ServiceNow’s Autonomous Workforce: From AI Assistants to Governed Virtual Employees

ServiceNow is putting hard numbers behind a vision many enterprises are still only piloting. Inside its own organization, the company says 90% of employee IT requests are now resolved autonomously, and those cases are closed 99% faster than when handled by human agents. With a new architectural framework called Autonomous Workforce, a new employee-facing product named EmployeeWorks, and a concept it calls “role automation,” ServiceNow is aiming to export that model to enterprise customers that have struggled to move from AI pilots to real execution.

For enterprise IT and security leaders, the pitch is not just faster help desks. It is about embedding governance, permissions, and workflow logic directly into the AI execution layer so that “virtual employees” can act with the same guardrails as human staff—without asking organizations to relax their compliance standards.

From tickets to virtual employees: What ServiceNow is actually announcing

The latest announcement combines several strands of ServiceNow’s evolution into a single strategy aimed at production-grade AI execution.

First is EmployeeWorks, an employee-facing product that lets staff describe an issue in plain language and have it resolved without ever filing a traditional ticket. Instead of deciding which tool, form, or portal to use, employees use a single entry point, and the system routes to the right capabilities in the background. EmployeeWorks is built on ServiceNow’s December acquisition of Moveworks, which already had 5.5 million enterprise users on its AI assistant platform.

Second is Autonomous Workforce, a framework for having AI execute work end-to-end, not just suggest answers. This is where ServiceNow is trying to move beyond “assistive” AI (which drafts responses or suggests steps) into AI that actually performs the steps: resetting passwords, provisioning software access, troubleshooting network issues, and more.

Underpinning both is role automation, the architectural layer that defines how these AI “specialists” operate inside the enterprise. Rather than letting agents dynamically reason their way into permissions, role automation binds AI workers to the same access controls, configuration context, SLAs, and entitlements that govern human roles on the ServiceNow platform, from the moment those AI roles are created.

Together, these pieces are meant to address a familiar pattern: after years of experimentation, many organizations still have AI agents that can identify a problem and recommend a fix, but then hand execution back to humans because the agents either lack the necessary permissions or the organization does not yet trust them to act autonomously inside a governed environment.

Why most agentic AI pilots stall at the execution layer

Many enterprises have invested the past three years building and piloting AI agents. Those agents can often parse natural language, classify requests, recommend next steps, or surface relevant knowledge. But at the moment of execution—creating accounts, modifying systems, changing configurations—projects frequently stall.

The barrier, according to the framing around ServiceNow’s announcement, is not primarily capability. Modern models can reason about tasks and propose actions. Instead, the friction is around governance and workflow continuity:

• Agents do not have the permissions needed to complete tasks without human intervention.
• Security and compliance teams are uneasy about agents deciding, at runtime, which systems to touch and how far to escalate their own privileges.
• Organizations treat governance as an overlay—policies, documents, review boards—rather than something embedded directly into the execution architecture.

The result is an awkward hybrid. AI speeds up classification and recommendations, but humans are still the only trusted actors for completing work. That cuts deeply into the ROI promised by automation and keeps IT and operations teams stuck in manual loops.

ServiceNow is explicitly targeting this execution-layer gap. Its internal claim—90% of IT requests autonomous, 99% faster resolution—serves as a reference point for what it believes is possible when execution is designed to be governed and auditable from the outset.

Inside role automation: Inheriting governance instead of reasoning into it

Role automation is ServiceNow’s proposed answer to the governance problem, and it differs from the task-oriented agents many enterprises are already experimenting with.

Typical AI agents are goal- and task-oriented: they are given an objective, reason through the steps, and decide at runtime what information and actions they need. In consumer or lightly governed settings, this flexibility is a strength. In enterprise environments with strict audit, compliance, and access controls, it is a liability.

ServiceNow’s role automation model flips this: the AI specialist does not negotiate or infer its permissions—it inherits them. From the moment an AI role is deployed on the platform, it is bound by the same structures that apply to human workers:

• Existing access control frameworks
• CMDB (configuration management database) context, defining systems, relationships, and ownership
• SLA (service level agreement) logic, dictating response and resolution expectations
• Entitlement rules that define which roles can do what, and where

Within this model, the AI specialist:

• Cannot exceed its defined scope.
• Cannot self-escalate privileges during a task based on what it “learns.”
• Operates with pre-inherited governance baked into its role definition, not as an afterthought.

ServiceNow describes a three-tier structure for agentic work on its platform:

• Task agents handle individual automation steps.
• Agentic workflows mix deterministic automation with probabilistic, AI-driven decisions to orchestrate those steps.
• Role automation sits above both as a virtualized employee role—a governed entity with defined responsibilities, permissions, and constraints.

The first product built on this approach is the Level 1 Service Desk AI Specialist. It is designed to handle common IT requests end to end—such as password resets, software access provisioning, and network troubleshooting—documenting each resolution and escalating only when a request falls outside its defined scope. For IT leaders, this is a concrete example of what a “virtual employee” looks like when bound by enterprise-grade governance.

EmployeeWorks and Moveworks: One entry point instead of fragmented tools

While role automation focuses on execution and governance, EmployeeWorks addresses a different but related problem: fragmented employee experiences across multiple AI tools.

Today, many enterprise AI assistants—whether from Microsoft, Google, or other vendors—require employees to know which tool to use for which problem. This creates context-switching and friction just to get basic tasks done.

Moveworks, acquired by ServiceNow in December and now the basis for EmployeeWorks, was built around a single entry point that resolves that ambiguity. Employees describe their issue in natural language, and the system automatically routes to the right underlying workflows and capabilities. Before the acquisition, Moveworks already supported 5.5 million enterprise users on this model.

Bhavin Shah, founder of Moveworks and now SVP at ServiceNow, underscored the fragmentation problem in a briefing with press and analysts. Over the last two years, he noted, organizations have rushed to adopt AI, but in many cases that has resulted in “fragmented tools, disconnected AI experiences and employees bouncing between systems just to get simple things done.”

EmployeeWorks, layered on top of Autonomous Workforce and role automation, is positioned as a unified front door to this governed AI execution layer—abstracting complexity from employees while giving IT and security teams tighter control over how work is actually carried out.

Governance, not hype: Lessons from CVS Health’s AI deployment

The emphasis on governance is not purely theoretical. Alan Rosa, CISO and SVP of infrastructure and operations at CVS Health, brought a large-enterprise, highly regulated perspective to the discussion. He manages AI deployment across a 300,000-employee healthcare organization where compliance is non-negotiable.

CVS Health was already a customer of both ServiceNow and Moveworks before the acquisition. Rosa said the combination is encouraging and that the potential is “coming to life,” though CVS Health has not publicly committed to deploying Autonomous Workforce itself.

Rosa’s framework for scaling AI aligns closely with ServiceNow’s architectural claims:

• “Boring is beautiful.” For Rosa, predictable and stable systems are a virtue. He stressed the need to start with responsible, explainable AI—minimizing bias, avoiding hallucinations, and enforcing clear guardrails that everyone understands.
• “Don’t chase butterflies.” Rosa warned against chasing the newest AI capabilities before governance foundations are in place. Instead, he advocated focusing on “gritty, unsexy, operational use cases”—areas where there is clear ROI and real impact on people’s lives.
• Continuous governance. CVS Health runs every AI use case through clinical, legal, privacy, and security review before it ever touches production. Governance is treated as dynamic rather than static. As he put it, “Static review doesn’t cut it when AI is learning and adapting. Wash, rinse, repeat.”

The key point for enterprise leaders is that governance must be embedded in the deployment architecture from the start, not retrofitted after incidents occur. That is precisely what ServiceNow is asserting with role automation: if AI specialists inherit existing permissions and workflow logic, they are structurally less likely to cross governance boundaries than agents that determine their own scope on the fly.

Practical implications for CIOs, CISOs, and automation architects

For organizations evaluating agentic AI—whether from ServiceNow or any other vendor—the underlying question is direct: Does your AI governance live inside your execution layer, or is it sitting on top of it as policy that agents can reason past?

ServiceNow’s Autonomous Workforce and EmployeeWorks are an attempt to answer that question by baking governance, CMDB context, SLAs, and permissions into the same layer that actually performs work. For practitioners, this suggests several practical steps:

• Start with governance architecture, not demo capabilities. Before deploying agentic AI, map where your permissions models, workflow logic, and audit requirements actually live. If these are fragmented or implicit, the AI execution layer will inherit that fragility.
• Define AI roles like you define human roles. Instead of treating agents as abstract tools, design them as virtualized roles with specific scopes, access rights, and escalation paths—then bind them to your existing controls.
• Focus initial use cases on high-volume, well-governed workflows. Service desk Level 1 requests—passwords, access, basic troubleshooting—are natural starting points because the tasks are repetitive, the rules are well understood, and the impact of errors can be contained and audited.
• Integrate continuous review into your AI lifecycle. As Rosa’s “wash, rinse, repeat” mantra suggests, assume models and patterns will evolve. Build clinical/legal/privacy/security review into the ongoing lifecycle of AI use cases, not just initial approval.

As Rosa summarized, “Scale and trust go together. If you lose trust, you lose the right to scale.” For enterprise IT and security leaders, ServiceNow’s latest move is less about a single product and more about an architectural thesis: the only sustainable path to scaled AI execution is to treat virtual employees with the same rigor, constraints, and oversight as their human counterparts—starting at the role definition itself.