Alanna Laforet takes a look into why questions around agentic AI are sidestepping the one that actually keeps CMOs up at night...

An industry obsessed with agent-to-agent infrastructure is skipping the harder question: for what should the agents actually be optimising?

The last six months have produced two of the most ambitious infrastructure initiatives in recent ad tech memory. In October 2025, a coalition of more than twenty companies (including Yahoo, PubMatic, Scope3, and Magnite) launched the Ad Context Protocol (AdCP): an open standard enabling AI agents to declare, brief, negotiate, and transact programmatically, without human intermediaries. Shortly after, the IAB Tech Lab released its Agentic RTB Framework (ARTF), a containerised single server architecture for the buy and sell side. Instead of a DSP and SSP passing data back and forth across separate servers, containerisation lets one party deploy their code directly inside the other's infrastructure. In an agentic context, this means an AI agent can execute decisions (bidding logic, enrichment, targeting) locally where the data lives, rather than shuttling information across the internet and back. 

Both are genuinely impressive. Both represent serious engineering bets by serious players. And both are solving for speed, scale, and interoperability while largely sidestepping the question that actually keeps CMOs up at night.

What happens to my brand?

Look more closely at who is actually in the room, and the gap becomes hard to ignore. AdCP's founding and launch coalition is predominantly supply-side: SSPs, publishers, data companies, contextual vendors, and streaming platforms. The buy side is thin. The major agency holding companies (WPP, Publicis, IPG, Omnicom, Dentsu, Havas) are conspicuously absent, as are the large brand advertisers whose media budgets these protocols are ultimately designed to move. The companies helping to define how AI agents will negotiate and transact on behalf of advertisers are, in large part, not the advertisers or their direct representatives. That is not a minor footnote. It is a structural imbalance that will shape what the protocol optimises for, and whose interests it is most fluent in expressing.

ARTF, developed under the IAB Tech Lab, has drawn broader institutional support, including Netflix, Paramount, The Trade Desk, and Yahoo, but the pattern holds: the engineering coalition is infrastructure-first, and the voices of brand marketers remain largely advisory at best. When the people who write the specs don't include the people writing the checks, the resulting standards tend to speak the language of delivery efficiency rather than brand stewardship.

The protocol that doesn’t know what a brand is

AdCP is best understood as plumbing. It gives AI agents a common language to communicate across platforms, doing for the emerging agentic layer what OpenRTB did for real-time bidding a decade ago. To be clear, AdCP does not replace OpenRTB, rather just complements it. In theory, an agent representing a buyer can brief an agent representing a publisher, negotiate terms, and execute a transaction without a human touching the keyboard. ARTF reimagines the communication flow, containerising the architecture of programmatic trading itself so that one company’s code can be deployed as an agent of the entire system, catching problems earlier and faster than any human workflow could.

These are not incremental improvements. They are genuine infrastructure shifts, and the industry is right to take them seriously.

But infrastructure is neutral. It moves whatever you put into it faster and at greater scale. And that is precisely where the problem begins.

The performance trap: When the algorithm is right and the brand is wrong

Performance metrics are easy for an agentic system to optimise. Click-through rates, ROAS, completion rates, conversion signals: these are numbers. Agents are very good at chasing numbers.

Brand is not a number.

Consider a scenario the industry is not stress-testing nearly enough: an unsavoury site, one that a human media planner would flag and block without hesitation, performs exceptionally well on every metric an agent has been instructed to optimise. Strong completion rates. High CTR. Low CPMs. Clean conversion data. No signals that trip the taxonomy filters.

An agentic system with a performance mandate will find that site, favour it, and scale spend against it. Because by every measurable signal, it is the right call.

But your client’s brand just ran at scale next to content that would have been caught in any human-reviewed media plan. And when the client calls to ask how it happened, the answer (the agent optimised toward the brief) is both technically accurate and completely useless.

The challenge isn’t that agents are reckless. It’s that they’re precise, and brand suitability is inherently imprecise.

Brand suitability lives in judgment calls, editorial context, audience perception, and reputational heuristics that resist clean quantification. The IAB Tech Lab has updated its Content Taxonomy to address inefficiencies in content classification, and those taxonomies will integrate with agentic layers. But taxonomies classify content. They do not understand brand identity, brand values, or the reputational cost of an association that is technically "safe" but strategically corrosive.

And critically: who is accountable when an agent makes a brand-damaging buy that was technically within spec? This question has no clean answer today. Not from AdCP. Not from ARTF. Not from any of the frameworks currently being drafted.

We’ve been here before

The industry’s impulse to fix brand safety with better signals is understandable. More granular taxonomies, improved contextual scoring, richer pre-bid data: all of this helps at the margins. But it is still the wrong frame.

We did exactly this with programmatic. The promise was efficiency and precision. What we got was efficient delivery of the wrong ads to the wrong environments at enormous scale, followed by years of brand safety scandals, MFA site proliferation, and advertiser trust erosion that the industry is still recovering from.

The pattern with agentic is identical, only faster. A new capability emerges. Vendors build it into their pitch. Agencies add it to RFPs. Everyone chases the feature instead of the outcome. The IAB Tech Lab and the AdCP coalition are not the villains here; they are building necessary infrastructure. The failure is not theirs. The failure is the industry’s collective refusal to do the harder strategic work before turning the agents on.

Open standards only work when everyone builds on them with intention. The real work of defining objectives, validating partner compliance, establishing human oversight, and agreeing on what brand means as an operational input has to happen before the agents go live. Right now, almost no one has done it.

What the stack gets wrong about strategy

The positioning problem runs through every layer of the supply chain, and the agentic push is accelerating it.

Agencies are being pressured to implement agentic workflows without a clear brief on what success looks like beyond performance KPIs. They are fielding client questions about agentic brand safety they cannot answer, because neither they nor their vendor partners have defined the terms. They are using AI to police clean rooms without using the appropriate guidelines. 

All this pressure, and no seat at the table where the standards are being written. The holding companies, the organisations that collectively manage hundreds of billions of dollars in annual media spend and theoretically serve as the most powerful advocates for brand advertiser interests, were not founding members of AdCP. They were not the architects of ARTF. They are largely being handed protocols developed by the infrastructure layer and asked to integrate, rather than being co-authors of what those protocols should require. That inversion matters. When the buy side is a consumer of standards rather than a co-creator, the standards will reflect supplier priorities: fill rates, auction efficiency, agent-to-agent throughput. Not brand integrity, not advertiser accountability, not the reputational cost of a misplaced impression.

Large brand advertisers, the Unilevers, the P&Gs, the financial services and pharmaceutical companies that have historically driven brand safety policy, are similarly absent from these conversations at a meaningful level. Their procurement and marketing technology teams are not in the working groups. Their brand guidelines are not informing the schema. The result is a set of protocols that are technically sophisticated and strategically naive about what their most important end users actually need. This is not a new failure mode. The programmatic standards that enabled a decade of brand safety crises were also built without sufficient brand advertiser input. The industry is repeating the same structural mistake, only now with agents that will execute at a speed and scale that leaves far less room for course correction.

DSPs and SSPs are racing to build agentic capability into their roadmaps while their brand positioning (what actually makes them the right partner for a given advertiser) becomes increasingly indistinguishable from the competition. When every platform claims better reach, better data, and better ROAS, and now also claims agentic infrastructure, the differentiation collapses entirely. Price becomes the default selection criterion. That is bad for everyone.

Publishers are being told to make their inventory agent-readable, to game the answers LLMs deliver so they can appear first in the list. This inventory manipulation often means compressing the contextual richness that is their actual value proposition into a schema a machine can parse. The editorial environment, the audience relationship, the trust signals that premium publishers have spent years building: these are being flattened into taxonomy codes and bid floor signals while walled gardens continue to capture the brand budget that should be flowing their way.

Agentic tools will commoditise execution. Strategy and positioning cannot be automated.

What getting it right actually looks like

The companies positioned to win the agentic era are not the ones with the most sophisticated agent infrastructure. They are the ones who have done three things before turning the agents on.

First, define what brand means as an operational input, not just a mood board or a list of blocked keywords. Which environments signal brand alignment? Which adjacencies are disqualifying even when performance data is strong? This needs to be codified, tested, and agreed upon by both the agency and the client before an agent can be expected to apply it. If you cannot write it down clearly enough for a human to act on it, an agent will not be able to act on it either.

Second, build accountability into the agentic workflow, not bolted on after the fact. Every agent needs a defined purpose, a version history, and an audit trail that can answer three questions when something goes wrong: who changed what, when, and why. That is not a technology requirement. It is a governance requirement. It belongs in the strategy conversation, not the product roadmap.

Third, separate performance optimisation from brand stewardship and treat them as distinct mandates with distinct oversight. An agent can handle the former with genuine efficiency. The latter still needs a human with context, accountability, and a point of view: someone who can look at a site that is performing well and say, with authority, "we are not running there". Equally, the same can be said for clean room data. There needs to be a checksum to make sure the data used is not stolen, corrupted, or unsavoury.

The question nobody is asking

AdCP and ARTF are impressive answers to a real question: how do agents communicate with each other efficiently across a fragmented ecosystem? The engineering is sound. The coalition-building is impressive. The ambition is genuine.

What nobody has answered yet is the harder one.

What should the agents be optimising toward, and who is responsible when the algorithm gets it technically right but strategically wrong?

The companies that will win the next phase of ad tech are not the ones with the most integrations or the most agent-to-agent transactions per second. They are the ones whose buyers, sellers, and partners can each answer, in one sentence and without hesitation, why they made the choices they made. Strategy first. Infrastructure second.

The agents are ready. The industry is not. Stop the hype cycle and start thinking. 

About the Author

Alanna Laforet is a media and ad tech executive with over a decade of experience across publishing, programmatic, and emerging technology. She is CEO of Laforet Productions, a boutique consulting firm focused on the intersection of technology and humanity, with an emphasis on storytelling, market mechanics, tokenisation, ad tech systems, artificial intelligence, and human attention.