Artificial intelligence has moved from the research literature into dermatology, and the field now faces a harder question than whether the technology works: how to fit it into everyday clinical practice.

That is the central argument of a new Medscape report, "Dermatology AI Is Here. Now Comes the Hard Part," which surveys the state of machine learning in dermatologic care at the moment the technology crosses from proof of concept to deployment.

The title itself frames the tension. The first half — "Dermatology AI Is Here" — is no longer controversial. Image-classification algorithms for skin lesions have been tested against dermatologists in repeated studies over the past several years, and regulatory pathways have opened for software that assists in lesion assessment. AI-assisted triage, teledermatology image analysis and diagnostic support tools are commercially available.

The second half — "Now Comes the Hard Part" — is where the report concentrates. Building an algorithm that performs well on a curated image set is a different task from running one in a busy clinic, on unstandardized smartphone photographs, across diverse patient populations and under real-world time pressure.

For dermatologists, the practical questions arrive quickly once a tool is in hand. Where in the workflow does an AI assessment belong — before the clinician examines the patient, or after, as a second read? Who is accountable when the algorithm and the dermatologist disagree on a suspicious pigmented lesion? How should clinicians document AI-assisted findings, and will payers reimburse for encounters that lean on automated analysis?

Validation outside the laboratory is the recurring obstacle. Algorithms trained predominantly on lighter skin tones have historically underperformed on darker skin, a disparity dermatology has had to confront directly as AI tools spread. Performance metrics reported on research datasets do not automatically transfer to the variegated lighting, angles and camera quality of patient-submitted photographs, which now feed much of teledermatology.

The specialty also faces a workforce and training dimension. Residents entering practice now will use tools that did not exist when many of their attending physicians trained, and curricula are still catching up. Understanding what an algorithm's output means — its sensitivity, its false-positive rate, its limitations — requires a level of statistical literacy that traditional dermatology training did not emphasize.

There is also the question of scope. Skin-cancer detection has attracted the most attention, driven by the clinical stakes of missing a melanoma, but the report's framing suggests the technology's reach extends across the specialty. Psoriasis severity scoring, eczema monitoring, acne grading and treatment-response tracking are all tasks that image-based AI could partially automate — each with its own validation and integration hurdles.

Skepticism and enthusiasm both have evidence behind them. Dermatology has been an early and heavy adopter of teledermatology and digital imaging, which gives the specialty unusually rich visual datasets. At the same time, clinicians who watched earlier decision-support tools overpromise remain cautious about algorithms marketed with performance figures drawn from idealized conditions.

The Medscape report lands at a moment when the regulatory, reimbursement and medico-legal frameworks for clinical AI are still under construction. Dermatologists who adopt these tools now are, in effect, drafting the de facto standards their colleagues will inherit — deciding through daily use where AI helps, where it misleads and where a trained human eye remains indispensable.

The technology's arrival, in other words, settles the easy debate. What comes next — validation in real clinics, equitable performance across skin types, defensible workflows and clear accountability — will determine whether dermatology AI becomes a durable clinical instrument or an impressive demo.