By TGM Team · Fri Oct 02 2026 · 7 min read
Agentic advertising has two rival standards, AdCP and AAMP, racing to govern how AI agents transact. Here's what the standoff means for publishers and SSPs.
By late 2026, agentic advertising has stopped being a thoughtful experiment.
We state this with firm understanding because according to the IAB's 2026 Outlook Study, 96% of buyers are now aware of agentic AI ad buying, with roughly two-thirds of them actively focused on it for campaign execution specifically.
The real budget is already moving through AI agents, and the question that actually matters now isn't whether that continues, it's whose rulebook the agents use to do it.
That's not an abstract question from where we sit. TGM operates on the sell side of exactly this transaction as an SSP, and directly through Swift TV'sown FAST inventory, our infrastructure is one of the things an agent will eventually need to discover, evaluate and transact against, under whichever protocol ends up carrying real volume.
We've been tracking this standard question closely for that reason, and it's worth writing about publicly because almost none of the coverage so far has come from the sell side. The early evidence that this is real, not hype, is already public: PubMatic launched AgenticOS at CES in January, after piloting a live agentic campaign the previous month.
That's one data point among a fast-growing list, agencies, DSPs and platforms across the industry have moved in the same direction through the year, but it captures the shift well: agent-to-agent negotiation moving from pilot to production inside a single quarter.
What's had far less coverage is the question that actually determines how this plays out for publishers, not "will AI agents buy ad inventory," but "whose rulebook will they use to do it."
Right now, there are two competing answers, built on genuinely different philosophies, and almost nobody has written about what either one means from the sell side.
AdCP (Ad Context Protocol) is an open standard published by AgenticAdvertising.org, released under an Apache 2.0, built on Anthropic's Model Context Protocol, operates at the campaign, management layer, above real-time bidding infrastructure, defining how AI agents discover, compare, plan, buy, sell and measure advertising.
Its defining design choice is that it runs asynchronously: an agent's request doesn't need a millisecond response, so human-in-the-loop approval is architecturally built in rather than bolted on afterward.
In practice, an advertiser's agent might send something close to a plain language brief; publisher and platform agents respond with inventory packages that fit, and a human can approve before anything transacts.
AAMP (Agentic Advertising Management Protocols) is IAB Tech Lab's answer, and it starts from the opposite premise: don't replace the programmatic stack, agentify it.
Anthony Katsur the CEO of IAB Tech Lab has been publicly skeptical of AdCP's from-scratch approach, arguing the industry doesn't need another standards body reinventing infrastructure that already handles billions of transactions reliably.
AAMP builds directly on top of existing IAB Tech Lab standards, OpenRTB, AdCOM, OpenDirect, VAST, rather than replacing them, and is organized into three pillars: Agentic Foundations, Agentic Protocols, and Trust & Transparency.
IAB Tech Lab named the initiative on February 26, 2026 at the Beet Retreat event, with the fuller framework following in the weeks after.
Adnami CPO Thomas From has described the split as a straightforward "burn it down" versus "build on top" tension, with AdCP betting agentic trading is a genuine paradigm shift and IAB Tech Lab betting it's an incremental evolution of infrastructure that already has overwhelming scale and network effects behind it.
Neither side disputes that agents are coming. They disagree about whether the plumbing needs to be replaced or upgraded, and neither side has stopped moving since the split first became public.
IAB Tech Lab shipped AAMP 2.3 in late July, adding enterprise-grade deployment and pricing-integrity features; AdCP has kept shipping versioned releases in parallel across the same stretch. Whichever protocol you're watching, treat this as a live, ongoing build-out, not a settled debate, the gap between "which one got more press" and "which one is actually carrying volume" is still very much open.
Here's the constraint that makes this more than a branding exercise: programmatic auctions have to resolve in well under a second, and a live large language model call is nowhere near fast enough to sit inside that loop.
AAMP's execution layer, the Agentic Real-Time Framework (ARTF), was released for public comment in November 2025 specifically to solve this, not by making agents faster at thinking, but by moving execution into containerized, co-located environments that keep the familiar real-time-bidding auction mechanics while dramatically cutting the latency of agent-driven decisions.
IAB Tech Lab has stated a goal of reducing real-time bidding response latency by roughly 80% through this approach.
AdCP sidesteps the problem differently: it simply doesn't try to operate inside the millisecond auction window at all, keeping agent negotiation at a slower, asynchronous layer above it, and leaving OpenRTB-style execution to do what it already does.
There's a trust layer forming alongside both efforts, too. IAB Tech Lab launched an Agent Registry on March 1, 2026, letting companies register and verify their AI agents, an identity and accountability layer for a world where agents transact directly with each other rather than through a human trader.
LiveRamp separately donated its User Context Protocol (UCP) to IAB Tech Lab, an open standard for how agents exchange identity, contextual and reinforcement signals with each other; that work has since been folded into IAB Tech Lab's broader Agentic Audiences Framework.
Whichever transaction protocol ultimately wins more adoption, something resembling this trust and signal-exchange layer has to exist underneath it, agents can't negotiate in good faith with counterparties they can't verify.
It's tempting to treat this as a buy-side problem to watch from the sidelines. That's a mistake, for three reasons specific to the sell side.
Those are different engineering problems, and a publisher that's only prepared for one of them is betting on an outcome that hasn't actually been decided yet.
US programmatic spend is forecast to exceed $220 billion in 2026, and walled gardens have long captured the large majority of that spend, a share most forecasts don't expect to shrink quickly even as open-web agentic tooling expands.
Independent publishers and SSPs have a shrinking window to prove their inventory is worth an agent's attention before that share concentrates further. It's the same pressure we watch closely on Swift TV's side of the business, not just TGM's SSP side.
Publishers building direct, hard-coded integrations to either protocol today are taking on technical debt that a genuinely competitive standards race makes unusually risky. It's the same posture TGM is taking with its own roadmap, build the abstraction layer first, commit to a protocol allegiance later, once volume rather than press coverage tells you which one to bet on.
This is the single highest-leverage thing a publisher can do right now, and it's the first thing we'd recommend to any publisher asking us where to start.
Agentic advertising coverage so far has mostly asked whether AI agents will buy and sell ad inventory. The more useful question, especially if you're on the sell side, is which rulebook they'll use to do it, and right now that's genuinely, publicly unresolved between two well-funded, differently, motivated efforts.
The publishers who come out ahead won't be the ones who guessed the winner first. They'll be the ones who built inventory, identity and integration infrastructure flexible enough not to have needed to guess, which is exactly the bet TGM is making with its own roadmap, and with Swift TV's inventory, while this race plays out.
Q1. What's the difference between AdCP and AAMP in one sentence?
AdCP is a from-scratch, MIT-licensed protocol for agent-to-agent negotiation built on Anthropic's MCP; AAMP is IAB Tech Lab's framework for extending existing standards like OpenRTB to support agentic execution instead of replacing them.
Q2. Do publishers have to pick one protocol now?
No — and the most credible analysis of this standoff recommends against it specifically. Building an abstraction layer that can support either is the safer near-term move than committing to one.
Q3. Is agentic advertising actually happening yet, or is this still theoretical?
Both, depending on the layer. Real campaigns (PubMatic's AgenticOS pilots, for instance) are already running; the standards that would let this work consistently across the whole open web are still being built and contested.
Q4. Does this affect CTV and FAST inventory the same way as web display?
The underlying protocol question is the same, but CTV's real-time bidding environment already runs on tighter latency and less mature identity infrastructure than the open web — which makes the ARTF-style execution problem and the UCP-style identity problem both more acute for CTV/FAST publishers specifically, not less.
