Adobe CX Enterprise Coworker in Microsoft Copilot: Admin Setup, Work IQ and Data Access

The broader theme comes from CX Today's roundup of Adobe, Orange France, Zendesk and Oracle. The question for buyers is whether a platform can tie intent, data, workflow, inventory, policy and human approval together without making operations harder to run. Or does it just make the conversation about the work sound smarter?

Adobe puts CX Enterprise Coworker inside the AI tools staff already use​

Adobe's press release dated November 4 says CX Enterprise Coworker is now available within several leading AI platforms. The named platforms are ChatGPT, Claude, Databricks Genie, Gemini for business and Microsoft Copilot. Amazon Quick, IBM and Snowflake are expected soon. Those three should not be treated as live yet.

Adobe describes Coworker as an agentic system built on Model Context Protocol (MCP) and Agent2Agent standards. It is meant to run customer-experience orchestration workflows across Adobe CX Enterprise, AI platforms and enterprise systems, with human oversight. The listed tasks include campaign performance analysis, audience forecasting, journey optimization and conflict resolution.

Adobe also announced prebuilt MCP connectors for Amazon Ads, Databricks Genie, Demandbase, an ads MCP server for Meta campaigns, Microsoft Work IQ, RainFocus and Snowflake Cortex AI. Adobe says marketers can use a single request to pull data and take actions across several systems. That is Adobe's description of the product. Nobody independent has validated the outcomes or the security.

Microsoft's Rob Howard, Corporate Vice President of Copilot Platform and Extensibility, is quoted in Adobe's release. He says the work is about helping enterprises move from insight to action across customer data, content and journeys.

A chronology note helps here. One roundup says the ChatGPT plugin went live on September 29, 2026, so the November announcement widens availability. It is not the first ChatGPT integration.

What a Microsoft 365 admin actually has to do​

Adobe's documentation shows this is not an end-user toggle. For Microsoft 365 Copilot, an administrator must complete these steps before anyone can use it:

1. In the Microsoft 365 admin center, confirm that tenant settings allow apps and agents built by external publishers.
2. Approve Adobe CX Enterprise Coworker in both the Microsoft 365 admin center and the Teams admin center. Adobe says it can typically be found through the Microsoft Commercial Marketplace or the Integrated Apps list.
3. Deploy it to specific users or groups, or to the whole organization.

Users also need access to Microsoft 365 Copilot. Once enabled, Adobe says Coworker appears across Copilot surfaces including Teams, Word, Excel, PowerPoint and web chat. Users add it by choosing Agents, then More agents, then selecting the agent and choosing Add.

Adobe adds that if Microsoft releases a Coworker update requiring new permissions, an administrator must approve it in the Agent Registry before it rolls out.

Other platforms follow the same pattern. In ChatGPT, the organization's administrator approves Coworker, and users then install it from the Plugins catalog. In Claude, the administrator approves it, and users connect via Customize > Connectors with their Adobe sign-in. In Gemini Enterprise, the admin goes through the AI Agents marketplace and a private-offer process with Adobe. Adobe's documentation says there is no extra cost for Coworker in AI platforms. It is included in the Coworker license and draws on the same AI credits.

Data access, limits and the Work IQ question​

Adobe's documentation says Coworker in AI platforms inherits the access tied to each user's Adobe identity. It says the integration returns read-only insights and does not modify customer data. Prerequisites include Adobe Experience Platform and at least one of Real-Time CDP, Journey Optimizer or Customer Journey Analytics. Known limitations include no support for custom agent harnesses outside the listed platforms. The Firefly Image Generation skill is also unavailable through Claude.

There is a tension to note. Adobe's launch language talks about taking action across partner systems, while the AI-platform documentation describes read-only insight for customer data. Admins should ask which specific connectors and tools can write, and where, rather than assume either description covers everything.

Work IQ is the Microsoft piece that deserves scrutiny. A WindowsForum analysis points to Microsoft Learn's description of the Work IQ MCP server as exposing Microsoft 365 capabilities to AI agents. It says Work IQ MCP uses Microsoft Entra authentication and the signed-in user's Microsoft 365 permissions. It also says tenant administrators can switch Work IQ MCP policy on or off in the Microsoft 365 admin center. I could not independently confirm the details of how Adobe's connector uses Work IQ beyond Adobe's own announcement.

A reasonable review checklist for admins:

• Who can approve external agents in your tenant, and is that approval logged?
• Which users get the agent, and do they already have the Adobe and Copilot entitlements?
• Does the Work IQ MCP tenant policy match what you intend to allow?
• Are agent updates that request new permissions routed through the Agent Registry for review?

Orange France: the unglamorous layer behind the promise​

The Orange France story is not an AI announcement. In a November 3 announcement, Accenture said Kbrw and Accenture were selected to modernize order and inventory management across online, retail and B2B channels. The first phase aims to give customers more accurate availability and more reliable delivery dates for smartphones, SIM cards and telecom equipment. It also supports pre-ordering and inventory reservation.

The architecture splits transactional and operational duties. Orange France's ERP stays the main record for financial and transactional processes. Kbrw's Order Management System handles orders, inventory and fulfillment in real time. Accenture is integrating it with the existing environment and testing it across sales, inventory and fulfillment.

The rollout is phased:

• Retail-store replenishment orders come next.
• B2B and enterprise orders are planned for 2027.
• After-sales services follow over time, covering returns, repairs and refurbishment.

Supply Chain Director Lionel Benezech describes a consistent view of stock that is in stock, in transit or expected. He also says the system automates scenarios that previously needed manual intervention. These are company statements, and no measured results were published.

The lesson holds regardless of the vendor. A chatbot or agent can only promise what the back-end systems can deliver.

Zendesk and Oracle: two takes on autonomous service​

The CX Today text supplied here cuts off in the Zendesk section, so the following comes from the companies' own materials.

Zendesk. Zendesk announced on November 5 that it was named a Leader in Gartner's 2026 Magic Quadrant for CRM Customer Engagement Center for the second year running. Its release describes an Autonomous Service Workforce, voice AI agents and employee-service agents that work through Slack or Microsoft Teams. It reports that 300 early-access customers drove 6 million workflow executions in five weeks. Zendesk estimates that saved 500,000 hours. A footnote says the estimate rests on a conservative benchmark of five minutes saved per execution. So the hours figure is an assumption multiplied by a count, not a measured result.

Gartner's own disclaimer says its research reflects the opinions of its research organization. It should not be read as statements of fact or as an endorsement of any vendor. Microsoft is among the vendors listed in the same Magic Quadrant.

Oracle. Oracle's October 21 post describes Service Execution Workspace in Oracle Fusion Service 26D. Representatives can review AI-prepared resolutions and accept or modify them before anything reaches the customer. Oracle says the workspace respects existing access controls. It also says the workspace does not replace the full service-request experience, and that knowledge authoring, SLA management and stuck-request management are not part of the release. I could not retrieve Oracle's posts directly, so these details come from the research packet and should be read as Oracle's own description.

What this means for IT teams​

Taken together, the four stories show different layers of "execution":

None of this shows the products are interchangeable, generally available to every customer, or consistently accurate. The common thread is governance. Every connector adds a place where permissions, data quality and accountability have to hold. In a Microsoft 365 shop, that work lands on whoever manages the admin center, Entra and the Agent Registry.