The bots are buying and selling the ads!
Wait, I thought we hated "bots."
Yeah, but not these bots. We call these bots: agents.
The ANA says that our average programmatic campaign runs on something like 44,000 websites. I think that number is a bit low in reality, but let's go with it.
That's way, way too many.
Spraying a buy across more sites than you could possibly name is not a strategy.
This issue, among hundreds of others, is what AdCP, or Ad Context Protocol, aims to fix. It was launched last month (October 2025) and intends to formalize how buyer agents, seller agents, and creative agents work together to complete an advertising transaction.
If successful, publishers should earn more revenue per impression because the buying path can be more direct. Advertisers should get access to better inventory with greater transparency at a fair price.
That's the hope, anyway.
What is AdCP?
The working group's website says that we should think about AdCP as the "USB-C of ad buying."
It is the protocol, language, and methodology for which AI agents talk to eachother to get things done in the ad ecosystem regardless of channel.
Instead of having to log into 40+ different platforms, exporting a boat load of spreadsheats, manually merging them, crafting targeting groups and waiting days (weeks) for approvals…AI agents will handle it. End-to-end.
Here's what that looks like:
"Find outdoor adventure enthusiasts with high purchase intent for outerwear, compare prices across platforms, then activate the best options to drive eComm sales."
^^ That's the input the buyer gives to their buyer's AI agent.
From there, the agent reaches out to all available seller AI agents to initiate the process. Together, the agents handle all the discovery, value comparison, and campaign activation. Along the way, humans may be asked to weigh in on certain decisions, but the bulk of the work is autonomous.
This is AdCP at work.
Why does this matter? Today, your in-house or outsourced teams spend far too much time focused on platform gamesmanship and not enough time on actual strategy. AdCP would flip that, enabling humans to spend time on what they are best at.
What is AdCP going to solve?
The short answer: lots of stuff. Actually, so much stuff that we cannot even predict what it will be just yet. That is because AdCP is an open-source project that anyone can pick up and build with. Consequently, the use cases are limitless.
But let's start with one. Waste.
As our industry has belabored for years, only 36 cents of every dollar reaches the end publisher. Transaction fees eat up the rest, take rates, etc. And a bit of that money seems to vanish into the ether, too.
These integration costs, for lack of a better term, are killing you. They're weighing down your investment in a way that hamstrings your campaign's success before it even begins.
Streamlining that workflow is the key to success in AdCP.
Today, you have junior humans on your team doing this stuff:
- Logging into platforms.
- Search for the audiences that sound like the ones you need.
- Export your findings into Excel.
- Recreate the targeting parameters based on your findings.
- Kick that up for approvals.
- Wait for approvals.
- Wait. For. Approvals.
- Did legal see this?
- WaIt fOr aPpROvals.
- Repeat for the other platforms you may want to use.
By the way, this is what "diverse client experience" adds up to these days.
With AdCP, unified protocols are established that allow AI agents to handle all of the above. The spec aims to standardize processes, communication, and replace the chaos.
How did we get here?
There are essentially four major developments in the market that have given rise to the opportunity AdCP presents.
Cookie deprecation forced everyone's hand. Even though we continue to live through the long saga of the cookie's final death, many ecosystem players have moved on. Safari and Firefox already booted third party cookies. The sell side went absolutely ham with "curation" being the new hotness. As a result, AdCP assumes that cookies will not be essential from day one.
Google's antitrust drama opened a window. If you assume that there will be some shakeup in the near future, Google's response to its monopoly position is an opportunity for the emergence of new methodologies. AdCP is open-source and vendor-neutral.
AI as a whole works well enough to handle…humans. Or rather, it works well enough to handle significant workload components currently required in the advertising ecosystem. Thus, agent-to-agent advertising is born.
The agencies see the writing on the wall. They've already invested over a billion dollars in AI and laid off 15,000 staffers. They know it is going to transform their workflows.
Good news
Bringing buyers and sellers closer together is a win-win. In my opinion, we've added an insane number of intermediaries into every transaction and destroyed a lot of value along the way. We do not need 6 to 12 middlemen involved to sell a single impression.
I want to see more revenue making it to content creators, not the aggregators that move money around.
The transparency we've droned on about ad nauseam is baked into the core architecture of AdCP. Perhaps we will finally have the audit trails and standardized reporting protocols we've been begging for.
Your AI buyer's agent will be able to negotiate better media deals. It will also adhere to your governance frameworks more effectively than a human would, regardless of how detailed they are. Set the rules, define the constraints, approve the spend, then let the agents do their work.
I know. It sounds like we are trying to get people out of the workflow. What we are really trying to do is have humans focus on what they do best. Humans are in the loop.
We want human judgment without human bottlenecks.
The catch
Guys, it is going to be hard. It is going to take work. Yes, this will require more effort initially. But only in the beginning.
There's absolutely an initial hump that everyone adopting an agentic future will need to get over. However, once you do, the gains are limitless.
All of that initial effort will be around crafting and codifying your explicit governance frameworks. The dos and don'ts for your agents and any downstream partner. That includes defining when a human needs to be brought into the loop or when the agent can act fully autonomously on your behalf.
Implementation, in the early days, will be more difficult than the initial press releases suggest, but that's always the case. That should not mean a default "wait-and-see" approach for your brand.
As I always say, today will be the hardest it will be. Every week, things get more streamlined, implementations multiply, deployments improve, debugging turnarounds improve, and we early adopters document the learnings.
According to industry analysts, the critical mass for AdCP may be 20-30%. We have the classic ad tech chicken-and-egg scenario on our hands. Advertisers won't build their buying agents until publishers make inventory available through the protocol. Publishers won't bother building their seller agents until there is some demonstrated demand. And monopolistic walled gardens have zero incentive to participate, so they'll join the party late.
Luckily, we do have a bit of an early groundswell with multiple buy-and-sell site parties willing to take a leap of faith.
Which advertisers are best suited for AdCP?
If you are spending $50M+ in paid advertising
Make AdCP evaluation a key agenda item for 2026. Commit to participating in the working groups. AdCP is not yet production-ready. Founding members are conducting pilots, and could use your help.
Ask your agency about their AdCP strategy. They'll either light up with excitement because they've been wanting to experiment with it, or they will squirm because they have not been paying attention.
Start thinking about agent governance now. Who approves what (humans or bots)? When? Under what conditions? What are the escalation paths when a human needs to step in and make a decision?
Start building your governance frameworks and iterating now. You'll be ahead of the competition.
If you are spending less than $50M
Follow along. You probably do not have the internal resources to support this level of effort. Let the larger marketers fund the early learning curves, but do follow along from the sidelines.
Everyone
Everyone should be questioning every "AI-powered" solution pitched to them. Internally and from external solutions providers. Have a strong bias for open source or highly transparent models. Begin asking now if your vendors have plans to explore AdCP.
Remember that just because a platform offers transparency, it does not necessarily mean that it is actually delivered. No, transparency depends on implementation, contract terms, and your willingness to demand it continually.
The rules for agentic advertising do not exist because they have not been written.
Yet.
