My thoughts on AdCP and why agent-based protocols are replacing platform UIs
This morning, AdCP (Advertising Control Protocol) launched—and I haven’t been this interested in an adtech announcement in years. Below is my recap and interpretation on what we’re experiencing.
I was part of the team that commercialized RTB back when that protocol transformed display advertising. We went from faxed insertion orders to millisecond auctions. The shift wasn’t about better software—it was about establishing a standard that everyone could implement. Once enough platforms adopted RTB, the entire industry’s logistics fundamentally changed.
AdCP has that same architecture. But this time, the protocol solves for agent coordination across platforms instead of auction dynamics within platforms.
What Protocols Actually Do (And Why They Matter More Than Features)
Before going further, let me explain what we mean by “protocol” for those who don’t live in the technical weeds.
Think of a protocol as a common language different systems agree to speak. Email works because Gmail and Outlook both follow the same protocol (SMTP). You don’t think about it, but that shared language is why email works seamlessly across every provider.
In advertising, protocols solve coordination problems at scale. Before RTB, buying from 50 publishers meant 50 different relationships and custom processes. RTB created a standard auction language—suddenly you could access thousands of publishers because everyone spoke the same way.
The magic happens when everyone adopts the standard: everything gets easier for everyone.
That’s what AdCP does for agent-based coordination. It’s a shared language that lets any buyer agent work with any seller agent, regardless of which platforms or interfaces anyone uses underneath.
What Actually Happened Today
Here’s the core concept: instead of humans logging into platform dashboards, AI agents coordinate media transactions using a standardized protocol.
A marketer’s agent discovers inventory across multiple publishers, evaluates options based on historical performance, negotiates terms with seller agents, coordinates targeting, manages creative, and monitors delivery. The marketer reviews recommendations and approves key decisions. The agents handle the coordination complexity.
The seller agent maintains approval workflows, enforces pricing rules, and coordinates with their ad server. The buyer agent isn’t locked into any specific platform—it works with any seller speaking the protocol.
Yes, the demo showed Claude as the interface. But that’s not the interesting part. The interesting part is the protocol layer underneath that lets any buyer agent coordinate with any seller agent—regardless of what interface the buyer uses or what ad server the seller operates.
The Ematini Story That Made It Real
The presentation showed an Ematini beverage brand campaign executed in one day instead of the typical 3-4 weeks. Here’s what actually happened:
The marketer described campaign objectives in plain language. Their agent queried multiple seller agents to discover available CTV inventory. Seller agents responded with options—formats, pricing, audience matches. The buyer agent evaluated these against historical performance data, made recommendations, created a $20K media buy, coordinated targeting signals, trafficked creative assets, and set up monitoring—all while the marketer just reviewed key decisions and approved.
The marketer never thought about which platforms to log into or how to coordinate across systems. The protocol abstracted that complexity entirely.
This only works when everyone speaks the same language for discovery, negotiation, execution, and monitoring. That’s what protocol standardization delivers.
Why Protocol Standardization Matters More Than AI Capabilities
Here’s the insight most people are missing: The innovation isn’t that ChatGPT can understand natural language requests. The innovation is that standardized protocols let any agent coordinate with any other agent without custom integration.
Let me explain why this distinction matters:
AI capabilities are commoditizing. ChatGPT, Claude, Gemini—they’re all converging on similar natural language understanding. In 2 years, dozens of models will handle conversational interfaces competently. That’s not differentiation.
Protocol standardization creates defensible positions. Once a protocol achieves ecosystem adoption, it becomes extremely difficult to displace. TCP/IP emerged in 1983—40+ years later, it’s still the internet’s networking standard despite numerous technically superior alternatives. HTTP became the web standard in the 1990s and remains dominant. SMTP enabled email in 1982 and is still how email works today.
Why? Because ecosystem coordination cost for migration is enormous. Once thousands of companies build systems around a protocol, once millions of users expect things to work a certain way, once the entire industry operates on shared standards—switching becomes nearly impossible even if something better exists.
The race isn’t about building the best AI interface. The race is about establishing the protocol standard that everyone builds on. AdCP isn’t competing to be the best chatbot for media buying. It’s competing to be the HTTP of agent coordination—the foundational standard that enables the entire ecosystem.
This is why timing matters. First viable protocol in a new category tends to become permanent standard. Not because it’s technically perfect, but because protocol effects create winner-take-most dynamics through network effects and switching costs.
The Progression: Every Major Protocol Solved A Coordination Problem
Let me show you the clear progression of advertising infrastructure:
- RTB (2009): Standardized programmatic auctions so buyers could access thousands of publishers without custom integrations. Before RTB, you needed separate relationships with every publisher. After RTB, one integration accessed the entire ecosystem.
- Header Bidding (2015): Enabled parallel evaluation across demand sources instead of sequential waterfalls. Publishers could run simultaneous auctions rather than calling buyers one at a time until someone bid.
- Ads.txt (2017): Created verifiable authorization so buyers could trust seller legitimacy. The protocol made supply chain transparency possible without centralized verification.
- AdCP (2025): Enables agents to coordinate across any platforms without platform-specific expertise. Buyers don’t need to know each platform’s API, data formats, or integration requirements—agents speak the protocol.
Each addressed something that couldn’t be solved by building a better platform. You needed a standard everyone could adopt. That’s when coordination transforms.
The key insight: platforms optimize within their walls. Protocols optimize across the entire ecosystem. Platforms want you locked in. Protocols enable you to switch freely while maintaining coordination capability.
Why Sellers Should Pay Attention
Here’s the design choice that matters: AdCP is explicitly asynchronous and human-in-the-loop.
Unlike RTB’s millisecond auctions that can force race-to-bottom dynamics, AdCP’s protocol allows operations that take hours or days. Publishers can auto-approve standard campaigns under $5K but route larger ones to their sales team for review. They can set dynamic pricing based on fill rates and demand patterns. They can require approval for sensitive categories.
The protocol standardizes coordination logistics while preserving business model control. Automation handles the tedious parts—discovery, proposals, configuration—while human judgment stays in the loop where it matters for quality and pricing.
That’s fundamentally different from “automate everything for efficiency.” It recognizes that sometimes speed matters less than thoughtfulness.
The Network Effect That Makes Protocol Standards Inevitable
Once enough participants adopt a protocol, network effects accelerate adoption:
- For buyers: Every new seller becomes immediately accessible without new integration work. Your agent already speaks their language.
- For sellers: Every buyer agent can discover and transact with your inventory. You implement the protocol once, gain access to the entire agent ecosystem.
- For platforms: You compete on actual value—inventory quality, audience data, optimization—rather than switching costs.
This is exactly how email won despite Microsoft and Google having platform advantages. SMTP was an open standard anyone could implement. The best interface won, but the protocol stayed constant.
The implication: if AdCP achieves meaningful ecosystem adoption in the next 18-24 months, it becomes extremely difficult to displace even if technically superior alternatives emerge. Protocol effects create durable competitive positions independent of who builds the best features.
Why This Moves Faster Than Previous Waves
The 2-year timeline isn’t optimistic—it’s realistic based on three factors:
The infrastructure already exists. Unlike RTB, which required building new ad servers and exchange technology while an entire computing paradigm was emerging, AdCP runs on top of existing infrastructure. Platforms just need to implement agent interfaces—ChatGPT, Claude, and other LLM frameworks are already production-ready.
AI adoption is compounding faster. It took RTB 3-4 years to reach meaningful adoption. But AI agents are being adopted across industries simultaneously. Every organization building agents for customer service, operations, or analytics can extend them to media buying once the protocol exists.
Economic pressure accelerates change. Current platform architectures face severe economic challenges—OpenAI projecting massive losses, enterprise AI pilots failing at high rates, inference costs scaling faster than pricing allows. When status quo is unsustainable, adoption happens fast.
The Flywheel Already Spinning
As campaigns execute through AdCP, the performance data that informs agent recommendations improves. Which publishers deliver best outcomes for which targeting? What pricing is competitive? Which creative approaches work in which contexts?
This isn’t theoretical—it’s happening today. Every campaign through AdCP creates data that makes the next recommendation smarter. That feedback loop accelerates adoption because agent quality improves with every execution.
Early movers get better agents faster because they’re building the performance history that makes recommendations valuable. That creates competitive pressure: if your competitors’ agents are making smarter decisions because they have more data, you need to adopt to keep pace.
My Take: Protocol Standard Status Within 24 Months
After decades in this industry, I recognize the difference between incremental improvements and structural innovations. Platform improvements make workflows easier. Protocol innovations change what coordination is possible.
RTB changed what was possible by standardizing programmatic auctions. AdCP could change what’s possible by standardizing agent coordination across platforms.
The architecture is right: distributed agents coordinating through open protocols rather than centralized platforms orchestrating through proprietary systems. The timing is right: technology enables distributed architecture just as economic pressure makes centralized approaches unsustainable. The approach is right: human-in-the-loop asynchronous coordination that preserves business model control while automating logistics.
In two years, logging into ad platforms will feel as outdated as faxing insertion orders feels today.
Not because platforms die suddenly, but because protocol infrastructure provides better experiences with better economics for enough use cases that platforms become the exception rather than the default. You’ll still log in when something breaks or requires manual intervention. But standard media buying will happen through agents coordinating via protocols.
I’m optimistic because this addresses real problems—publisher sustainability, buyer efficiency, platform lock-in—through standardization that aligns everyone’s interests. When protocols work, everybody wins because the coordination problem gets solved for everyone simultaneously.
And this time, the infrastructure for rapid adoption already exists. We’re not waiting for technology to mature—we’re just waiting for the industry to recognize that the coordination problem has a standard solution.