Part Three: A Wake-Up Call for the Whole Ecosystem.
The Agentic AI Asymmetry Dilemma.
By Stuart Bailey | Part Three of Three
Originally published on 6th August 2026 in [B&T]
In Part One I introduced the Agentic AI Asymmetry Dilemma: the structural gap between agencies building proprietary AI and publishers optimising for systems they can't see. In Part Two I turned to advertisers, arguing the same asymmetry runs through that relationship, and that cost cutting is the wrong response to a capability problem. This piece asks the obvious next question: what happens when all three parties are in the room together?
The answer isn't technology. It's governance. The industry is running out of time to choose its own architecture before one gets chosen for it.
The Data Has Made Up Its Mind
The scale is no longer in question. Two-thirds of advertisers are now focused on agentic AI for ad buying, and token volumes, the raw measure of AI workload, are growing fourteen times annually, driven almost entirely by agentic use cases (IAB 2026 Outlook Study; Exponential View).
It's a step change in scale and autonomy, arriving faster than the governance built to manage it: BCG's AI at Work survey found agent integration into workflows has more than doubled in a year, while half of employees say their organisation still has no clear guidance for managing human-AI teams.
The cost of that gap is already visible. CMOs now put 71% of budget into short-term performance media, nearly the inverse of the 60/40 brand-to-activation split the IPA's thirty-year effectiveness dataset finds most profitable (Deloitte/AMA CMO Survey, 2025), and marketers over-indexing on short-term gains may be forfeiting as much as half their potential return (Ipsos MMA, for Google and WARC).
The industry isn't just building AI without shared governance. It's defunding the work that builds durable growth, at the same time.
The infrastructure of the agentic era is already under construction; the industry just hasn't agreed on whose. IAB Tech Lab, the incumbent standards body, has launched AAMP (Agentic Advertising Management Protocols), extending the existing rails that already power programmatic advertising rather than building new ones.
At the same time, a separate, independent effort, AgenticAdvertising.org, has emerged with its own competing protocol, AdCP, built from outside the incumbent system entirely — fronted, notably, by the IAB's own former long-time CEO.
Two credible governance efforts, launched within months of each other, each claiming the job. That's not an industry that has solved this. It's an industry still arguing about it. Which means the rules aren't written yet.
That was true when I wrote this article. Then, on 28 July, before the article was published, the Model Context Protocol (MCP), the underlying plumbing that both these efforts are built on, announced its biggest update since launch. That is how fast this is moving.
Two things in it matter to us. It now runs on standard web infrastructure, which makes this kind of shared access far cheaper for everyone to stand up. And it can track a single request all the way through the system, so you can see who asked for what, and when.
That is the accountability everyone says they want before they share anything. It also breaks some existing builds, so anyone running an AdCP pilot has work ahead of them. This is the rules being written, in real time, while most of our market isn't watching.
The Governance Vacuum Nobody Is Talking About
Most of this industry's AI conversation is about tools: which models, which platforms, which agencies have the sharpest capability. BCG's research says that's the wrong fixation. Employees with clear AI strategy but limited tools outperform those with strong tools and no strategic direction by 25 percentage points on measurable business impact. Strategy beats tools. Clarity beats sophistication.
This is a governance problem, not a technology one. Agencies are building autonomous systems that plan, recommend and activate media using logic that advertisers and publishers can't see, against criteria that haven't been shared. Gartner predicts more than 40% of agentic AI projects will be cancelled by 2027 due to unclear value and weak governance. Sophistication doesn't fix bad inputs. It amplifies them.
We've Done This Before
This isn't the first time the industry has had to build shared infrastructure it didn't fully trust yet.
The Australian Digital Advertising Practices and the standards that made programmatic trading possible both required competing parties to agree on common ground, because the absence of standards cost everyone more than the compromise required to build them. We're in the same testing phase now, before the patterns calcify. The window is open. It won't stay open indefinitely.
The Interconnected Advantage
The tempting reading of this moment is that the winners will be whoever has the most sophisticated standalone AI. That's the wrong frame. High-AI-intensity firms are growing revenue 92 percentage points faster than firms with no AI spend (Exponential View), but the mechanism isn't the tool. It's what the tool is connected to.
Agentic allocation only scales if the market converges on shared standards for products, identity and permissions, so agents can transact across environments without rebuilding integrations company to company.
The winners won't be the agencies, publishers or advertisers with the smartest standalone AI.
They'll be the ones with the best connections across the ecosystem. We've spent twenty years in media building moats. The next twenty will reward bridges.
What Each Party Needs to Do
The solution isn't complicated. It's uncomfortable, because it asks each party to give something up.
Agencies need to share the criteria, not the architecture: publishers and advertisers don't need proprietary systems, just to know what signals those systems weight, so they can participate fairly.
Publishers need the infrastructure to participate on equal terms, giving sales teams real-time access to performance data rather than waiting for agencies to explain the system.
Advertisers need to become intelligent co-pilots: not builders of AI systems but active participants in the ones being built on their behalf, understanding what the system optimises toward and when to redirect it.
You don't need to build the engine. But you do need to learn to drive.
And the industry needs minimum transparency standards, so buyer and seller agents share a common foundation for faster, better decisions.
The Institutional Memory Problem
One part of the governance problem rarely gets named directly: the industry loses knowledge every time someone leaves. Every relationship that resets with agency churn, every brief that starts from scratch, every onboarding that rebuilds context an agency already had.
In an agentic system, that loss degrades the quality of the AI's decisions: systems trained on incomplete, discontinuous data don't just underperform, they confidently underperform.
The fix isn't another training programme. It's building systems that hold what every engagement learns, persist through personnel change, and make that intelligence queryable on demand. The governance layer this industry needs isn't only a set of standards. It's a shared memory.
The Choice Is Still Ours
In Part One I drew the parallel to the walled gardens: a decade spent reacting to a system that accumulated information asymmetry at scale, and we're still dealing with the consequences.
The Agentic AI Asymmetry Dilemma is forming right now, and we're early enough to choose a different architecture.
Agentic AI can become a trusted repository that supports better decisions and fairer markets across the ecosystem. Or it can become another false prophet: confident outputs from flawed inputs, value concentrated in whoever built it first, and another decade of lobbying to fix what we chose not to prevent.
The difference isn't the technology. It's the choices we make now.
Australia has a small, connected ecosystem and a genuine history of building collaborative industry standards. We've done it before. We can do it again, and show the rest of the world what a collaborative AI ecosystem actually looks like.
The winners in this next era won't be the smartest. They'll be the ones who collaborated best.
We've always known how to do that in this industry.
The question is whether we choose to do it again; and whether we can afford not to.
Read Part One: [Catching Publishers Sleeping] · Read Part Two: [The New Thing Keeping Advertisers Awake]
