The AI Clinic · Issue 004AI in your workflow
The upgrade already landed.
OpenAI's new models just moved into your clinic's software
A new model family showed up inside tools you already use, MSK imaging AI keeps sorting out what's proven from what's still experimental, and two clinic-owner podcasts made the case that AI is now a bottom-line decision, not a tech one.
This issue went out later than usual this week. A full clinic week ran long and the news queue kept growing while I fell behind on it. Thanks for your patience, here's what actually mattered.
- OpenAI's new GPT-5.6 model family launched this week and is already running inside Microsoft 365 Copilot.
- A new review maps which AI applications in musculoskeletal imaging are clinically validated and which are still experimental.
- Two clinic-owner podcasts this week made the case that AI is now a financial decision, not a technology one.
The Big Story
OpenAI's new models just moved into your clinic's software
OpenAI shipped the GPT-5.6 family this week: three tiers called Luna, Terra, and Sol, with gains in reasoning, cybersecurity analysis, and general workplace tasks. Terra and Sol are already powering Microsoft 365 Copilot, so if your practice uses Copilot for anything, a scribe, a drafted email, a summarized note, you are running a new model under the hood as of this week whether you noticed or not.
Mark's read: The tiered pricing is the real story for a cash-pay practice. Terra gets you frontier-level reasoning without frontier-level cost, which finally makes AI worth a real line item instead of a curiosity. I run most of my own clinical reasoning work through Claude and haven't found a reason to switch. GPT-5.6 is a legitimate upgrade if you're already running OpenAI's tools, it just still isn't Fable 5 territory for how I actually use these tools day to day.
From Modern Pain Care
Want the tools this newsletter only talks about?
This newsletter is the free layer. Modern Pain Pro is where the AI workflows, and the clinical reasoning behind them, actually get built out, monthly drops, live coaching, and clinicians doing the work alongside you.
In the Clinic
AI is already reading your patients' MSK imaging, here's where it's proven
A new review maps every major AI application in musculoskeletal imaging since 2020, across rheumatoid arthritis, psoriatic arthritis, spondyloarthritis, and osteoarthritis. It covers automated diagnosis, severity classification, and prediction of disease progression, and sorts out what is validated for clinical use from what is still experimental.
In the clinic: Know which is which before you trust a report. Will AI clean up the inconsistency we already see in radiology reads? That remains to be seen.
The Business of Care
Four clinic owners put real numbers on AI ROI
Paul Gough's podcast features four physical therapy clinic owners sharing actual costs, growth, and trade-offs after adding AI tools. The central argument: a missed patient call costs more than the AI answering service that would have caught it.
For owners: Peer numbers beat any sales demo. If you're still debating a front-desk AI, this gives you real comparables instead of vendor promises.
Practice Ops
The admin burden strangling your growth has a fix, starting with a time audit
Doc Danny walks through a real cash-pay clinic case study on where admin time actually disappears to, then lays out a weekly time-audit method and the 80/20 principle applied to clinic operations.
For owners: Run the audit before you buy anything. The bottleneck in most practices is the process, not the missing software.
Research Watch
Anthropic built an AI researcher, not just an AI writer
Anthropic launched Claude Science, built to run autonomous, multi-step research tasks in computational biology and drug development with minimal direction, the same relationship Claude Code has to software engineering. Anthropic says it will also use the tool to develop its own drug candidates.
For researchers: This is a preview of what research tools look like in a year or two, agents that design experiments and synthesize findings, not just search literature and draft text. Stay ahead of your institution's policy on this, not behind it.
Under the Hood
Anthropic found where Claude "thinks" before it answers
A new interpretability method Anthropic calls the J-lens revealed an internal space inside Claude Opus 4.6 where the model holds and works through concepts before producing an answer. MIT Technology Review calls it a meaningful step toward practical AI interpretability.
Worth noting: This is groundwork for the day you can ask an AI tool to show its work and actually trust the answer, which matters for anyone weighing when to trust, question, or override an AI output in a clinical setting.
AI for Researchers & Faculty
A how-to guide for putting AI into your syllabus, written for any health profession
A peer-reviewed article lays out practical strategies for pharmacy faculty integrating generative AI into teaching: prompt design, AI-generated case studies, formative assessment, and responsible-use instruction, driven by accreditor expectations.
For educators: It's written for pharmacy programs, but every strategy here transfers directly to PT, OT, or medical education. Worth a read before you write your next AI policy for a course.
Classroom & Critical Thinking
Is AI quietly making your students worse thinkers?
Educator Andy Stapleton argues that routine AI use can erode the critical thinking, synthesis, and argumentation skills that make an expert clinician or scholar valuable, and frames the fix as how you assign the tool, not whether you ban it.
Worth watching: Relevant for anyone teaching clinical reasoning or evidence-based practice. One caveat, he's building a paid course around this argument, so weigh the pitch accordingly.
The last word
A new model family showed up inside tools you already run, MSK imaging AI keeps sorting out what's proven from what's still experimental, and two clinic-owner podcasts made the case that AI is a bottom-line decision now, not a tech one. Different stories, same thread: the tools keep getting embedded quietly into places you already work.
Thanks for your patience with the late send this week. If you have thoughts on the newsletter, good or bad or something missing, reply to any issue and let me know. I read every one.
Get the next issue.
One email a week. The AI that matters for clinical work, in plain language.