What happened to agentic media buying?

When MCP (Model Context Protocol) and AdCP (Advertising Context Protocol) first showed up in my feed last fall, I got swept up in it like everyone else.

I envisioned a world where I could type a plain language brief into a chat window and watch a media plan assemble itself in seconds — no deck, no spreadsheet, no three-hour meeting to approve it.

I wrote about the death of the media plan back in April. I still think that future will come to pass, but it might take a bit longer than we thought while we were caught up in the excitement of it all.  That’s probably a good thing.

Things feel different right now. The chatter is quiet. The panels tackling agentic “this-and-that” have settled down, or at least are less publicised than they were. The LinkedIn posts about the agent-to-agent future slowed down.

I first assumed the hype had simply run its course and not lived up to the expectations.  After diving in a bit, I think the conversation hasn’t faded away, but has definitely shifted to a new place.

Real campaigns

AdCP, the open protocol built on Anthropic’s MCP framework, reached general availability and a 3.0 release earlier this year.

Real campaigns have run on it, not demos. Butler/Till and PubMatic executed one of the first fully autonomous, end-to-end agentic campaigns built from a natural-language brief and optimised in real time, with no human touching a single line item.

NBCUniversal, FreeWheel, and RPA used agents to automate a chunk of NFL playoff inventory. Omnicom’s CTO confirmed on an earnings call that the agency is running live agent-to-agent buys for multiple clients. There does seem to be adoption, but the conversation is not as hot as it once was.

What went quiet is the marketing of the products. Underneath the surface, AdCP and the Interactive Advertising Bureau Tech Lab’s competing framework, AAMP, are locked in a genuine standards brawl. The Tech Lab has called AdCP deeply flawed. AdCP’s backers think the Tech Lab is protecting its old turf.

Backroom debate

A backroom debate is far less interesting than test after test and press releases talking about them. The conversation became a governance and compliance debate and moved from the main stage to committee, and committees do not make for exciting articles in MediaPost or anywhere else.

After doing a little homework, I’ve come to believe that agentic buying can be real and will matter. It will probably be incremental, not the overnight replacement of the media plan I half expected when I first read about it.

That aligns with what I’m seeing on my own team with AI more broadly. Nobody on my team has been replaced by an agent. What has happened is that the work gets done a little faster, the first draft of a plan shows up a little more quickly, and with more initial detail.  The people doing the work have more time to think about the parts that actually require judgement.

An iterative approach

That is not a revolutionary approach.  It is an iterative one. It is an improvement, and I will take improvements of any kind.

Agentic advertising is indeed “inevitable, but incremental” as someone recently put it. I’d add another word: Inevitable, incremental, and under appreciated, because there are people doing real work to bring this to bear, but the conversation and debate are happening in rooms where no one is watching, and therefore fewer people are posting about it.

They’re quietly making their teams a little better every week, letting the standards fight play out in the background while they keep buying media.

That’s usually how the real transformations happen — not with a bang, but with a bunch of small, unglamorous wins that nobody bothered to write a column about.