Though some recent studies cast doubt on the ability of AI agents to complete complex tasks, ServiceNow boasts that its bots are better, because they can rely on 20 years and 80 billion workflows worth of experience. The underlying model, they say, is just a small part of the product.

ServiceNow’s president and chief operating officer Amit Zavery knows Vishal Sikka, a longtime veteran of several enterprise tech companies and is aware of the paper Sikka recently co-wrote casting doubts on the efficacy of AI agents. But he says that Sikka’s conclusions don’t apply to his company.

“I know Vishal very well,” he told The Register. “We talk quite often …. It’s a small industry. And we argue. Vishal has some ideas. It doesn’t mean all of them are right. Some of them make sense, and we learn from each other in some cases too.”

Zavery told The Register that ServiceNow’s agents rely less on an underlying LLM than the ones studied by Sikka and co-author Varin Sikka for their July 2025 paper.

Sikka is currently the CEO of Vianai Systems as well as an Oracle boardmember and advisor to Stanford University on the subject of AI, and has a long history in enterprise tech, having served as the CEO of Infosys and CTO of SAP. He and his co-author – his son Varin – wrote that LLMs and agents based on LLMs hallucinate when ordered to complete tasks that are too complex for their model. The paper, which was surfaced by Wired last week, doubts whether AI agents are capable of carrying out multi-step tasks or checking to see if a task was done correctly.

“Our argument, in essence, is: if the prompt to an LLM specifies a computation (or a computational task) whose complexity is higher than that of the LLM’s core operation, then the LLM will in general respond incorrectly,” the paper states. “Also, any LLM-based agent (i.e. in the Agentic AI sense) cannot correctly carry out tasks beyond the complexity of the LLM.”

Zavery said that’s not a problem for ServiceNow, which has run 80 billion end-to-end workflows for customers to understand what the final result should be. Because of that, ServiceNow can verify whether the agent performed it correctly.

“When we build our products, we have to guarantee outcomes,” Zavery told The Register. “We cannot have things which are unpredictable or very probabilistic without having some kind of controls built in. This is the value of our 20 years of workflow data, and our understanding of what the results should be. So when we produce something, even with the Build Agent, it’s not just that we use the LLM and put a wrapper around it. There’s a huge amount of IP we build on top of that.”

On its website, ServiceNow says that it offers out-of-the box AI agents as well as build-your-own agents. Zavery estimates that when ServiceNow builds an agent for customers, the underlying LLM is carrying about 10 percent of the lift while ServiceNow’s own IP and internally developed software based on billions of business transactions is carrying the rest.

“Ninty percent is the IP and the technology we are building from governance, compliance, predictability of the outcome, who needs to monitor it, what data you need to track it, so when someone says ‘I want to onboard an employee,’ we guarantee the employee gets onboarded to the right applications and the right systems because we know what good looks like,” he said.

Zavery said that the company has a new offering with Anthropic that is designed for ServiceNow app developers. Last year, the company rolled out its build agent to let users vibe code bolt on applications that improve their ServiceNow experience. That resulted in 6,000 applications being introduced to ServiceNow’s portal.

Now, the company is adding Anthropic’s Claude Code beneath the covers of its build agent to energize that effort and Zavery said that they are expecting a four fold increase in the number of applications that should be produced.

“We are adding a lot of context in terms of security, domain, as well as prepackaged workflows, so you can expand and extend those things easily using the build agent and we’re powering that through Claude,” he said. “What this does is now evolves it into much more of a vibe coding kind of mindset. You wake up in the middle of the night and want to build something on ServiceNow. That’s our market with this one.”

ServiceNow is also looking to work with Anthropic in health care and life sciences, to better streamline case management, claim authorization, research analysts, where Opus has shown effectiveness in understanding that industry.

“We’re adding the data model as well as the workflow elements of it because we’ve been building technologies and products for life sciences and health care and we try to bring them together into one integrated offering for our customers in those regulated industries,” he said. ®