AdCP: the standard that wants to change what you buy, not just how fast

In adtech, the technology is almost never the problem. The operational layer always is. This series is about that gap. One buzzword at a time.

Picture the buyer running a 50K campaign across three publishers. Last time, it took two weeks. Four logins, four dashboards, four proposals that never came back in the same format, and a spreadsheet built by hand just to compare them. You know this job. Most of it isn't buying. It's reconciling.

That grind is what AdCP is trying to kill. And once you see what it wants to replace it with, you realize the interesting part isn't the speed. It's that it quietly changes what you'd be allowed to buy.

What everyone's saying

"AdCP is how agents will buy media." "It's the OpenRTB of the AI era." "Built on MCP, backed by the big SSPs, it's the new layer." You'll hear it in the same breath as the agentic hype, with the usual implication that it's already here and you're already behind. The reality is more interesting than the slogan, and worth ten minutes of your attention.

What it actually is

AdCP, the Ad Context Protocol, is an open standard launched in October 2025. It runs over MCP, the connection layer we took apart last time, and other agent-to-agent transports. If MCP is the universal adapter that lets agents reach platforms, AdCP is the specific language they speak on top of it to do one thing: trade media. MCP (and its cousins) carry the conversation. AdCP defines what the conversation is about.

Start with the analogy that does the heavy lifting, because the industry reached for it immediately: AdCP is "OpenRTB for the AI era." OpenRTB is the standard that made programmatic possible, the shared language that let machines auction a single ad impression in milliseconds, at scale. It gave everyone one way to speak that. AdCP wants to be that shared language for the next layer up, not auctioning impressions, but negotiating whole deals between a buyer's agent and a seller's agent.

In practice, the promise is that our buyer stops logging into four systems. They'd brief once, in plain language, and competing sellers' agents would answer in the same structured format: here's the inventory, here's the audience, here's the price. The four-dashboard reconciliation job would collapse into one conversation. That's the part you'd feel on Monday morning, if it works.

Why it matters when you're the one spending the budget

Here's the part most explainers skip, and it's the part actually worth learning.

The speed grabs the headlines. The real change is in what you can transact on. Today, programmatic mostly means bidding on impressions: you buy the chance to show one ad to one person, millions of times. AdCP is built so agents negotiate terms instead, which opens the door to buying on different units entirely, an audience segment, an engagement rate, even an outcome like brand lift. Read that again, because it's the whole point: this isn't just a faster route to the same impression-based buy. It's the possibility of buying the result rather than the inventory.

That's a genuine shift, and it's why this is more than hype-with-a-protocol. If you can transact on "eco-conscious car buyers who actually engaged" instead of "a million impressions that hopefully reached them," the thing you're purchasing is closer to the thing you actually want.

Now the second thing worth understanding, because it confuses people: AdCP is deliberately slow. RTB clears an impression in under 100 milliseconds, a reflex. AdCP runs over seconds, sometimes days. That sounds broken until you see why. RTB automates an auction, a split-second decision. AdCP automates a negotiation, the kind where terms get discussed, a human might approve, and the output isn't a bid but a packaged deal that holds for a campaign. The slowness isn't a flaw. It's what negotiating, rather than bidding, actually requires.

So the operational read: AdCP doesn't touch the millisecond machinery you already run. It targets the part that still runs on email, logins, and patience, the strategic, slow, human part. That's both its promise and, as we'll see, its problem.

And here's the boundary that decides how much of your world this touches: AdCP is an open-web standard. Google, Meta, and Amazon are building their own agentic buying inside their own systems, and they have no reason to speak someone else's protocol. So whatever AdCP becomes, it becomes on the open web, not inside the walled gardens where a large share of budget already sits. That's just where the line falls, and it tells you which part of your spend this conversation is actually about.

What nobody selling you AdCP will admit

Two things the keynote leaves out.

First, the one that sounds like good news but isn't, on its own: a faster, smarter path to inventory is not better inventory. AdCP compresses how you brief, source, and transact. It has no opinion on whether what you're buying is any good. Point this elegant process at a thin, poorly-curated, fraud-adjacent supply pool and you've built a more sophisticated way to spend money on bad media. The protocol moves the deal. Judging the deal is still your job.

Second, the deeper one, and this is the part that makes you smarter than the pitch. The thing most likely to kill AdCP isn't technical. It's human. The veterans who've watched this movie point out that media-buying automation has stalled before for one stubborn reason: the incentives don't line up. Buyers don't want to be "price takers" who accept whatever rate is published. Sellers don't want to reveal their real rates at all. One person who tried to build essentially this at Google years ago ran into "hostility and indifference from both sides." A protocol is a technical answer to what is fundamentally a business standoff, and no schema, however elegant, forces two parties to share what they'd rather hide.

Notice the tell: the parts of AdCP almost everyone agrees are useful are creative handling and audience signals, the cooperative tasks. The contested part is media buying itself, the adversarial one. That split tells you exactly where the technology ends and the human problem begins.

So the stance that makes you look smart in the room isn't excitement or dismissal. It's this: AdCP is a real, well-built attempt at a real bottleneck, and its fate rests on adoption and incentives, not engineering. Watch who actually implements it, not who announces support. The demo will always look clean.

The test to carry into any meeting where it comes up: does this change what I can buy, or only how fast I can buy it? One reshapes your strategy. The other just speeds up your Monday. They carry very different price tags, and the pitch will charge you for the second as if it were the first.

The real question nobody's asking

Here's where I'll be honest, because that's the point of this series.

AdCP is built to make the transaction better, faster, smarter, broader in what it can express. And that's real progress. But watch the assumption underneath: that the hard part of media buying is the transacting. For a lot of buyers, it isn't. The hard part is knowing which inventory is actually worth the money, across a fragmented supply landscape that no protocol maps for you.

AdCP makes the conversation between systems richer. It doesn't make the supply underneath that conversation any clearer. And if you can now brief, negotiate, and buy outcomes at machine speed across a mess you still can't fully see, you don't automatically get better results. You get more sophisticated access to the same blind spots.

That's the pattern under every acronym in this series: in adtech, the technology is almost never the problem. The operational layer always is. AdCP is a genuine step, and a more interesting one than most. But the protocol is the part that gets built. Knowing what's worth buying, and seeing across all of it, is the part no standard hands you, and the part that decides whether any of this helps. Which is why the honest posture, the one the market itself has landed on, is this: more interesting than urgent. The demo is clean. The adoption is unproven. Both are true at once.

And AdCP is only one answer to the question the whole industry is now circling: who gets to decide how agents buy media, and in whose interest? A coalition of SSPs wrote one answer. The biggest standards body in the industry wrote a rival one, on the opposite philosophy. The walled gardens are quietly writing their own. That fight, not the spec sheet, is what shapes what you can buy in two years. It's where this series goes next.

---

Maxime Khalfallaoui, Supply Finder. I came up on the SSP side. Now I help agencies find their way through it.