Overview
A credible AI agent used to set a vendor apart. Now nearly every CX vendor offers a platform for building one, and many enterprises are running more agents, tools, and automations than any single team can track. The hard problem has moved, from building agents to getting them to work together.
In this independent report, Opus Research analysts Derek Top, Ian Jacobs, and Amy Stapleton make the case that architecture, not agent quality, will decide the next phase of enterprise CX AI. What matters is whether the agents, people, workflows, and systems already in your stack can operate as one coordinated system.
At the center of their argument is the CX AI control plane: a shared operating layer that carries journey and intent state, identity and consent, policy and guardrails, knowledge governance, and evaluation, testing, and audit across everything acting on a customer's behalf. Without it, the failures are specific and familiar. Context disappears at handoff, and the same policy is interpreted in three different ways. A workflow accepts an action that another system later rejects. Leaders can see that something happened, but not why.
As customers begin arriving with AI agents of their own, that same layer matters even more. Identity, consent, and authority must be established for software you didn't build, and a connected architecture makes that manageable where a fragmented one struggles.
Along the way, the report puts vendor openness claims to a practical test. It closes with a nine-dimension readiness framework you can score today.
Inside the report:
- The five layers of a CX AI control plane
- Verifying a customer’s AI agent, its intent and permissions
- Why fragmentation starts as a series of good decisions
- Three practical tests for any vendor's openness claim
- A nine-dimension readiness framework you can score today
Architecture determines whether your CX AI scales
