Adobe brings CX Coworker marketing agent to ChatGPT and Copilot

Expanding the AI workspace

Adobe recently announced that it is bringing its CX Enterprise Coworker marketing agent to several major third-party AI platforms. While the company made the agent generally available in June, this new update focuses on where employees can actually interact with the technology. Instead of staying inside a dedicated Adobe workflow, the agent is now accessible within tools like ChatGPT, Claude, and Microsoft Copilot.

The goal of this distribution is to address the issue of marketers using different AI tools in isolation. By placing Coworker inside these popular assistants, Adobe aims to connect those separate work surfaces to a company's core customer-experience systems. This allows teams to start campaign planning and analysis from the interfaces they already use daily.

According to Adobe, the agent is designed to draw on specific brand, customer, and channel context. It is intended to coordinate marketing work while remaining under human oversight. This move signals a shift toward a more decentralized way of interacting with enterprise marketing software.

How MCP connectors work

A major part of this update involves the introduction of prebuilt Model Context Protocol (MCP) connectors. MCP is a standard that provides a way for AI software to interact with various tools and data sources. These connectors allow the Coworker agent to link with specific advertising and marketing systems like Amazon Ads, Demandbase, and Meta campaign tools.

For example, a marketing operations team could use an assistant to ask for a performance assessment across different channels. The agent would then use the approved connected systems to pull the necessary information and coordinate the next steps. This creates a path for the AI to move from simply providing information to helping with actual work orchestration.

Adobe also references Agent2Agent (A2A) as a standard for how different agents can work together. While these connectors make it easier for software to talk to each other, they do not automatically handle security. The benefit of these protocols is in how they facilitate the flow of data, rather than replacing the need for manual rules.

Governance and permissions

Moving an enterprise agent into a third-party platform like ChatGPT introduces significant questions about data permissions. It is not guaranteed that every enterprise account will have the same permissions to perform every action just because the agent is available in a new interface. Marketing operations teams must still ensure that access rights and audit records are strictly maintained.

The company emphasizes that the presence of a connector does not remove the need for approval checkpoints. Organizations must still map out which specific actions require a human to sign off before the AI can proceed. The focus for many companies will shift toward defining which systems and measurement rules an assistant can safely access for a specific client.

This creates a new type of work for systems integrators and agencies. Instead of just building new dashboards, they may need to focus on defining the boundaries of what an assistant can do. The practical competition in the market is becoming less about the tools themselves and more about how well a company can govern those connected workflows.

The competitive landscape

Adobe's move puts pressure on other major enterprise martech vendors to meet teams where they already work. Companies like Salesforce, Oracle, and Microsoft already sell various parts of the customer-data and CRM stack. Adobe is attempting to maintain its position by keeping the customer-experience context even when the interaction happens on a competitor's platform.

The company has identified several systems integrators and agencies as part of its ecosystem, including names like Accenture, Deloitte Digital, and WPP. These partners will likely be involved in helping customers set up these complex connections. However, these are reported participants and do not mean every customer will have the same configuration.

For agencies specifically, this shift could change the nature of integration briefs. Rather than creating a new place to see data, the task becomes ensuring that the AI assistant follows the correct rules. The challenge lies in making sure the reach of the AI comes with predictable costs and measurement for the company's existing software stack.

Next steps for marketers

Procurement and marketing-operations teams should begin by reviewing official product documentation to see which surfaces are enabled. It is important to confirm which specific connectors are active for a particular account. Teams should request a sandbox demonstration to see how the permission boundaries actually function in a live environment.

There are four primary checks that teams should perform before connecting Coworker to their systems. They need to know which customer data is accessible from an external assistant and which actions require a human to give explicit approval. They also need to know where the event logs and rollback records are stored if something goes wrong.

Finally, it is important to distinguish between read-only reporting and actions that actually change things. Teams should be clear on whether the AI can only look at data or if it has the power to change paid-media campaigns. Understanding these distinctions is key to using the new platform safely and effectively.