Mark Geene understands there’s a big difference between an agent that books your personal travel and one that processes a million claims a month for a health insurer. The distinction matters because the latter must be accurate, auditable, and able to work autonomously inside mission-critical systems.
Geene, senior vice president and general manager of AI Products and Platform at UiPath, arrived at the company in 2021 via the acquisition of Cloud Elements, the API integration firm he co-founded. That deal was early evidence of UiPath’s shift from an RPA provider to a broader platform capable of connecting, governing, and orchestrating AI-driven agents.
Today, the company’s approach reflects a convergence of its RPA heritage, API integration know-how, and a growing focus on deploying AI agents that automate high-volume, repeatable processes that form the foundation of an enterprise’s operations.
Building a Unified Automation Platform
When UiPath bought Cloud Elements, the goal was more than just expanding connectors. It was about bolstering API integration in a platform long associated with RPA.
“When you’re building a workflow, you might be dealing with a well-behaved SaaS app with solid APIs, or a legacy system that needs UI automation,” Geene says. “The key is being able to combine those in one workflow.”
That combination, he explains, isn’t just about connectivity—it’s about governance.
“If I create a connector to Salesforce or SAP,” he explains, “I want to centrally control access: who can use it, which data objects they can see, what they can update. We treat APIs with the same governance and role-based controls we’ve always had for RPA bots.”
This unified governance model, a cornerstone of UiPath’s approach, Geene says, ensures that both API and UI automations are secure and manageable, paving the way for more complex, enterprise-wide automation strategies.
Bringing Agentic AI into Enterprise Workflows
If the first era of automation was deterministic—bots following predefined steps—the agentic era is about handling the “non-deterministic” work that once required human judgment. Geene uses healthcare claims as an example:
“When a claim is denied, a provider might have to research why, analyze the case, and respond. That’s hundreds of millions of dollars at stake for large organizations,” he notes. “We can now have an adversarial agent look for reasons a claim might be denied, and another agent prepare a response. Humans remain in control, but they’re reviewing recommendations instead of doing all the research.”
The key, Geene stresses, is scoping agents to deliver measurable ROI without overloading them. “If you give an agent too much to do, you lose accuracy and predictability. If you give it too little, you won’t see real value. The art is in breaking down processes to find that balance.”
Orchestration: From Bots to Agents
UiPath’s orchestration capability—long used to manage fleets of RPA bots—now governs agents as well.
“I don’t think there’s another company that’s governing as many autonomous workers as UiPath,” Geene says. “Any month, we manage hundreds of millions of digital workers—historically robots—controlling what they can access, monitoring how they run, and shutting them down if needed.”
The same orchestration applies to agents, whether built natively in UiPath or in other frameworks. “We can manage your own coded agents, low-code agents, or even third-party ones like LangGraph or CrewAI. They all run under the same governance and compliance model.”
This continuity, Geene explains, matters when agents are operating in high-stakes environments. “It’s not about making a PowerPoint for you. It’s about working inside your ERP, your loan origination system, your patient records—systems where accuracy, auditability, and control are non-negotiable.”
Trust Builds Adoption
Despite a small uptick recently in doubt around the readiness of agentic AI in business automation, UiPath understands that adoption of the technology depends on confidence in how they handle data. The company’s “trust layer” enforces that.
“No data is used to train a third-party model unless you explicitly allow it,” Geene says. “You choose which models to use for which purposes, and you can bring your own.”
The trust layer also controls what data is sent to models. “If you never want PII sent to a model, we can mask it before sending, then rehydrate the response. And those policies are centrally enforced across the platform.”
UiPath is also pursuing ISO 42001 certification for AI management, which Geene says reflects both vendor and customer compliance needs. “We give you full auditability—so you can explain and prove how decisions were made.”
Measuring and Improving Agent Performance
Building an agent is one thing; knowing whether it’s performing is another. UiPath applies scoring at both “design time” and “runtime.”
“At design time, we measure accuracy, tool use, adherence to policies—not just whether the output is correct, but whether the agent is using the right context and avoiding hallucinations,” Geene explains. “We generate a health score and can recommend specific prompt changes, tools, or guardrails to improve it.”
Runtime scoring then tracks live performance. “You might start an agent with full human review. If, over a month, humans approve 100 percent of its work without material changes, you can increase its autonomy. That’s how you evolve agents into truly productive workers.”
Importantly, the scoring extends beyond UiPath’s own agents. “You can orchestrate a mix of agents from multiple platforms. Our goal is to give you one place to see whether each is doing exactly what you asked.”
Looking Ahead: Scaling Agentic Work
Geene believes the next frontier is deploying agents at scale in bulk, repeatable processes—not just in ad hoc, personal productivity scenarios.
“High-volume processes demand autonomy,” he says. “You can’t have people touching a million transactions a month. That’s why governance, compliance, health scoring, and predictability are where we focus.”
In his view, the enterprise agent market will mature around those principles. “The proliferation of agents will look a lot like the proliferation of apps. The winners will be the ones that can orchestrate across that sprawl, keep it compliant, and prove it’s delivering value.”
And as for what comes after that? Geene hints at deeper integration between agentic AI, deterministic automation, and human oversight—all in the same workflows. “We’re entering a phase where digital workers, human workers, and AI agents operate side-by-side,” he says. “The question now is how to make them work together in a way that’s both productive and trusted.”
