UiPath AI agents blend with RPA amid industry hype, doubts

UiPath's controlled agency

However, Ward-Dutton said UiPath's approach to AI agents differs from some competitors' in taking a pragmatic view of where the technology is useful to enterprises while integrating it with existing IT automation systems that are still more effective for certain tasks.

Mark Geene, senior vice president and general manager of AI products and platform at UiPath, called this strategy "controlled agency."

"You can say, 'I'm never going to let the agent go update and make a payment out of my SAP system. I will always rely on a reliable API or UI automation that I know and trust to do that,'" Geene said. "But for the steps in between, I'll let the agent make recommendations or do analysis."

A UiPath Maestro process intelligence feature helps users start with limited AI agent autonomy and expand it incrementally.

"Our process intelligence looks and says, 'Hey, you're passing this to a human to review every time, and they're not making any changes. They're just accepting everything the agent's doing. Would you like to give the agent now more responsibility?'" Geene said. "We'll still send it to a human if we're uncertain, and we give the agent some [boundaries] and context, because we have these very detailed evaluations through our platform of how an agent is running and how accurate it is."

UiPath's Maestro and Agent Builder include guidance for users, in some cases delivered through AI assistants, about which parts of business workflows AI agents are best suited for and low-code/no-code templates to help users get started. This is similar to Atlassian's approach this month, when it embedded its Rovo AI tools into its broader cloud subscriptions and updated its Studio automation builder to include more guided workflows.

LLMs get a rethink as AI agent workloads grow

AI agents sprang from generative AI, and both trends emerged in an explosion of industry hype beginning with OpenAI's ChatGPT in 2022. But in the real world, the hype has yet to live up to its promises of agentic AI automation magic, according to Ward-Dutton.

People are confused about what AI agents are, what they're capable of, what they're not capable of and what that means in terms of where AI agents fit in the context of what people have already.

"People are confused about what AI agents are, what they're capable of, what they're not capable of and what that means in terms of where AI agents fit in the context of what people have already," he said. "So much of the content and the promotional stuff that gets shared online, it kind of assumes that because this stuff is AI-based, then it's worth doing, regardless of whether it's solving a problem that's maybe been solved already using more established, more predictable, lower-cost technology."

This confusion is reflected in recent industry surveys. A December report from EY and a January report from Informatica found that business leaders had difficulty implementing AI, especially in demonstrating its ROI and value to the business. A survey of 300 e-commerce business leaders conducted in March by content management system vendor Storyblok found that they have spent an average of $369,916 on AI in the past year, but 32% reported only slight improvements to customers' digital experiences.

Some vendors in observability and cybersecurity are working to develop specialized large language models and small language models that can perform faster, deliver results optimized for specialized applications using more targeted training data sets, and run more cheaply than large frontier LLMs, but Ward-Dutton said that work is still very much in progress.

"In two to three years, we'll get to the point where AI agents can play more expanded roles in workflows and processes," he said. "But right now, with hybrid implementations, we're going to see all these things working as parts of the overall toolkit along with existing automation technologies like API integration, RPA and workflow automation."

Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.