The Agentic Shift Is Already Happening Inside Publisher Operations

Originally published by IAB Tech Lab as a Member Perspective.

When we launched the Burt MCP server in early March, agentic workflows in advertising still felt fairly experimental.Five months later, things have changed rapidly.

Publishers are already using agents to democratize governed data in their organizations, explore complex datasets, investigate anomalies, and build and automate essential parts of their day-to-day with LLMs. For Burt, MCP has become the fastest-adopted new offering we have ever released.

That speed suggests something broader is going on: AI is moving from answering questions about the business to participating in how the business operates.

The first use case is straightforward. Someone who previously needed to understand a complex data model, navigate multiple dashboards, or ask an analyst can now investigate a question conversationally.

But the more interesting use cases begin when the agent can do something with the answer.

We have just released a new version of our MCP implementation with significantly expanded tooling, including full create, read, update and delete (CRUD) capabilities for reporting, along with a range of improvements behind the scenes. More operational capabilities are coming, including around forecasting and other publisher workflows.

This points toward a very different interface for ad tech.

Instead of teaching every user how every system works, we can increasingly give an agent access to governed capabilities across those systems and allow the user to describe the outcome they want.

• Why did revenue decline yesterday?

• Which campaigns are at risk?

• Build a report explaining the difference.

• What inventory can I confidently sell next month?

Each of those questions can involve multiple datasets, systems and analytical steps. Increasingly, that complexity can sit behind the agent rather than in front of the user.

For example, one of our publisher partners built an internal support agent for its sales team to ask questions to. The agent has all their internal context and documentation combined with the performance data in Burt, allowing the sales team to retrieve insights into their customers themselves directly via the agent, without having to pull the ad ops team into every question.That is exactly the kind of data democratization that makes agents so powerful. It also raises an obvious question: as access to data gets easier, how do you make sure it remains governed?

Democratization Needs Governance

Publishers have spent years building complex data environments across ad servers, SSPs, DSPs, order management systems, CRM platforms and many other adtech systems.

Agents dramatically lower the technical barrier to accessing that data, but lowering the barrier cannot mean lowering the standards around how that data is accessed and used.

As I recently wrote in AdExchanger, this becomes particularly crucial when AI touches revenue data. Finance, billing and client reporting require auditability. The source of the data, the logic applied to it and the permissions around it all need to rest on solid foundations. Those requirements become even more important as agents move from reading information to taking action.

This is also where industry coordination starts to matter. We published the Burt MCP in the IAB Tech Lab Agent and MCP Registry early in its development, and it has given us a useful vantage point on how quickly the ecosystem is taking shape.

The Registry provides a common place for the industry to discover these new capabilities. Efforts such as IAB Tech Lab’s AAMP initiative can help establish the shared conventions around how those capabilities are described, trusted and ultimately used together.

Adtech already has enough fragmentation. The agentic layer should not recreate another generation of proprietary interfaces and one-off integrations.

We are still early. There is plenty left to solve around authentication, permissions, interoperability and governance.

But judging by what publishers are already doing, the agentic shift is moving from experimentation into day-to-day operations remarkably quickly.

The infrastructure underneath it now needs to mature just as quickly.