NYU Langone Builds Bot-Powered Playbook for Clinical Transformation

With clinician-informaticist collaboration, enterprise-wide governance, and a deep bench of automation use cases, NYU Langone Health is pushing the boundaries of robotic process automation and agentic AI.

When a bot at NYU Langone Health identifies a medication reconciliation gap, it doesn’t just log an alert. It initiates a secure chat among the attending physician, quality champion, and frontline staff. The goal: resolve the issue in real-time and discuss how to prevent similar gaps going forward.

This is one of several enterprise-wide deployments of intelligent automation that NYU Langone has launched in recent years. The bots are built by a central team led by Ajay Mansukhani, senior director of RPA, intelligent automation, and QA within Medical Center IT. They are deployed across clinical, research, education, and administrative functions, in close partnership with clinician informaticists like Dr. Jonah Feldman, medical director for clinical transformation and informatics.

From RPA to Agentic AI

Initially focused on traditional RPA tasks—rule-based, repetitive activities—NYU Langone’s automation program has evolved. According to Mansukhani, the team is now moving toward agentic AI, which introduces a decision-making layer to automation workflows. “Agents have the power to make decisions on the fly,” he said. “They aren’t just following scripts.”

The line between bots and agents, he explained, is drawn by the level of embedded intelligence. While bots tend to follow rules-based commands, agents can adapt to variable conditions and make autonomous decisions based on contextual parameters. “What we’re doing now is taking our bots, adding intelligence to them, and trending toward agency,” he said.



Feldman explained that some of their use cases fall into the category of “automation of the obvious”—taking repetitive work off human plates. But others were long-standing ambitions never realized due to lack of resources. The quality and safety huddle bot, for instance, enables a level of coordinated response that would be impossible to sustain manually.

“This was always on our dream list,” Feldman said. “Now we have scalable resources to make it happen.”

The system has also seen measurable improvements. Medication reconciliation accuracy increased from 86% to 96%. First contact provider accuracy improved to 99%. Operating room cleaning times have been reduced, freeing up roughly half an hour of surgical time per week.

Balancing Ambition with Governance

NYU Langone runs its automation program like an internal consultancy. While many requests bubble up from end users, Mansukhani’s team also proactively educates staff about what’s possible—demonstrating bots as virtual assistants, not replacements. “We want users to see bots as partners,” he said. “When the job is done, the bot reports back. It’s a handshake, not a handoff.”

Still, not every use case makes the cut. The team evaluates requests through a structured process that includes business viability, process optimization, and return on investment. “We don’t build on top of broken processes,” Mansukhani said. Each opportunity must go through architecture, security, and business reviews before being approved.

Feldman noted that some proposed automations are rejected not because they are unhelpful, but because the underlying clinical infrastructure isn’t ready. In one case, the team had to pause a bot rollout due to the lack of foundational clinical decision support in the EHR. “We realized the groundwork wasn’t there,” he said. “We had to go back, build up the basics, and come back to automation later.”

Requests are evaluated and prioritized through established IT governance processes. This ensures that clinical needs are addressed through the most appropriate solutions—which may or may not involve bots. “If something comes in as a bot request, it might go to clinical decision support,