As the industry leaves the summer months, companies are now showcasing why their solutions and approaches are fit for the coming year.
These include the growing expectations for responsible AI adoption as concerns over its capabilities outside of human oversight grow.
As organizations prepare for 2027, leaders will be required to monitor how agents perform in real customer journeys, identify unexpected behaviors, and understand interactions happening beyond their own channels.
Vangie Cleversey, Chief Customer Officer at Acquia, explained to CX Today why organizations must keep observing and learning from AI agents after they have been deployed.
“The companies I worry about are the companies that agents run free,” she said.
“They’re doing great things, they’re providing efficiency, but they’re not observing, watching, and learning from the patterns of those actions.”
Why You Shouldn't Leave Your AI Alone After Launch
Organizations must now recognize that deploying an AI agent should be treated as the beginning of an intelligence cycle rather than the final stage of an AI project.
Once an agent is operating in live customer journeys, there must be clear oversight into whether it is achieving the outcomes it was designed to deliver.
This means deciding your AI goals in advance: “What are the outcomes that you’re trying to achieve through our platform?” Cleversey asked leaders.
“And then where do agents step in to help you with those outcomes?”
This includes defining objectives, permissions, data boundaries and behavioral guardrails while continuing to assess those controls against real-world interactions.
Ensuring continuous observation also enables businesses to uncover behaviors that weren’t visible during testing.
“We’ve seen customer paths for certain issues that we never thought would happen, but they do,” she said.
These unexpected journeys can reveal service, product, and data gaps, meaning organizations without a mechanism for capturing and analyzing those patterns may see efficiency gains while missing cx deterioration evidence.
She emphasized:
“You’ve got to build that loop.”
Introducing a loop connects deployment with observation, analysis, and ongoing improvement, allowing customer interactions to inform how the agent operates and where human intervention is required.
The need for this discipline is becoming more pressing as organizations move agents into more important roles, as McKinsey’s 2026 AI Trust Maturity Survey found that only around 30% of organizations had reached mature levels of strategy, governance, and controls for agentic AI.
