Adobe released Coworker, a "super agent" for marketing that can manage a stable of task-based agents to define campaigns guided by analytics data and then design, tee up, adjust and -- with human approvals -- launch them. All, Adobe claims, on the fly.
Not to be confused with Adobe Agent Orchestrator, Coworker is a set of agents built with the same CX Enterprise agent harness. The harness adds deterministic business rules to probabilistic workflows -- such as gathering required approvals, deploying agentic skills (reusable instruction sets that teach AI the order of operations), and orchestrating agents from multiple vendors.
Agent harnesses unlock the potential for more sophisticated workflows than task-based agents can, said Daniel Sheinberg, senior director of product management for Adobe Experience Cloud.
"You can just say to the orchestration layer or the harness, 'Here are all the APIs and the tools that are available to you, and on the fly you can figure out these agentic workflows,'" Sheinberg said. "What's happening with these harnesses is, yes, you can give it prebuilt agents, but you can also just expose it to a library of skills, and it will figure out how to use these skills to get what you want done."
Coworker agents serve as the manager of Adobe tools -- such as Audience Agent and Journey Agent -- as well as agents from other tools that plumb outside data sources.
Another agent, Coworker Chat, executes complex edits in Creative Cloud apps like Photoshop and Illustrator from plain-language prompts.
Where A2A, MCP come in
Coworker is Adobe's answer to agent management systems such as ServiceNow's AI Control Tower, albeit without an agentic "kill switch," said Liz Miller, analyst at Constellation Research.
Adobe customers hoping to automate, for example, the process of generating social media posts based on fresh website content, don't just need support for model context protocol (MCP) and Agent-to-Agent (A2A) agentic interoperability standards, she said. They need tools customized to their marketing work, because many processes are fraught with complications when humans do them -- and the agents need to fix that, not add to the complexity.
Coworker helps marketers tap into those A2A and MCP standards by embedding them into familiar, everyday marketing workflows, Miller said. These workflows can span creative processes, deployment and execution, guided by analytics and audience data along the way.
"We've got an MCP server, they've got a server. I'm looking at A2A, you're looking at A2A. Where do I actually start connecting these things?" Miller said. "It's one thing to be able to say, 'We are looking at protocols that enable one agent to hand off to another agent.' It's completely another thing to actually attribute that to a work product important to a marketer."
Users of agentic AI in general -- not just Adobe's -- are discovering that security and governance are crucial elements to keeping agentic AI workflows in check. As agents proliferate in marketing, automations that incorporate agents, tap MCP servers and autonomously generate content or take other actions need to stick to guardrails.
Coworker also has workflow templates that govern individual use cases, such as campaign creation or retargeting, to keep them consistent across an enterprise's marketing operations, Sheinberg said.
Coworker, built on headless architecture, is available standalone or as an add-on to existing Adobe stacks. Adobe builds a custom interface for each brand using it. A self-service offering is available for smaller business or marketing departments.
Don Fluckinger is a seasoned B2B technology journalist with more than 30 years of experience specializing in enterprise IT, digital experience and content management. As a senior news writer at Informa TechTarget, he delivers award-winning analysis that helps IT and business leaders navigate complex technologies to enhance customer and employee experiences. Got a tip? Email him.