Adobe Launches Cross-Platform Agentic AI Coworker for CX Teams

Adobe has introduced an agentic AI "coworker" built for enterprise customer experience teams, and designed it to work across multiple third-party AI platforms rather than a single model provider. The system is being made interoperable with AI platforms from Anthropic, Google, Microsoft and OpenAI, among others, according to Adobe and reporting on the announcement.

The move positions Adobe's agent as a cross-platform layer for CX work rather than a standalone tool tied to one model vendor. Rather than requiring enterprises to standardize on a single AI provider, the coworker is meant to plug into whichever large-model ecosystem a given organization already uses.

This is fundamentally a platform and interoperability story: Adobe is betting that enterprise AI adoption will be multi-vendor by default, and that CX tooling needs to sit above that fragmentation rather than force a choice. For technology and transformation leaders, the announcement signals that "agentic" AI — software that can take on coworker-like tasks rather than just answer prompts — is moving from experimental pilots into vendor-backed enterprise products built for integration at scale.

For CX and service-design leaders specifically, an agent that can operate across Anthropic, Google, Microsoft and OpenAI's ecosystems lowers the switching cost of adopting agentic AI in customer-facing and operational workflows. It also raises a harder question that technology alone won't answer: how much decision authority an organization is willing to hand to an AI "coworker" in service interactions, and how that authority is governed, audited and escalated when it fails.

Cross-platform compatibility is a smart go-to-market move, but it quietly shifts the hard work from Adobe's engineers to each customer's operating model.

Most coverage will frame this as an interoperability win, and it is — but the real test isn't whether the agent talks to five AI platforms, it's whether the humans around it know when to trust it and when to pull it back. Agentic tools fail quietly: they don't crash, they just make confident, wrong calls inside a workflow nobody is watching closely enough. Any operator adopting this should spend less time evaluating model compatibility and more time designing the escalation path, the audit trail and the moment a human re-enters the loop before the agent becomes the default decision-maker by accident.