In a recent TFiR interview, Swapnil Bhartiya spoke with Nenshad Bardoliwalla, Group Vice President of Product Management for AI Products at ServiceNow, about the architectural and organizational shift required to move from AI assistance to full autonomous execution in enterprise IT operations.

THE SHIFT FROM TASK AUTOMATION TO ROLE AUTOMATION

What autonomous workers actually do: Most enterprise AI deployments today automate individual tasks or sequences of tasks within a process. ServiceNow’s autonomous workers operate at a fundamentally different level—they automate entire roles and responsibilities within an organization. A service desk autonomous worker doesn’t just summarize a ticket or suggest a resolution; it triages the incident, categorizes it, queries the knowledge base, generates a response, and closes the request—just as a human agent would.

“What ServiceNow has done is moved us into the next wave of this revolution, which is to automate roles and responsibilities that exist in an organization. If you imagine that a given person in an organization like a service desk worker is responsible for many different tasks—triaging incidents, categorizing those incidents, being able to look up information in a knowledge base and respond to somebody—all of those are individual processes. So what ServiceNow has done is said, because of the power of our ability to combine deterministic workflows, we can insert the agentic technology.”

How the autonomous worker is deployed: The system is assigned a literal user ID within ServiceNow’s platform. It pulls tickets from the same queue as human agents. It is assigned work using the same assignment mechanisms already in place. When scoped correctly to priority 3 and priority 4 tickets—the high-volume, lower-risk incidents that consume disproportionate team time—autonomous resolution rates can exceed 80%.

DETERMINISTIC WORKFLOWS AS AI GUARDRAILS

Why guardrails matter more than capability: A common failure mode in enterprise agentic AI deployments is building powerful systems without governance. ServiceNow’s differentiation is the inverse: the agentic layer is infused into a deterministic workflow backbone that enterprises have already been operating for years. RBAC, escalation paths, regional routing policies, assignment rules—all of these continue to govern what the AI can and cannot do.

“The part that we don’t want hallucinated is: did we actually close the correct request? Did we route it to the right person? Did we shut down the correct system? All of those types of very real-world, deterministic use cases—you want the agentic technology to be governed by that framework. Even if the agent hallucinates, the guardrails, policies, and compliance procedures prevent bad outcomes from happening.”

Checks and balances as architecture: Bardoliwalla explicitly frames the ServiceNow approach using the analogy of constitutional checks and balances—the AI operates with significant autonomy but within a system designed to prevent runaway decisions.

THE REAL BARRIERS TO PRODUCTION DEPLOYMENT

Technical barriers: The most underestimated requirement for autonomous IT support is knowledge base quality. For a system to autonomously answer questions, the knowledge base must be well-structured, non-overlapping, and internally consistent. Many enterprise knowledge bases fail this test: articles contain conflicting answers, excessive hyperlinks that create lookup chains, and inconsistent metadata. The same confusion a human agent would experience when reading contradictory documentation is experienced by the AI agent.

“Your knowledge base has to be very high quality—and a lot of times, our knowledge bases are not. There’s overlapping metadata. There are inconsistent answers to questions. People put many links into the knowledge base article. So having a very well-structured knowledge base, having a very well-structured service catalog, is really important.”

Organizational barriers: The harder problem is human trust. Bardoliwalla describes the same psychological arc enterprise teams go through as new drivers experience with self-driving cars—initial anxiety, brief releases of control, rapid recapture of the wheel after any unexpected behavior, and only gradual expansion of autonomous latitude over repeated positive interactions.

“Allowing a system to work autonomously means that human beings have to have a level of trust such that they are willing to allow the system to do the work for them—and that trust has to be earned. The way that we humans work is we need to be part of the change management process. We need to see the machine making the right decisions. And once we have a comfort level, we start allowing progressively increasing amounts of autonomy.”

THE MOVEWORKS ACQUISITION AND EMPLOYEEWORKS

What MoveWorks brings: The acquisition addresses employee-facing productivity—a parallel but distinct surface area from IT ticket resolution. MoveWorks provides a high-powered conversational AI layer with enterprise context: it can book travel, look up time-off policies, query payroll, and access procurement systems like Ariba or expense platforms like Concur—all through a single interface, with enterprise security and access controls intact.

What EmployeeWorks adds: EmployeeWorks extends the MoveWorks foundation with a next-generation portal experience. Rather than a static, list-based employee portal, EmployeeWorks generates dynamic widgets in real time based on user requests. A question about getting to a different office generates a map. Routine benefit enrollments surface proactively. The system becomes a persistent, contextual workspace rather than a navigation menu.

“Employee works becomes that single place where if you need to converse, if you need to search, if you need to push proactive notification, and you need a workspace that covers all your enterprise use cases as an employee—EmployeeWorks brings the full power of MoveWorks and ServiceNow’s historical strength in employee-facing solutions together in a single environment.”