In 2025, you probably heard the word “agents” used more often in conversations about AI. In a newsroom context, that might mean agents that look at content you’re creating and provide additional context that may be relevant in the research and editing process.

It’s worth noting that the concept of an AI agent is not new. It remains at the heart of AI research. As Russell and Norvig define it, “an agent is something that perceives and acts in an environment,” and an “ideal intelligent agent takes the best possible action in a situation.”

Much of the recent discussion has been fueled by rapid developments in the underlying technology. Anthropic released the Model Context Protocol (MCP) in November 2024, offering a framework that allows an application to interact more fluidly with an agentic service to retrieve information or complete tasks. IBM followed with the Agent Communication Protocol (ACP) in March 2025, and Google introduced the Agent2Agent Protocol (A2A) in April. In August, work on ACP and A2A was consolidated under the auspices of the Linux Foundation. These newer protocols focus on enabling agents to communicate and coordinate with one another.

Looking ahead, I expect well-resourced newsrooms to accelerate experimentation with agents in 2026, using these emerging protocols to build tools tailored for newsroom workflows. Editorial use cases may require evolving current protocols to delineate between tasks and goals that require a deterministic, auditable response and those that don’t. A copyeditor agent applying AP Style, for example, must behave consistently and predictably. And if a research agent needs to coordinate with a copyeditor agent, their decisions must be transparent and repeatable, given the same information and context.

One of the largest opportunities, in my view, is the creation of an “institutional knowledge agent.” Such an agent would be trained on a newsroom’s internal and published archives. During major breaking news events, it could surface similar historical coverage and, working with a design agent, generate a ready-to-use package of related text, photos, and video. The goal is to preserve and operationalize the institutional knowledge that newsrooms accumulate — knowledge that too often walks out the door when journalists retire or move on.

I recognize that these visions will be within reach first for the most well-resourced news organizations. At the Associated Press’s workflow solutions team, our mission is to bridge that gap by taking on much of the research and development needed to deliver powerful, affordable, and relevant AI agents to our members and customers.