As artificial intelligence moves into diabetes care, the key question for physicians may be less whether technology can support insulin titration than how it changes the work surrounding it.

UpDoc’s recent FDA clearance highlights a growing role for AI-supported insulin management, while keeping clinical dosing decisions within clinician-defined protocols.

The platform includes a patient-facing large language model (LLM), but the insulin instructions generated are based on treatment parameters established by the patient’s health care provider.

The software is intended to assist patients with insulin therapy management by providing instructions based on a health care provider-specified treatment plan. UpDoc said the platform will initially be deployed at Cleveland Clinic, Allegheny Health Network, and UCSF Health.

The FDA clearance is notable for the role artificial intelligence (AI) plays—and does not play—in the platform, according to Johnson Thomas, MD, FSCE, FEAA, Editor-in-Chief of AACE Endocrine AI. “It represents a notable step forward for diabetes clinical AI, but also a considerably narrower one,” he said. 

For endocrinologists, the key distinction is between the conversational technology interacting with patients and the software determining insulin doses, noted Dr. Thomas. Although UpDoc incorporates a patient-facing LLM, FDA documentation indicates that insulin instructions are generated based on clinician-defined treatment parameters rather than independently determined by the LLM.  

Clinicians establish the treatment parameters, including insulin type, starting and maximum doses, glucose targets, adjustment algorithms, and safety rules. The conversational component can collect information from patients, while the platform's clinical software uses the clinician-defined parameters to generate insulin dose recommendations. 

That separation is explicit in FDA documentation. UpDoc is a Class II drug dose calculator and was cleared through the traditional 510(k) pathway after FDA determined the device was substantially equivalent to the d-Nav Insulin Guidance System, a predicate device that received FDA clearance in 2019. The system provides algorithm-based insulin guidance and has been evaluated in a randomized clinical trial in patients with type 2 diabetes. No clinical testing was performed for UpDoc's 510(k) submission. FDA said the substantial-equivalence determination was supported by software, cybersecurity, and human-factors testing. 

According to Dr. Thomas, the UpDoc platform could shift some routine insulin-management interactions from episodic office visits to ongoing patient interactions through the mobile application.  "The LLM may facilitate communication and information gathering, but the clinical dosing decision remains constrained by parameters established by the treating clinician," he said. 

The FDA also authorized a Predetermined Change Control Plan for the UpDoc device, allowing specified future software modifications. These include certain changes to default clinical values, supported insulin products and existing dosing workflow options, the user interface, and methods of data-input. However, the modifications must remain within the device's intended use and technological characteristics and meet predefined verification, validation, and acceptance criteria. Changes to dosing workflows must remain within the cleared dosing logic. 

Algorithmic insulin titration itself is not new. Dexcom also received FDA clearance in November 2025 for Smart Basal, a continuous glucose monitor-informed insulin dose calculator for adults with type 2 diabetes requiring long-acting insulin. 

These products illustrate different approaches to the same clinical problem, according to Dr. Thomas. Dexcom Smart Basal uses continuous glucose monitoring data, logged basal insulin doses, and health care provider-entered treatment parameters to generate daily basal insulin dose recommendations. UpDoc, by comparison, incorporates a conversational patient interface for collecting patient-reported information and can also receive glucose data from connected devices. 

"Both approaches nevertheless leave the clinician in a central role," said Dr. Thomas. 

For endocrinology practices, the immediate question may be less whether AI can titrate insulin than how protocolized titration changes clinical workflow.  

“Automating routine dose adjustments could address therapeutic inertia and reduce repeated low-complexity contacts between visits,” said Dr. Thomas. “But practices still must determine who establishes protocols, reviews documentation, receives escalations, and assumes responsibility when patients move outside predefined parameters.”