Right now, every adtech company is rolling out an AI agent for buying, selling, measurement, you name it. It sounds amazing, but there’s a catch: Can they actually talk to each other?

For this to work, these agents need a “shared language” so they can finish a campaign from start to finish without getting lost in translation.

Lately, everyone is buzzing about AdCP (Ad Context Protocol) as the solution, but it’s also caused some serious head-spinning. People are panicking, asking: “Is this replacing RTB? Do I need an MCP? What is the IAB doing?”

If your brain feels a little melted, don’t worry. We’re going to take a step back, cut through the hype, and break down what’s actually game-changing.

To keep it on theme, let’s explain it the way we’d talk to a chatbot…one prompt at a time.

PROMPT: So what’s the big idea behind agentic transactions anyway?

The idea is that human ad buyers (advertisers, agencies, DSPs, etc.) can give natural language briefs with target audiences and goals to buyer agents, who will then build, negotiate, and execute campaigns based on interactions with specialized agents: seller agents, audience targeting agents, measurement agents, creative agents, etc. 

Basically, the goal is to truly automate all the different aspects of campaign setup and activation via agents. In the near term, there will still be a lot of humans involved to get the guardrails right.

PROMPT: Huh. OK, that makes sense, and I guess AdCP is key here. Can you explain what AdCP is to me, like I’m 3 years old?

AdCP is like a magic rulebook that helps robot friends in the ad world talk and play together super easily.*

PROMPT: Um… That’s not actually helpful. Why don’t you explain it to me like I’ve been in adtech for a while, and all this new jargon is making my head spin?

Sure. In spirit, AdCP is a lot like OpenRTB, just forget all that bidding and auction stuff for a moment.

In the early days of real-time bidding, transactions were handled via proprietary server-to-server connections. Every exchange had its own response/request schemas and mapping tables, and custom integrations were required when bringing on-demand or supply sources. To standardize and scale the nascent technology, the industry came together around the OpenRTB specification, enabling divergent tech to easily work together.

AdCP is meant to serve as a “universal ads API” to unify agentic advertising workflows across a campaign lifecycle, including campaign development, inventory discovery, activation, optimization, and even reporting. Like OpenRTB, it’s an open protocol that’s not owned by a specific company, but serves as a shared rulebook defining how different tech systems build, send, and interpret communications. This way, tech from different companies. In this case, AI agents performing a variety of functions can work with each other, the famed “interoperability.”

It’s easiest to think about AdCP as a communication layer where buyer, seller, and other types of agents come together in a unified interface (rather than scattershot integrations) and speak the same language regarding campaign details: briefs, inventory avails, activation, measurement, etc.

But let’s note another difference between OpenRTB and AdCP: while major SSPs like Magnite and PubMatic have built seller agents that leverage AdCP for communication, it is not the industry-accepted and widely adopted communication protocol for agentic transactions. Buyers and sellers could adopt other protocols or build and share proprietary ones.

PROMPT: I keep hearing about this MCP thing. What’s that gotta do with it?

MCP stands for Model Context Protocol; though it’s now open and open-source, it was developed by Anthropic of Claude fame to serve as a standard framework for connecting AI agents to various external systems, tools, and data sources – anything from databases of content to business management software. You can even think about MCP as a super-sophisticated, high-functional next-gen API.

While AdCP was built on similar principles to MCP, its functions are different. 

MCP is about connecting AI agents and systems/sources, while AdCP focuses on defining specific ad transaction automation tasks (e.g., brief delivery, inventory avails). 

If AdCP is the meeting room where ad-related agents communicate in a shared language, MCP is the express shuttle that brings them to that room. That’s why you’ll hear people say, “AdCP is built on top of the MCP.” 

PROMPT: Ah, this is starting to come together! Though I hear there are separate protocols inside AdCP. What?

Yes, AdCP has three internal protocols (with a fourth proposed) to enable more robust agentic transactions.

1. Media Buy Protocol. This is for agents to trade information on campaign briefs, inventory availability, activations, and performance monitoring.
2. Signals Activation Protocol. This is for agents to discover, activate, and manage first-party data and other audience signals, including leveraging “signals” agents –  specialists for audience targeting, identity resolution, etc.
3. Creative Protocol. This is for automating the generation and management of creatives, including specs and a standard formats library that extends across video, display, audio, DOOH, and more.

In Q2 2026, the Curation Protocol standardizes how contextual targeting and brand safety requirements are both requested and applied to inventory packages.

PROMPT: Sheesh, so many protocols, but how does it actually work?

It’s more straightforward than all these standards and protocols stuff would have you believe. Let’s focus on getting a campaign running in the Media Buy Protocol, which is helpfully very illustrative of how buyer agents can interact with the wider agent landscape.

Here’s a simplified workflow.

STEP 1: Write a natural language brief. The AgenticAdvertising.org offers this helpful example.

“Mike’s Plumbing Services needs to reach homeowners in the Denver, Colorado area who might need plumbing services. We have $8,000 USD to spend from October 15-31, 2024. Looking for display and native formats to drive phone calls.”

STEP 2: The “get_product” function will find inventory that matches the brief. The agent, likely with human oversight, will review offers, select applicable products, or revise the initial brief to get more suitable offerings.

STEP 3: Once satisfied with packages, use the “create_media_buy” function to… Well, create a media buy. Share your timing, overall budget, then wait for approval from the seller agent’s human monitor.

STEP 4: Once a human gives the thumbs up, upload creatives, activate the campaign, then monitor and optimize.

PROMPT: There seems to be a lot of human activity required with these agentic transactions.

Would you not want humans involved? Considering problems with AI hallucinations, would you want to find out you just allocated 50% of your campaign budget to a publisher that doesn’t exist? AdCP has many “human approval hooks,” and it’s widely agreed that humans will be intensely involved as the use of agents ramps up. In time, agents should develop best practices and will need less human hand-holding.

PROMPT: Right, right. Also, this all sounds like it’s about… automating direct, guaranteed campaigns.

Yes! AdCP is about empowering direct buys/sales across publishers at scale.

Agencies frequently complain they can’t scale I/Os for direct buys because going publisher to publisher is time-consuming and inefficient. They want to buy audiences and inventory en masse across a wide spectrum of media. While programmatic guaranteed has taken a lot of the pain out of the process by streamlining direct-buying on individual publishers, AdCP, as a communication layer, can theoretically facilitate guaranteed buys at scale across a wide range of publishers.

PROMPT: Then I guess RTB is a goner, huh?

Despite all the change, AdCP is not replacing existing programmatic infrastructure.

Instead, it sits above it.

• RTB continues to handle real-time auctions.
• PMP and direct deals continue to exist.
• AdCP helps agents create and manage those deals more efficiently.

Not at all! If you think about it, private marketplace deals (PMPs) are direct campaigns, that is, negotiated deals that leverage RTB. AdCP can theoretically be used to establish PMPs and Deal IDs across publishers and at scale. Agentic transactions could make the setup and execution of Deal IDs far more efficient. 

AdCP establishes the deal, and RTB is used to execute.

PROMPT: OK, then AdCP isn’t replacing RTB… But what about open programmatic?

There will always be room for open programmatic, and not just potentially for “unsold inventory.” Open programmatic bids should compete against PMPs and guaranteed campaigns so publishers can better understand yield to price inventory and audiences. 

PROMPT: So that’s why there’s this ARTF thing? Isn’t that about AI, too?

AI agents are driving the need for a major modernization of RTB workflows. The  IAB Tech Lab’s Agentic RTB Framework (ARTF) modernizes ad auctions to keep up with the speed of AI. Currently, auctions crawl because they constantly “call” outside vendors for data. ARTF replaces these slow calls with containerization.
Let’s break it down: 

• The Switch: Instead of calling a vendor, ad platforms “plug in” the vendor’s code directly into their own servers.
• The Speed: Because everything happens locally, latency drops by 80% (from ~700ms to just 100ms).
• The Win: It’s cheaper, more secure, and allows AI to make complex decisions instantly without the “lag.”

Bottom line: ARTF moves the tech to the data, instead of sending the data to the tech.

PROMPT: Oh boy. I thought my head hurt before when I was ingesting all that info about AdCP.

And we’re only scratching the surface regarding what’s happening at IAB Tech Lab! That industry standards organization has put together an umbrella of standards called the Agentic Advertising Management Protocols (AAMP), a kind of master playbook for how buyer, seller, measurement, creative, and audience agents plug into a multifaceted programmatic environment, including RTB.

Yes, ARTF is a part of that, as well as Agentic Direct, which offers guardrails for buyer and seller agents… to negotiate and execute direct deals.

PROMPT: WHAT?!? You mean… Agentic Direct is a competitor standard to AdCP?!?

Bet you didn’t see that twist coming AdTech sure is dramatic! It’s a bit of a plot twist: Agentic Direct and AdCP aren’t actually the same thing, but they can be BFFs.

• The Difference: AdCP is like a high-level “universal remote” for ad deals. Agentic Direct is more “on the ground,” plugging directly into the IAB’s existing rulebook to help agents handle things like open auctions and private deals.
• The Vibe: AdCP operates “above” the technical standards, while Agentic Direct lives right inside the workflow with the publishers.
• The Best Part: They don’t have to compete! A buyer agent might use AdCP to plan the big picture, then hop into Agentic Direct to actually execute the deal with a publisher.

It really just comes down to which “meeting room” the agents decide to hang out in. Plus, you can bet the buyers will have a huge say in which one becomes the favorite.

PROMPT: Oh man. I can’t take any more. My brain is full.

Yeah, it’s a lot, right? And once we start digging into the weeds, we lose focus on the true potential of AI: making media buying and selling faster, more efficient, better performing, and easier. 

We get it… it’s easy to get lost in the “alphabet soup” of all these protocols, but don’t let the tech talk scare you. At the end of the day, AI is just here to make buying and selling ads faster and way less of a headache.

Here’s the “vibe check” on what actually matters:

• History repeats itself: This shift is just like when we stopped hard-coding ads and started using ad servers. It felt weird then, but now we can’t live without it.
• You don’t need to be an engineer: You don’t need to build these protocols; you just need to know they exist so you can ask your partners, “Hey, can we plug into this?”
• Flexibility is king: Don’t marry one single protocol. The winners will be the ones whose tech is flexible enough to speak every “dialect”—whether it’s AdCP, AAMP, or whatever comes next.
• Talk the talk: Since AI runs on natural language, if you can explain what you want, your agent can handle the rest.

Your choice of self‑serve or direct‑sales platform will determine how quickly you can plug into whichever rulebooks win out:  AdCP, AAMP, or others, without ripping out your whole stack. 

PROMPT: Wow, that makes me feel better. Wait, did AI actually write this?

AI definitely helped with the research, but this is very human-created. See? AI isn’t even replacing the job of bloggers yet.

Speaking of humans and AI working together, we’re not just talking about this shift, we’re helping build it. That’s why DanAds is officially a Founding Member of AdCP.

By becoming a Founding Member of the Ad Context Protocol (AdCP), we’re helping ensure that self-serve advertising infrastructure is represented in the evolution of AdCP, and that publisher-direct monetization remains a core component of the emerging agent-driven advertising ecosystem.

AdCP signals a shift toward open standards designed to enable interoperability between publishers, advertisers, and adtech platforms, replacing isolated systems with a more connected, agent-ready infrastructure layer.

To explore what this means in practice and why it matters now, read the full DanAds announcement.

* This was an actual response from Perplexity.