AI Surgical Agent Cuts Documentation Time and Reproduces Published Research in Minutes, Desai Sethi Urology Institute Reports at AUA 2026
At the American Urological Association’s 2026 meeting, Archan Khandekar, M.D., shared new results showing how a HIPAA-compliant artificial intelligence agent, now live in University of Miami operating rooms, drafts operative notes from surgical video, answers complex research questions in minutes and points to why clinical AI must be grounded to be safe.
Archan Khandekar, M.D., assistant professor of urologic oncology at Desai Sethi Urology Institute (DSUI), part of the University of Miami Miller School of Medicine, shared the latest results from the institute’s surgical AI program at the 2026 American Urological Association (AUA) Annual Meeting. The work centers on a single, multimodal AI agent that links intraoperative video with the electronic health record (EHR) and now runs in production within UHealth — University of Miami Health System, drafting operative notes, accelerating clinical research and informing how these tools can be deployed safely.
The agent applies computer vision to robotic operative video and feeds the results to large language models inside the institution’s Epic EHR. It performs real-time surgical step segmentation, instrument tracking and anatomical landmark detection, then generates structured operative narratives directly within the medical record.
Reducing Operative Documentation Time
In the live deployment, the team reported that:
• Median time to a signed operative note fell from roughly 26 minutes to less than two minutes per case.
• An AI drafted report was generated for every case, and surgeons reviewed, edited and signed each one without manual dictation.
• Surgeons made about six edits per case, most often correcting artery versus vein labeling.
• The system reliably captured frequently missed elements, including secondary procedures, unusual instrument use such as intraoperative ultrasound and anatomic variations.
The work builds on the institute’s finding that written operative reports omit roughly one in five clinically meaningful intraoperative events, gaps that affect billing, quality measurement and clinical communication.
From Operative Notes to a Research Engine
The same agent does more than write the record. It can read and analyze it. Through a natural language interface, with no manual data abstraction or programming, the agent reproduced a previously published, peer-reviewed study of warm ischemia time during robotic partial nephrectomy across 61 cases.
In a single query session lasting three minutes and 26 seconds, it recovered every reported correlation at greater than 95 percent concordance. The original, hand-produced work required weeks. It also identified a finding not in the original study. Average postoperative creatinine rose steadily with longer ischemia, from 6.8 percent at 15 minutes or less to 29.8 percent beyond 40 minutes.
“We asked the agent, in plain English, to reproduce a study that originally took our team weeks. It returned every published correlation in about three and a half minutes, then surfaced a relationship we had not reported,” Dr. Khandekar said. “That is the difference between a tool that documents care and one that helps you learn from it.”
Why Grounding Matters for Safety
The team also examined how safely large language models answer urologic questions from patients, comparing a general purpose model with versions grounded in verified medical sources across five high-stakes scenarios scored against AUA and NCCN guidelines. The grounded models scored 25 and 24 of 25. The ungrounded model scored 14 of 25 and produced a clinically dangerous error, confusing a prostate procedure for benign disease with a bladder cancer operation. All models ran locally, with no patient information leaving the institution.
“When we grounded the same class of model in verified urologic guidelines, that dangerous error disappeared,” Dr. Khandekar said. “For anything that faces patients directly, grounding is not an enhancement. It is the safety floor.”
From Innovation to Clinical Implementation
The effort extends a multiyear line of work at DSUI that began with the institute’s 2024 validation of computer vision for measuring warm ischemia time, published in BJUI Compass. A full manuscript on the deployed agent is in preparation for a leading medical AI journal. A multicenter expansion is planned, with the University of Miami serving as one of three national validation sites under a federally funded surgical foundation model program.
“Our work has moved from evaluating new technologies to testing them in real clinical settings, and DSUI is leading the way,” Dr. Khandekar said. “The emphasis now is on validation and understanding what works reliably and where these tools can make a meaningful difference in urologic care.”
The University of Miami conducts this work with a commercial surgical intelligence platform under an institutional agreement.
More from the Desai Sethi Urology Institute
University of Miami Miller School urologists presented AI-powered AR guidance for MRI fusion prostate biopsy at AUA 2026.
Nima Sharifi, M.D., co-authors studies revealing how glucocorticoids influence breast cancer and glioblastoma, opening new treatment avenues.
University of Miami urologists led AUA 2026 sessions on overactive bladder therapy, neuromodulation and precision medicine.
University of Miami researchers report that biomarkers and MRI show similar performance in predicting prostate cancer risk.
Tags: AI, artificial intelligence, AUA2026, cancer research, Desai Sethi Urology Institute, Dr. Archan Khandekar, technology, urology