The AdCP Hype Problem: Why Standardized AI Workflows Don’t Equal Better Media Outcomes

Recorded: Jan. 26, 2026, 7 a.m.

Original

Summarized

The AdCP Hype Problem: Standardized Workflows Don’t Equal Better Outcomes

image/svg+xml:

Topics
Latest
Marketers
Agencies
Publishers
Technology
Platforms
Identity
Measurement
Data Privacy
Artificial Intelligence
CTV
Commerce
AdExplainer
Exclusive Report
Daily News Roundup

Opinion
All Columns
Data-Driven Thinking
On TV & Video
The Sell Sider
Content Studio
Comic
Contributor Guidelines

About Us
Advertise
Newsletter
AdExchanger Advisory Board
About Us
Contact Us

Events
Programmatic AI Las Vegas
AdExchanger Awards
Webinars
All Events
Network Events

Podcasts
AdExchanger Talks
The Big Story
Inside the Stack

NEW! Programmatic AI 2026

Become an AdHero

Topics
Latest
Marketers
Agencies
Publishers
Technology
Platforms
Identity
Measurement
Data Privacy
Artificial Intelligence
CTV
Commerce
AdExplainer
Exclusive Report
Daily News Roundup

Opinion
All Columns
Data-Driven Thinking
On TV & Video
The Sell Sider
Content Studio
Comic
Contributor Guidelines

Events & Awards
Programmatic AI Las Vegas
AdExchanger Awards
Webinars
All Events
Network Events

Podcasts
AdExchanger Talks
The Big Story
Inside the Stack

Subscribe Free
Sign Up

About Us
Advertise
Newsletter
AdExchanger Advisory Board
About Us
Contact Us

CONNECT

Home Data-Driven Thinking The AdCP Hype Problem: Why Standardized AI Workflows Don’t Equal Better Media Outcomes

OPINION: Data-Driven Thinking
The AdCP Hype Problem: Why Standardized AI Workflows Don’t Equal Better Media Outcomes By Dr. Aaron Andalman, Cognitiv

Monday, January 26th, 2026 – 12:35 am
SHARE:

Dr. Aaron Andalman
Chief Science Officer & Co-Founder

If you are feeling pressure to “do something” about Ad Context Protocol, you are not alone. Since its launch a few months ago, AdCP has been framed as a foundational step toward an “agentic” future, one where AI systems plan, execute and optimize media buys without human hand-holding.
It is a compelling vision, but one that relies on capabilities that do not exist yet. AdCP may one day make buying easier, but it will not make buying better by itself.
What AdCP was built to fix
AdCP is built on top of Model Context Protocol (MCP), an open standard introduced by Anthropic and broadly adopted by the industry. MCP addresses a well-known limitation of large language models. While AI models have become much better at language and reasoning, they still struggle to take reliable actions inside real software. Interactions with external tools can be brittle, with models misunderstanding what actions do or invoking them incorrectly.

MCP tackles that problem by standardizing how tools describe the actions they support. And AdCP applies this idea to advertising, giving ad tech platforms a shared way to describe actions like creating a strategy or activating an audience. As a result, AI models can operate across platforms without learning a bespoke API for each one.
AdCP standardizes the interface between models and advertising systems. It makes it easier for a model to discover available actions and invoke them consistently, like ordering items off a menu. That helps workflow automation scale, but it stops at execution.
Standardizing how an action is triggered does not change the quality of the action itself. If a bidding strategy is poorly designed or an audience definition is weak, AdCP can help invoke it more reliably, but it cannot improve the underlying logic.
That distinction matters because it reveals two bigger questions hiding behind the AdCP hype.
Are “agents” ready to run advertising?
Much of the excitement around AdCP is tied to autonomous agents. Yet the “agents” making headlines today are typically large language models connected to tools. These systems are excellent at interpreting instructions and coordinating actions, such as to “set up a CTV campaign,” but they are not inherently optimizing toward a business objective. They are trained to model language, not to learn through trial and error in noisy, complex environments.
Today’s LLM-based agents often lack the internal feedback loops required to learn from outcomes and continuously improve performance. And those limits quickly bubble up in real-world scenarios. In a recent Wall Street Journal experiment, a state-of-the-art agent tasked with running a simple vending machine was quickly manipulated into giving away inventory and ordering a hodgepodge of items, including a live fish, a PlayStation and some kosher wine.
If an autonomous system struggles to manage a vending machine – a closed environment with fixed inventory and transparent pricing – it is a stretch to expect it to reliably manage a multimillion-dollar media budget across fragmented channels, shifting auctions and adversarial market dynamics.
AdCP does not unlock full autonomy. It simply enables a new way to interact with existing platforms: chat instead of dashboards. While that can be useful for operational convenience and exploration, it is a long way from a hands-off performance revolution.

AdExchanger Daily
Get our editors’ roundup delivered to your inbox every weekday.

Daily Roundup

Daily News Roundup
TikTok US (Finally) Emerges; Forbes, For Bettor Or Worse

Where can AI actually improve performance?
The more interesting opportunity for AI in advertising sits inside the platforms themselves.
AI can materially improve how audiences are understood, how context is interpreted and how first-party data is activated. World knowledge, language understanding and reasoning can be used to enrich signals, inferring intent from content, resolving ambiguity in user behavior and translating messy business data into usable features for models.
This is where AI actually moves the needle: inside systems that score impressions, predict outcomes and balance trade-offs in real time. These models can learn directly from outcomes, testing bidding strategies, audience definitions and contextual signals, then adjusting based on what actually performs.
AdCP doesn’t compete with this work, but it does not replace it either. A cleaner interface for invoking tools only matters if the tools themselves are improving. Performance gains still come from better predictions and better decisions inside the platform.
Implications for 2026 planning
AdCP should stay on your radar. It is a sensible standard, and, over time, it may make AI-driven workflows easier to build.
But it should not be the primary focus.
If your goal is performance, the more important question is not how an AI chatbot invokes actions. For the foreseeable future, performance will come from platforms using AI to improve targeting, prediction and optimization at the core of programmatic buying. This is work advertisers will need to find on their platforms or build themselves.
AdCP may shape how those systems are accessed. It will not determine whether they work.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow Cognitiv and AdExchanger on LinkedIn.

Next In Data-Driven Thinking

Privacy Enforcement Crossed A Line In 2025 – And There’s No Going Back

Related Stories

OPINION: Data-Driven Thinking
AdCP And The Math Of Agentic AI: Building For Today's Economics, Not Tomorrow's Dreams

OPINION: Data-Driven Thinking
Why Your First AdCP Project Is Probably Going To Fail

Comic
Comic: "I Need Some Coffee Context Protocol"

Platforms
APIs Have Had Their Moment, But MCPs Reign Supreme In The Agentic Era

Must Read

PODCAST: The Big Story
Guess Its AdsGPT Now?

Ads were going to be a “last resort” for ChatGPT, OpenAI CEO Sam Altman promised two years ago. Now, they’re finally here. Omnicom Digital CEO Jonathan Nelson joins the AdExchanger editorial team to talk through what comes next.

Commerce
Hershey’s Undergoes A Brand Update As It Rethinks Paid, Earned And Owned Media

This Wednesday marks the beginning of Hershey’s first major brand marketing campaign since 2018

Programmatic
A Win For Open Standards: Amazon’s Prebid Adapter Goes Live

Amazon looks to support a more collaborative programmatic ecosystem now that the APS Prebid adapter is available for open beta testing.

Publishers
Gamera Raises $1.6 Million To Protect The Open Web’s Media Quality

Gamera, a media quality measurement startup for publishers, announced on Tuesday it raised $1.6 million to promote its service that combines data about a site’s ad experience with data about how its ads perform.

AI
CES 2026: What’s Real – And What’s BS – When It Comes To AI

Ad industry experts call out trends to watch in 2026 and separate the real AI use cases having an impact today from the AI hype they heard at CES.

Commerce
New Startup Pinch AI Tackles The Growing Problem Of Ecommerce Return Scams

Fraud is eating into retail profits. A new startup called Pinch AI just launched with $5 million in funding to fight back.

Popular

CTV
You’re Not Imagining It – There Are More Shoppable CTV Ads On YouTube Now

YouTube is finally delivering on the promise of shoppable CTV ads. Starting Thursday, they’ll roll out to all of Google’s Performance Max and Demand Gen campaigns, YouTube confirmed to AdExchanger.

AI
What We Know: ChatGPT Is Launching Ads. The Rest Is Speculation

ChatGPT’s initial ad rollout didn’t come as a surprise. Now, advertisers are figuring out how to make the most of this new channel.

Programmatic
A Win For Open Standards: Amazon’s Prebid Adapter Goes Live

Amazon looks to support a more collaborative programmatic ecosystem now that the APS Prebid adapter is available for open beta testing.

Marketers
Linkby Raises $15 Million For its Hybrid PR And Affiliate Marketing Network

Linkby, which connects brands with publisher content that advertisers pay for based on reader engagement, announced its $15 million Series B on Tuesday.

OPINION: Data-Driven Thinking
Privacy Enforcement Crossed A Line In 2025 – And There’s No Going Back

For most of the last decade, privacy compliance lived in a gray zone. Companies could point to a cookie banner, update a policy and reasonably believe they were doing enough. Not anymore.

Join the AdExchanger Community
Join Now

Your trusted source for in-depth programmatic news, views, education and events.
AdExchanger is where marketers, agencies, publishers and tech companies go for the latest information on the trends that are transforming digital media and marketing, from data, privacy, identity and AI to commerce, CTV, measurement and mobile.

NEXT EVENT
Convergent TV World
March 5-6, 2026The Times CenterNew York, NY
Learn More

ABOUT ADEXCHANGER
About Us
Advertise
Contact Us
Events
Subscribe
RSS
Cookie Settings
Privacy & Terms
Accessibility
Diversity, Equity, Inclusion & Belonging

CONNECT

The AdCP Hype Problem: Standardized Workflows Don’t Equal Better Outcomes

Dr. Aaron Andalman, Chief Science Officer & Co-Founder of Cognitiv, dissects the current fascination with Ad Context Protocol (AdCP) and highlights why its potential, while real, shouldn’t be overstated. The core issue is that AdCP, built upon Model Context Protocol (MCP), offers a standardized interface for advertising platforms, allowing AI models to interact with tools more reliably. This standardization addresses a recognized limitation of large language models – their tendency to falter when interacting with external software – but it doesn’t inherently improve the *quality* of the underlying actions.

AdCP’s function is to facilitate a “menu-like” interaction, where AI can consistently invoke actions across different platforms. However, the success of that invocation depends entirely on the quality of the action itself. A poorly designed bidding strategy or a weak audience definition, regardless of how seamlessly AdCP facilitates its execution, will still yield undesirable outcomes.

The excitement surrounding AdCP is largely tied to the concept of “agentic” AI, where autonomous systems manage advertising campaigns. However, today’s AI agents, typically large language models connected to tools, are primarily adept at interpreting instructions and coordinating actions, not at performing optimization. These systems lack the crucial feedback loops required to learn from iterative outcomes and adapt in the complex, noisy arena of real-world advertising. The well-publicized experiment with a state-of-the-art vending machine agent vividly demonstrates this limitation – even a seemingly intelligent system struggles to manage a closed environment like a vending machine.

Therefore, AdCP doesn’t unlock full autonomy; it merely provides a new way to interact with existing platforms—a shift from dashboards to a chat-based interface. While this can offer operational convenience and exploration, it’s a far cry from a hands-off performance revolution.

Looking ahead to 2026, the most significant opportunity for AI in advertising lies within the platforms themselves. AI’s true impact comes from its ability to refine audience understanding, interpret contextual signals, and activate first-party data. By leaning on world knowledge, language understanding, and reasoning, AI can enrich signals, infer user intent, resolve ambiguity, and transform messy data into usable features for models. This is where genuine performance gains will be realized – in systems capable of learning directly from outcomes, testing bidding strategies, and adjusting audience definitions in real-time.

AdCP’s value resides in its potential to simplify access to these platforms, but it won’t determine whether they actually work. The focus should remain on the platforms' capabilities – how AI improves targeting, prediction, and optimization at the core of programmatic buying. Advertisers will need to find these improvements within the systems themselves or build them independently. Dr. Andalman concludes that in 2026, the success of AdCP will depend on whether the underlying AI-powered platforms continue to mature and deliver substantive performance improvements.