Code to Script: AI Agents Are Now Writing Medical Prescriptions

Medical prescriptions used to require a medical degree, a grueling residency, and board certification. In Utah, all you apparently need now is a smartphone camera and an AI agent with execution privileges.

Healthcare startup Nolla Health has officially launched a pilot program that allows its AI system to analyze facial scans, assess acne severity, and autonomously issue medical prescriptions. While human doctors remain in the loop initially—two physicians will sign off on the first 100 automated prescriptions—the ultimate goal is to systematically dial back human oversight and let the agent handle the entire diagnostic-to-prescription pipeline.

This isn't just another conversational chatbot advising you to drink more water and wash your face. This is an agent taking definitive, high-stakes actions in the real world. For years, the health tech sector has been bogged down in diagnostic parlor tricks, celebrating benchmarks where models pass the USMLE or summarize medical charts. But diagnosis without execution is just advice. Nolla Health is pushing past the advisory stage and granting agents transactional authority in one of the most heavily regulated sectors on the planet.

Of course, the operational risks are glaring. Skin conditions are notoriously tricky to classify from consumer smartphone photos, where poor lighting and low resolution can easily mask underlying complications. Handing a model the authority to issue pharmaceuticals raises critical liability questions. If an agent misidentifies a subtle skin issue and prescribes an ineffective or harmful topical treatment, who takes the fall—the supervising physician who rubber-stamped the training batch, or the startup that deployed the code?

Despite the regulatory minefield, this pilot marks an important milestone for the broader AI ecosystem. It proves that agentic workflows are finally breaking out of developer tools and enterprise workflow automation, venturing directly into real-world, service-level execution. If autonomous prescribing gains traction in dermatology, it sets a clear precedent for agents taking autonomous action in legal filing, financial advisory, and clinical operations.

The AI market doesn't need another chatbot that acts like a doctor. It needs systems that reliably execute tasks. Nolla Health's experiment might be starting with acne, but the implications for agent autonomy are massive.

Photo: Szabo Viktor / Unsplash (https://unsplash.com/@vmxhu)