SAP CX's leadership has articulated a fundamental tension in enterprise AI adoption: the rush to deploy agentic experiences is masking deeper architectural failures that these systems will inevitably expose and amplify. Balaji Balasubramanian's framing of "vibe coding"—building AI interfaces that appear functional before they are reliable—captures a critical risk for CX teams already managing fragmented data ecosystems. When AI agents operate on disconnected systems and incomplete customer records, they don't simply fail gracefully; they fail at scale and speed, creating compounding operational damage. A support team lead deploying an AI agent into a Salesforce instance with siloed data sources isn't just automating existing processes—they're automating inconsistency, potentially offering promotions the business cannot fulfil, recommending unavailable products, or violating consent preferences at the velocity of conversational AI. The implication is stark: teams that have deferred foundational data governance work in favour of quick AI wins are now facing a reckoning where their technical debt becomes a customer-facing liability.
The strategic shift SAP is advocating moves CX leaders away from deployment-first thinking toward what might be called "foundation-first agentic design." This requires three interconnected capabilities: trusted, governed data; connected operational systems; and explicit governance frameworks that define what agents can access, when human intervention is required, and how decisions are audited. For Zendesk administrators and CX consultants, this reframes the AI conversation entirely—the question is no longer whether to deploy agents, but whether your underlying customer data architecture can sustain autonomous decision-making without creating compliance or trust violations. Teams already running Agentforce or similar platforms should be conducting urgent audits of data consistency, system connectivity, and governance controls, because the longer agents operate on weak foundations, the more customer-facing damage accumulates. The winners in this phase of AI adoption will be organisations that treat agentic CX as a forcing function for enterprise data discipline, not as a shortcut around it.
As AI agents become embedded in CX, deploying them on fragmented data and disconnected systems turns them into a liability at enterprise scale. The urgency to ship AI-powered experiences is creating a dangerous gap between what organizations think they are delivering and what the business can stan