For many clinicians, documentation has become one of the most frustrating parts of medicine. The visit ends, but the work continues: finishing notes, reviewing charts, responding to messages, placing orders, and trying to remember the details of conversations that happened hours earlier.

A recent JAMA study across five academic medical centers found that AI scribe adoption was associated with modest reductions in total EHR time and documentation time. Clinicians using AI scribes spent about 13 fewer minutes in the EHR and about 16 fewer minutes on documentation per 8 scheduled patient hours. Visit volume increased slightly, while EHR time outside scheduled hours did not significantly change.

That is helpful.

But it also shows why AI scribes, by themselves, are unlikely to solve specialty burnout.

In specialty care, the note is only one part of the problem. The larger issue is the structure of the clinic day.

The Hidden Friction of Specialty Clinic

In spine clinic, a large amount of time is spent before the actual clinical decision can be made.

Some of this time is obvious. Much of it is invisible.

Logging into a clinic computer may take 20 to 30 seconds. That does not sound like much, but if I see 30 patients in a clinic day, that can become 10 to 15 minutes spent just waiting for computers to become usable.

Opening imaging studies takes time. A single MRI, CT scan, or X-ray series may take 30 seconds or more to load. Many patients have multiple studies, sometimes from different facilities, often with outside images, reports, or prior comparisons that are not immediately available.

Then comes the search for context.

• Where is the referral note?
• What question is the referring clinician asking?
• Did the patient complete physical therapy?
• Were injections performed?
• What did the pain management physician document?
• Were prior operative reports uploaded?
• Is there evidence of medication trials, neurologic symptoms, or functional decline?
• Are we looking at the actual images, or only the radiology report?

Each step may only take a minute. But repeated across a full clinic, those minutes add up. More importantly, they add cognitive friction.

The physician is not only evaluating the patient. The physician is also searching, sorting, reconstructing, and verifying.

AI Scribes Help With the Visit, Not the Whole Workflow

An AI scribe can listen to the encounter and produce a draft note. That can be useful. It may allow the clinician to focus more on the patient and less on the keyboard.

But the scribe usually starts working once the visit begins.

In specialty care, much of the burden starts earlier.

For a spine surgeon, the key question is rarely just, "What did the patient say today?" The question is, "How does today's story fit with the imaging, prior treatments, physical exam, referral context, and the patient's goals?"

That requires synthesis.

A good note matters. But a good pre-visit summary may matter just as much.

What We Really Need: A Patient Packet

The next step is not simply a better AI scribe. It is an agent for clinical documentation synthesis.

In practical terms, this would be a system that prepares a concise, specialty-specific patient packet before the clinician enters the room.

For a spine clinic, that packet might include:

• Relevant imaging studies, already identified and easy to open
• Key radiology reports
• Referral notes and the reason for consultation
• Physical therapy documentation
• Pain management notes and injection history
• Prior spine surgery records
• Medication trials
• Functional limitations
• Neurologic symptoms or red flags
• A brief clinical timeline
• Missing information that needs to be clarified during the visit

This should not be a long AI-generated essay. Clinicians do not need more text to read.

They need the right information, organized in the right place, at the right time.

The goal is not for AI to decide who needs surgery. The goal is to reduce the time spent hunting through the chart so the clinician can spend more time examining, explaining, counseling, and making decisions with the patient.

The Clinic Day Is the Real Product

The risk with AI scribes is that health systems may treat them as the entire solution.

If a scribe saves a few minutes, those minutes may simply become more visits. But if the same missing imaging, scattered notes, slow logins, inbox work, and prior authorization burden remain, then clinicians may end up with a slightly more efficient version of the same overloaded day.

That is not a burnout solution.

The more meaningful opportunity is to rethink the clinic day itself.

• Can the necessary imaging be available before the visit?
• Can outside records be pulled into one place?
• Can prior treatments be summarized automatically?
• Can referral quality be improved?
• Can the clinician see a clear patient packet instead of searching through dozens of tabs?
• Can the post-visit plan be easier for the patient, staff, and referring clinician to follow?

That is where AI can become more useful in specialty care.

Not as a replacement for clinical judgment. Not as an independent decision-maker. But as infrastructure that helps organize the complexity of modern medicine.

The Real Opportunity for AI in Specialty Care

AI scribes are one of the first healthcare AI tools that many clinicians can understand immediately. They address a real pain point.

But specialty burnout is not just documentation burnout.

It is the accumulated burden of fragmented information, inefficient software, incomplete referrals, slow imaging access, and clinical decision-making under time pressure.

The next generation of tools should not only ask, "Can we write the note faster?"

They should ask, "Can we prepare the encounter better?"

For specialty care, that may be the more important question.

A well-designed AI system should help create the patient packet: the imaging, notes, prior treatments, referral context, and missing information organized in one place. That would make the clinic day more efficient, but also more clinically focused.

The future of AI in specialty care is not just ambient documentation.

It is clinical synthesis.

And that may be what clinicians need most.

This is general educational information and is not a substitute for evaluation by a qualified clinician.

Practical Takeaways

• AI scribes can reduce documentation burden, but they do not solve the full specialty clinic workflow.
• In spine clinic, small delays such as computer login time, imaging load time, and searching for prior notes can add up significantly over a full day.
• The next major opportunity is a clinical documentation synthesis agent that creates a concise patient packet before the visit.
• The best AI tools in specialty care will help clinicians find, organize, and act on relevant information. They should not replace physician judgment or decide who needs surgery.

I'm not interested in hype. I'm interested in what actually works, what we measure poorly, and what we could do better if we were honest about the gaps.

This site exists because I wanted a place to think in public about the parts of surgery that don't fit neatly into a journal article or a conference talk. The messy, practical, systems-level thinking that happens between cases — imaging interpretation as a workflow problem, the real cost of decision fatigue in elective spine surgery, what it means to adopt technology responsibly when the evidence is still catching up.

I write for surgeons, trainees, and builders. Patients are welcome, but this is the working-out-loud side of practice. If you're building tools for surgeons, I'd rather you understand how we actually think than how you imagine we do.

The categories here reflect the things I think about most: surgery and clinical decision-making, workflow and systems, and technology and innovation — but only the kind grounded in real problems. If it doesn't connect back to a patient, an OR, or a clinic, it probably won't end up here.

I'll write when I have something worth saying. No posting schedule, no content calendar. Just honest thinking from someone who operates, teaches, and builds.

In the rush to schedule, we often overlook the value of deliberation — and the decisions that happen before the OR matter more than most of what happens inside it.

Elective spine surgery is not emergency surgery. The word 'elective' means there is time — time to think, time to re-examine imaging, time to ask whether the surgical plan matches the clinical picture. And yet, in practice, that time is often compressed by scheduling pressure, patient expectations, and the momentum of a referral pipeline that rewards throughput over reflection.

I've started to think of the pre-operative period as its own kind of intervention. The decision to operate — and the specific plan — deserves as much rigor as the technical execution. A well-chosen operation done adequately will outperform a technically brilliant operation done for the wrong indication.

This isn't an argument against surgery. It's an argument for treating the decision itself as a skill worth practicing and protecting.

What would change if we measured the quality of surgical decisions as carefully as we measure complications? What if the most important outcome wasn't just whether the patient improved, but whether we chose the right operation for the right patient at the right time?

Who reads it, when they read it, what they compare it to, and how findings get communicated to the patient — each step introduces friction and potential error that has nothing to do with the scan itself.

Consider the typical flow: a patient gets an MRI. The radiologist reads it and generates a report. The referring physician sees the report and sends the patient to a surgeon. The surgeon pulls up the images, reads them independently, and forms a clinical impression that may or may not align with the radiology report. Meanwhile, the patient has already read the report in their portal and arrived with a mix of anxiety and misunderstanding.

Every handoff in that chain is a potential failure point — not because anyone is incompetent, but because the system isn't designed for the complexity of the information being transferred. We've built workflows around reports, not around understanding.

The fix isn't better AI reading scans. It's better systems for how imaging findings travel from acquisition to clinical decision. That's a workflow problem, and it deserves workflow solutions.

The gap between what's marketed and what matters is wide — and closing it requires surgeons who build, not just surgeons who buy.

Intraoperative navigation has improved meaningfully over the past decade. Registration is faster, accuracy is better, and integration with imaging has become more seamless. But the conversation around navigation technology is still dominated by feature lists and specifications rather than the questions that matter most: does this change how I operate, does it reduce errors, and does it fit into a real OR workflow without creating new problems?

Most navigation systems are designed by engineers solving engineering problems. The result is technology that is impressive on paper but awkward in practice — requiring extra setup time, creating new dependencies, and sometimes introducing a false sense of precision that can be more dangerous than helpful.

What surgeons actually need is simpler than what's being built: reliable confirmation of anatomy, fast registration that doesn't disrupt flow, and clear visualization that answers specific intraoperative questions. Everything else is noise.

The most useful navigation is the kind you forget is there — until the moment it matters.

The uncomfortable middle ground is where the real work happens — and where training falls short.

Shared decision-making is one of those concepts that everyone endorses and almost nobody does well. In residency, we learn to present options and obtain informed consent. But there's a vast difference between listing risks and benefits in a pre-operative visit and genuinely helping a patient navigate a decision that will affect the rest of their life.

The challenge isn't knowledge — it's communication under uncertainty. How do you explain that a surgery has a 70% chance of improving their pain without implying a 30% chance of making it worse? How do you honor a patient's preference to avoid surgery when you believe surgery is clearly indicated? How do you slow down enough to listen when there are twelve more patients in clinic?

Real shared decision-making requires time, humility, and a willingness to sit with ambiguity. It means sometimes saying 'I don't know' and sometimes saying 'I wouldn't do this surgery on my own family member.' It means treating the conversation as part of the care, not a hurdle before the care.

We don't train for this. We should.

Not interested in hype — interested in what holds up under scrutiny.

The conversation about AI in surgery tends to oscillate between uncritical enthusiasm and reflexive skepticism. Neither is productive. What's needed is a disciplined approach to identifying where AI might genuinely help, what evidence would be required to trust it, and what risks come with adoption — including risks we haven't thought of yet.

The most promising applications aren't the flashy ones. They're the boring ones: automating documentation, flagging imaging findings that might be missed in a busy workflow, predicting which patients are at higher risk for complications based on patterns too subtle for human pattern recognition. These are problems worth solving because they're real, measurable, and currently handled poorly.

The dangerous applications are the ones that promise to replace clinical judgment without understanding what clinical judgment actually is. Judgment isn't pattern matching — it's the integration of pattern recognition with context, experience, uncertainty, and values. Any AI system that claims to replicate that without acknowledging its limitations is selling something, not solving something.

I'm interested in AI that makes surgeons better at what they already do, not AI that pretends surgeons aren't necessary. The former is useful. The latter is fantasy.