AI as a Workflow Tool, Not a Decision Maker

In part 2 of a recent discussion with Dermatology Times, Faranak Kamangar, MD, discussed how clinicians can strike a balance between leveraging artificial intelligence (AI)’s efficiency and maintaining clinical rigor. Kamangar, founder of DermGPT and president of the San Francisco Dermatology Society, highlighted how AI is being integrated into clinic to improve workflow efficiency.

A central challenge with large language models is their tendency to generate confident, fluent responses—even when underlying data may be incomplete or uncertain. Kamangar noted that DermGPT took a more conservative approach, sometimes declining to answer questions altogether. There’s a happy medium,” she explained, emphasizing the importance of evidence-based responses paired with transparency about uncertainty. Guardrails, including tools that highlight unknowns or potential pitfalls, are critical to minimizing hallucinations while maintaining usability.

Adoption trends reflect growing interest but measured reliance. With thousands of dermatologists now using DermGPT—and a subset integrating them heavily into daily workflows—Kamangar observed that most clinicians are not turning to AI to redefine how they practice medicine. Instead, they are using it to improve efficiency, particularly in documentation, communication, and time management. The goal, she noted, is to tenfold productivity without fundamentally altering clinical judgment.

Patient communication represents a practical and relatively low-risk application. AI can assist in drafting responses to common inquiries or triaging minor concerns, such as mild postoperative redness, by providing structured guidance and clear thresholds for escalation. However, Kamangar drew a firm boundary: diagnostic decision-making and prescribing remain outside the scope of current AI capabilities in routine practice.

To safely integrate AI, she recommends treating it similarly to other members of the care team. Just as medical assistants, nurses, and advanced practice providers operate within defined protocols, AI systems should be governed by explicit rules outlining permissible tasks. Emerging regulations are likely to reinforce these boundaries, particularly around higher-risk clinical functions.

Despite rapid advancements, Kamangar cautioned against overestimating AI’s capabilities. “It’s very impressive technology,” she said, “but it’s not without flaws.” Understanding those limitations—and designing workflows accordingly—will be essential as dermatology continues to adopt AI-driven tools.