The billing code doesn't exist yet.

CMS has been pricing AI tools the same way it prices a CT scanner: bolt a code onto a piece of hardware and call it done. That has never made sense for software that updates every month and has no physical footprint. Its proposed 2027 outpatient and physician fee rules are the first real attempt to build payment mechanisms made specifically for AI-based diagnostic and clinical support tools, instead of forcing them into a category built for machines you can touch.

If your clinic uses or is weighing an AI diagnostic or decision-support tool, this is the rule to watch. New codes or payment categories could change what you can actually bill for the AI you're already running, or make a tool you've been holding off on suddenly worth adding. The proposal isn't final. Comment periods and vendor lobbying will still shape what lands.

I started a consulting practice this year for fixing the workflow friction that quietly eats a business's time: the same information typed into three systems, the handoff that drops a detail, the AI tool that gives a generic answer because it never had your actual context. Clinics carry this load as much as any other small business.

Take the Workflow Friction Checklist: ten plain questions, no email required, and it scores where the friction in your practice actually lives instead of you guessing at it. Takes about five minutes. See where you land, and if it points at something real, that's exactly the conversation a workflow audit is built to solve.

A new meta-analysis pooled the studies on deep-learning models that grade knee osteoarthritis severity on the Kellgren-Lawrence scale from a plain X-ray. The finding worth remembering: accuracy isn't uniformly high. It swings by grade, doing better at some severities than others instead of nailing every reading the same way.

Hospitals short on revenue-cycle staff are turning to AI vendors to keep the billing backlog from growing. Akasa, which uses generative AI and large language models to automate claims and billing work, has raised $205 million and now works with major health systems that can't fill the roles fast enough on their own.

Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-source model the company says performs competitively with the top proprietary systems from Anthropic and OpenAI. The release was timed ahead of the World AI Conference in Shanghai.

Researchers tested whether a self-adhesive, skin-conforming strain sensor called Motion Tape, paired with movement data, can objectively classify low back pain during functional tasks. That's a place where standard sensing technology has struggled, subtle movement impairments don't show up cleanly on the tools most clinics have access to.

Open a blank page. List every AI tool in your practice that touches a diagnosis, a triage decision, or a treatment recommendation, not scheduling, not scribing, just the clinical-decision layer. Next to each one, note whether it's billed today, and how. That list is exactly what CMS's 2027 rule will touch, and you want to know your own exposure before a payer rep tells you about it.