On June 10, 2026, Adobe announced the general availability of CX Enterprise Coworker — a milestone that marks a strategic shift in enterprise software. The Coworker works with “persistent agency”: it keeps long-term business goals in mind, breaks them down into multi-step plans, and executes them after approval, rather than merely reacting like earlier copilot assistants.
This is operationally relevant because many organizations have so far been running isolated AI tools whose results are hard to measure and do not lead to revenue growth or better customer retention. 44% of enterprise decision-makers name customer engagement as the top deployment area for agentic AI — yet without real orchestration, the benefit remains fragmentary.
Technical architecture
The defining architectural decision is openness: Adobe bases the platform on open standards such as the Model Context Protocol (MCP) and the Agent2Agent Protocol. This is not a rhetorical point, but relevant for integration: the Coworker runs on NVIDIA OpenShell, a policy-driven agent runtime for on-premises or cloud deployment. For DXP and CMS implementers, this means: the agents orchestrate across system boundaries, not inside a black box.
The system uses business context such as marketing briefs and past campaign performance to plan campaigns, create on-brand content, and design customer journeys in one workflow via natural-language prompts. This considerably reduces manual orchestration mapping.
Implications for projects
Three concrete points for CMS and DXP projects:
1. Clarify data integration up front: With fragmented data across marketing, sales, and service systems, it is difficult for AI tools to act on CX; without integration into real workflows, AI remains experimental. If you want to use Coworker productively, you have to build Adobe Experience Platform (AEP) as a real customer data platform, not as a decorative layer.
2. Take governance seriously: Agents run in policy-based, containerized sandboxes to keep execution controlled, observable, and auditable. Brand safety and compliance are not optional. For Magnolia or Strapi users: integrations with Adobe via APIs must be governance-aware from the start.
3. Treat agent skills as code: Skills specify not only how an agent works, but also design guidelines, acceptable parameters, and validation loops — they are living code that evolves with team learnings. This is not set-and-forget.
Assessment
Adobe positions the Coworker as the central orchestration layer for enterprise AI in CX, as a control system in an operational environment where AI executes ongoing marketing and engagement processes with measurable impact. Salesforce has pushed something similar with Agentforce, but Adobe's advantage lies in depth: more than 20,000 global brands have built their operations on Adobe; the Coworker rests on decades of domain expertise in data, content, and customer journeys.
This is not innovation for innovation's sake. While generative AI accelerates content creation, agentic AI enables outcome-driven workflows — customers need precision, brand governance, and seamless integration, not just speed.
Recommendation
For teams working on CMS or DXP projects with Adobe: evaluate now whether your AEP architecture is designed for multi-agent orchestration. This does not mean buying Adobe's Coworker right away, but reviewing the data foundations, the governance processes, and API interoperability. If these do not fit, even a powerful agent will fail at customer experience orchestration.
If you build these foundations consistently, you can use Coworker productively. If you ignore them, you will struggle with the same fragmentation problems as today — only faster.
Portalworks supports teams in the strategic planning and technical implementation of such agentic AI projects on Adobe Experience Platform, Magnolia, Strapi, and other DXP platforms. We would be happy to discuss the architecture requirements with you.
---
