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

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

SHARE:
Dr. Aaron Andalman, Chief Science Officer and Co-Founder, Cognitiv

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.

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.

Must Read

Sweetgreen Tapped Atmosphere TV To Introduce World Cup Viewers To Its Wraps

Sweetgreen turned the World Cup into a marketing moment for its new wraps, running video ads in public viewing spaces through Atmosphere TV.

Amazon Crushes Earnings And Reaches Almost $20 Billion In Q2 Ad Revenue

Amazon’s advertising businesses earned a total $19.8 billion in Q2, the company reported in its quarterly earnings on Thursday. That’s up from $15.7 billion in Q2 2025, and good for a 26% year over year growth rate.

Los Angeles, California - 26 February 2023: Reddit social media platform displayed on smart device

Reddit Had A Great Q2, But Investors Have AI Search Jitters

Guess there’s no pleasing investors. Despite Reddit delivering an objectively solid Q2, its stock cratered, in part because of search-related headwinds and low referral traffic.

Privacy! Commerce! Connected TV! Read all about it. Subscribe to AdExchanger Newsletters

Meta’s Expenses Are Growing Faster Than Its Revenue, Thanks To Lawsuits And AI

A combination of layoffs, lawsuits and AI operating costs set back Meta’s Q2 earnings, despite increased revenue.

Omnicom Investors Cheer IPG Sell-Off, Despite Weak Ad Spend In Q2

Omnicom is halfway through a major sell-off of IPG agencies. Its future looks healthier as it prunes lower-growth firms, including eliminating certain specialist firms and overlapping agencies in certain countries.

Hundreds of emails, depositions and other documents have been unsealed in the lead-up to the Google antitrust trial, providing a fascinating look at how Google talked about its own products when no one else was watching – especially tools to counteract the rise of header bidding.

Why PubMatic Ditched Its Prebid Web Wrapper, But Never Its SDK

Earlier this month, PubMatic shelved its Prebid integration wrapper, known as OpenWrap Web, and announced it would begin recommending Playwire as an offloading-onboarding partner for the 250-odd publishers that use its wrapper.