ServiceNow Lets Employees Build Their Own AI Agents. The Hard Part Is Trust.

ServiceNow wants the people closest to your company's daily problems to build the AI that solves them, and it is betting none of them need to write code. On October 15, the company announced major advances to ServiceNow EmployeeWorks, its employee experience platform, centered on a third-generation employee-centric AI harness that lets ordinary employees build and customize AI agents for their own jobs.

The pitch is aimed squarely at a problem almost every large employer knows by heart. Enterprise AI investment keeps rising, yet most employees still cannot build the AI their job actually needs. The reasons are structural: IT backlogs, developer constraints, vendor lock-in, and fragmented tools mean the people who understand the work best, the ones who live inside the process, cannot build AI that takes secure actions inside company systems. ServiceNow says EmployeeWorks changes that equation.

What changed

EmployeeWorks now gives employees no-code tools to build what ServiceNow calls Productivity Assistants, specialized AI assistants shaped by a role or a context, assembled by nonprogrammers within boundaries set by administrators. The company also ships ready-made examples: Talent Insights Assistant and Manager Assistant for HR service delivery, plus a Research Analyst and a Calendaring Assistant as standard EmployeeWorks fixtures.

Alongside assistants come Productivity Routines, agents built for recurring tasks. A routine runs in the background, follows the same steps every time, and delivers the result without human intervention. The distinction matters: assistants answer, routines finish. ServiceNow's argument is that the second category is where real hours get saved.

The connective tissue is the third-generation harness itself. ServiceNow describes it as providing intelligent context, trusted decision making, and robust enterprise actions so that agents orchestrate work safely. Bhavin Shah, senior vice president of the company's employee experience group, said the platform combines agentic planning with governed and auditable workflows, using the employee's own context as the foundation. The phrasing is deliberate: the selling point is not a smarter model, it is a safer leash.

The bottleneck was never the model. It was the wait: for the IT backlog, for legal's review, for someone with code access to care about your problem.

ToolSmith and the vibe-coding bet

For developers, ServiceNow introduced ToolSmith, an agentic developer environment for EmployeeWorks where tools get built by describing them in natural language. The company says ToolSmith takes complex tool-building from days to minutes, runs at enterprise scale, and is trained on more than 75,000 APIs. When something breaks, it retrieves the logs and suggests a fix, which is the practical version of vibe coding that enterprises have been demanding: speed with an audit trail.

The API count is the telling detail. A no-code builder is only as useful as the systems it can reach. By training ToolSmith across tens of thousands of APIs, ServiceNow is trying to solve the integration problem that quietly kills most citizen-developer programs: the moment a new agent needs to touch the payroll system and discovers it cannot.

A new Conversations API extends the whole stack into the applications employees already use, from internal apps to kiosks, so assistants and routines meet workers where they are instead of demanding yet another portal.

Governance is the actual product

Read the announcement twice and the real product starts to look like control, not creation. The productivity assistants, routines, harness, ToolSmith, and Conversations API all arrive in controlled availability, meaning customers get them in stages under ServiceNow's supervision. Employees build within administrator-set boundaries, and every agent is supposed to act inside signed, auditable workflows.

The same theme shows up in a related announcement this week: AI Passport, a system of signed credentials defining what an AI agent is allowed to do, across both ServiceNow and third-party agents, with controls that can restrict or stop activity outside those boundaries. NVIDIA is a design and launch partner, and the broader package adds AI discovery, monitoring, and cost controls, expected later this quarter.

This is the correct priority order. The history of enterprise software is a loop: a powerful new capability arrives, employees use it enthusiastically, and the security team discovers the shadow version six months later. No-code agents multiply that risk. If every employee can spawn an agent that reads email and writes to finance systems, the company needs a credential layer for agents before it needs anything else. Governance is not the brake on adoption. It is the adoption.

The ROI question nobody can dodge

All of this lands in a market where AI's business value is still largely a promise. McKinsey's latest State of AI report, cited in an industry release this week, found that only 37 percent of respondents said AI has contributed positively to EBIT, essentially unchanged from 2025. Just 6 percent qualified as AI high performers, meaning they attribute at least 5 percent of EBIT to AI and report significant value from it.

That gap explains the framing war. Druid AI, which expanded its own agentic platform this week with voice and governance features ahead of its October 20 Symbiosis event in London, argues that the industry should stop measuring activity and start measuring completed work: appointments booked, hours saved, revenue generated. ServiceNow's version of the same argument is the routine: value is not an assistant that answers questions, it is a workflow that finishes on its own.

Jacqui Canney, ServiceNow's chief people and AI enablement officer, framed the pitch around time, saying AI's real promise is to close the gap between intention and action, so employees spend less time on coordination and more time on customer problems that need human judgment.

What to watch

The announcement's honesty is in its availability status. Controlled availability means the no-code agent workplace is not here yet, it is being rolled out carefully, which is the right speed for software that acts inside payroll, HR, and finance systems. Watch for three things over the next two quarters: whether employees actually build agents or leave the toolkit to IT, whether the administrator guardrails hold up under real use, and whether any customer can point to a completed outcome, hours saved or errors avoided, rather than another chatbot deployment.

ServiceNow is making a structural bet: that the enterprise AI race will be won not by the best model but by the best wrapper around human intent, the safest way for a nontechnical employee to turn "I need this done" into done. The models are commodities now. The leash is the product.