography
IAB Tech Lab, "Agentic Advertising and AI Initiatives." Link
The Current, "IAB Tech Lab, not a fan of AdCP, releases an agentic advertising framework" (2026). Link
PPC Land, "IAB Tech Lab names its agentic ad initiative AAMP to end market confusion" (26 Feb 2026). Link
VideoWeek, "WPP Rejects HoldCo Label in New AI-Driven Strategy" (26 Feb 2026). Link
eMarketer, "Omnicom's latest report signals an agency shift from human talent to AI" (citing ANA, Gartner). Link
TensorOps, "Agentic AI in Advertising: A 2026 Field Guide" (21 May 2026). Link
The Drum, "As Omnicom becomes the biggest, Publicis wants to be the 'MVP' of holdcos" (2026). Link