You have walked past a Proposition 65 sign a thousand times. This facility contains chemicals known to the State of California to cause cancer and reproductive harm. It is on parking garages. It is on coffee shops. It is on a hotel lobby you stayed in last month and on the back of a receipt in your pocket right now.

You did not read it. Neither did I.

That is not stupidity and it is not indifference. That is what happens to a warning attached to nine hundred substances and posted on approximately everything. The sign stopped carrying information the day it stopped being possible for it to be absent. Whether the law changed anyone’s behavior is actively disputed and has been for thirty years, but the part nobody disputes is what happened to your attention. You learned to not see it.

Now put that same sign on a heart monitor, next to a person who is actually sick.

The number

In March of 2013, a research team at the University of California, San Francisco did something nobody had bothered to do. They counted.

Not sampled. Not modeled. Counted. Every alarm, audible and silent, from every physiologic monitor in five adult intensive care units, seventy seven beds, for thirty one consecutive days. Four hundred and sixty one patients.

Two million, five hundred and fifty eight thousand, seven hundred and sixty alarms.

Of those, 381,560 made a sound. Divide it out and you get the number that should be printed on the wall of every hospital that has ever convened an alarm safety committee:

One hundred and eighty seven audible alarms, per bed, per day.

That is one noise demanding a human being’s attention roughly every seven and a half minutes, around the clock, in the room of every single patient, for a month.

Then they did the harder part. They took 12,671 arrhythmia alarms and had trained annotators judge them, with ninety five percent agreement between raters.

Eighty eight point eight percent were false.

And here is the finding I almost missed the first three times I read the paper. Take ventricular tachycardia on its own, the alarm nobody would ever tell you to ignore. The annotators judged 502 of them to be true.

Three hundred and thirty four of those 502 came from a single patient in ventricular storm. One human being generated two thirds of every true ventricular tachycardia alarm in a 461 patient study. Set that patient aside and 168 true alarms remain across everybody else, real events, correctly detected, exactly as designed.

Ninety three percent of them did not last long enough to require treatment.

Read that again, because it breaks the usual argument. The standard defense of alarm burden is that a false alarm is the price of never missing a real one. Fine. But when they went and looked, the real ones mostly did not need anyone either. The system was not trading noise for safety. It was generating noise, and then generating a smaller pile of noise that happened to be technically accurate.

Nine times out of ten the alarm was wrong. On the tenth, it was right and it did not matter.

It has a name, and medicine named it a long time ago

I don’t want to present this as a discovery. It is not. This is one of the best documented problems in patient safety, and the people who documented it were not gentle about it.

Alarm fatigue is the accepted term. The Joint Commission issued Sentinel Event Alert 50 on 8 April 2013, and the numbers in it are not soft. Between January 2009 and June 2012 the Commission’s sentinel event database recorded 98 alarm related events. Eighty of them ended in a death. Thirteen more caused permanent loss of function. That alert also states, flatly, that the number of alarm signals per patient per day “can reach several hundred depending on the unit within the hospital.”

So the UCSF count did not surprise anybody. It confirmed, with a month of exhaustive data, a thing the field had already written down.

And medicine did respond. Alarm management became a National Patient Safety Goal. Hospitals stood up alarm committees. Manufacturers retuned defaults, added delays, tightened artifact rejection. Good people did real work and some of it helped.

The burden is still there. That is not because anybody was lazy. It is because everything we did was tuning, and the problem is not tuning.

What aviation did, and when

“Why can’t healthcare be more like aviation” is one of the laziest sentences in our industry, and I have heard it in a hundred conference rooms from people who had never read a single aviation document.

So let us read one.

In May of 1977, Boeing Commercial Airplane Company delivered a study to the Federal Aviation Administration called Aircraft Alerting Systems Criteria Study, report FAA-RD-76-222. They had gone and counted too. They tallied the alerts on nine aircraft types and found the number climbing with every new model. More systems, more sensors, more things worth telling a pilot about, and a flight deck that was starting to shout.

That was 1977. Medicine would take another thirty six years to run the equivalent count in an ICU.

Out of that work came a discipline with three parts.

First, a hierarchy with teeth. Warning, caution, advisory. Not a labeling convention but a definition of what the human is expected to do. Warning means immediate awareness and immediate response. Caution means immediate awareness and subsequent response. Advisory means awareness, and a response may follow. Those definitions are federal regulation now, 14 CFR 25.1322, Amendment 25-131, published 2 November 2010 and effective 3 January 2011.

Second, and this is the part medicine skipped: the deliberate withholding of true information. The FAA’s advisory circular on flight crew alerting, AC 25.1322-1, dated 13 December 2010, directs manufacturers to consider “inhibiting alerts for specific phases of flight (for example, takeoff and landing) and for specific airplane configurations.” An aircraft in a takeoff roll will not tell the crew about a condition that is real, correctly detected, and none of their business for the next forty seconds. The information is accurate. It is suppressed anyway, on purpose, because a pilot rotating at 150 knots has exactly one job.

Third, the philosophy that falls out of the first two. Airbus calls it the dark cockpit, and the rule is beautifully simple: if a system is doing what it should be doing, the light on its button stays out. A light means a fault, or a system switched off that ought to be on. Illumination is exception. Darkness is the affirmative statement that everything has been checked and everything is fine.

Look at that photograph again. That flight deck is not quiet because nobody is watching. It is quiet because everything is being watched, continuously, by something that has been given permission to say nothing.

Silence is the message. Lights out means ready to fly.

What medicine took, and what it left

In August of 2003 the International Electrotechnical Commission published IEC 60601-1-8, the collateral standard for alarm systems in medical electrical equipment. It defines three alarm priorities.

High. Medium. Low.

We took the tiers. We took the color conventions and the tone patterns and the compliance testing. We did not take the rule underneath them, and the rule underneath them is the whole invention.

So here is what one bed looks like.

I went through a photograph of a single intensive care bay and counted what was in it. Not a busy one. A bed rigged and ready with nobody in it, linens flat, ECG leads coiled on the rail, ventilator circuit hanging on its arm.

On the two poles beside it, four BD Alaris point of care units driving fourteen pump channels. Behind them a ventilator. In front, a console running arterial waveforms. Above the bed a physiologic monitor, and a second one to the right.

Eighteen devices at one bedside, every one of which can raise an alarm on its own authority. The bed was empty and the monitors were already lit.

I should say plainly that I have a personal stake in this. I spent the last stretch of my career as a vice president of connectivity at Alaris, then CareFusion, then Becton Dickinson. Those are my pumps. I worked on getting them talking to the electronic record, and that work was worth doing, and I would do it again.

And not one of those fourteen channels knows the ventilator four feet away exists. Or the monitor above the bed. Or the thirteen other channels bolted to the same pole.

There is no phase of care in a hospital during which a device will decline to tell you something true. There is no equivalent of the takeoff roll. There is no cross device prioritization, because the pump, the ventilator, the monitor and the bed are four products from four vendors and none of them knows the others are in the room, let alone which of them has earned the next four seconds of a nurse’s attention.

And there is certainly no dark cockpit, because in an intensive care unit silence does not mean everything has been checked. Silence just means nothing has crossed a threshold yet.

Everything speaks, all the time, and nothing is in charge of deciding who speaks next.

Why louder or smarter alarms do not fix this

Here is where I changed my own mind, and it took me an embarrassingly long time.

For most of my career I thought alarm fatigue was a signal processing problem. Better algorithms, tighter filters, smarter thresholds, fewer artifacts. That work is real, it has helped, and the people doing it are good at their jobs.

But look again at the UCSF finding. Ninety three percent of the true ventricular tachycardia alarms did not need treating. You cannot filter your way out of that, because those alarms were not errors. The detector did precisely what it was built to do. The machine did not get the answer wrong. Nobody ever asked it a question.

A threshold is not a question. A threshold is a comparison, and a comparison is what we have been shipping since I started in this business in 1985. Is this number above that number. Yes. Make a noise.

The question a clinician actually has is different in kind. Given this patient, on these drugs, with this history, at this hour, with those other thirteen channels running what they are running, does this matter to anyone right now?

That is not a threshold. That is reasoning over context. And for the forty years I spent acquiring medical data and normalizing it and moving it from one machine to another, there was nothing on earth that could do it. So we did the only thing available. We compared numbers to numbers and we made a noise, and we made a hundred and eighty seven noises per bed per day, and we called it vigilance.

It was not vigilance. It was the absence of anything better, wearing vigilance as a costume.

That part has changed. Not “is about to change,” not “shows promise in early studies.” Changed, in the last three years, and most of our industry is still designing alarms as though it had not.

The prediction is on a screen. The response to the prediction is on a piece of paper. Nothing in that room connects the two, and no amount of retuning an alarm threshold ever will.

One more thing, and I missed it the first three times I looked at that photograph. At the bottom of the toolbar on the left of the Edwards screen there is a bell with a line through it. The alarms on that monitor are silenced.

I am not going to tell you why, because I do not know and neither does anybody else looking at that picture. What I will tell you is that it is the only thing in the frame doing what aviation has done since 1977. And it was done by a person reaching up and pressing a button, on one device, for one shift, with no rule behind it, no record of it, and nothing whatsoever to stop the other fifteen machines in that bay carrying on without her.

Now look at the bottom of the frame, at the laminated sheet cable-tied to the pole. It is a flowchart, printed in red and black, and one of the boxes on it says MAP 65.

There has been a predictive algorithm at that bedside for seven years.

The monitor on the left is an Edwards HemoSphere. In a small box in the top right of that screen is the Acumen Hypotension Prediction Index, reading 42. That is a machine learning model, built from more than two hundred thousand patient events, telling the room how likely it is that this person’s blood pressure is about to fall. It is not a prototype and it is not a research tool. The FDA cleared it on the third of December, 2018 [9].

Before anybody writes to tell me the industry has not tried, look at this photograph. I did not take it. I went looking for a picture of an intensive care bay that was not a marketing shot, gave up on the stock libraries twice because every one of them is too clean, and eventually licensed this.

It is already in the room

What I would actually build

Not a better threshold. A layer above the devices that holds the whole picture, that knows what a takeoff roll looks like for this particular patient, and that is permitted to stay quiet.

The hard part is not the reasoning. The reasoning is close to free now, and I mean that literally: a reasoning pass over one patient’s full context costs somewhere between a cent and a dime. The hard part is the one it has been since 1985, which is being allowed to reach the data in the first place, and being permitted to say nothing.

We have never given a medical device permission to stay silent about something true. Aviation did it in 1977 and had it in regulation by 2010.

That is the gap. It is not a technology gap.

And a small piece of news. After forty years of this I finally wrote the book: The Technology Was Never the Problem, twenty two chapters on what we built, why so little of it moved outcomes, and what changed. It is with reviewers now and it will be on Amazon in mid September. I will say more closer to the date, and I will not turn this newsletter into an advertisement for it. Inside the Loop keeps doing what it has always done, and the reporting here stays new.

Sources

[1] Drew BJ, Harris P, Zegre-Hemsey JK, Mammone T, Schindler D, Salas-Boni R, et al. “Insights into the Problem of Alarm Fatigue with Physiologic Monitor Devices: A Comprehensive Observational Study of Consecutive Intensive Care Unit Patients.” PLOS ONE 2014;9(10):e110274. Five adult ICUs, 77 beds, 461 patients, 31 days, March 2013. 2,558,760 total alarms; 381,560 audible; 187 audible alarms per bed per day; 12,671 arrhythmia alarms annotated at 95 percent inter-rater reliability, of which 88.8 percent were false. Ventricular tachycardia: 502 true positives, of which 334 occurred in one patient with ventricular storm; of the remaining 168, 93 percent were not sustained long enough to warrant treatment.

[2] The Joint Commission. Sentinel Event Alert, Issue 50: Medical device alarm safety in hospitals. 8 April 2013. 98 alarm related events reported between January 2009 and June 2012: 80 deaths, 13 permanent loss of function, 5 unexpected additional care or extended stay. Also the source for “several hundred” alarm signals per patient per day.

[3] Veitengruber JE, Boucek GP, Smith WD. Aircraft Alerting Systems Criteria Study, Volume I. Report FAA-RD-76-222, Boeing Commercial Airplane Company for the Federal Aviation Administration, May 1977.

[4] 14 CFR 25.1322, Flightcrew alerting. Amendment 25-131, published 2 November 2010, 75 Federal Register 67201, effective 3 January 2011. Warning: immediate awareness and immediate response. Caution: immediate awareness and subsequent response. Advisory: awareness and possible subsequent response.

[5] Federal Aviation Administration. Advisory Circular AC 25.1322-1, Flightcrew Alerting, 13 December 2010. Sections 8(d) and 13(c)(1)(h) address inhibiting alerts for specific phases of flight, giving takeoff and landing as the examples.

[6] IEC 60601-1-8, Medical electrical equipment, Part 1-8: General requirements for basic safety and essential performance, Collateral standard: General requirements, tests and guidance for alarm systems. First edition, August 2003. Establishes high, medium and low alarm priority.

[7] California Safe Drinking Water and Toxic Enforcement Act of 1986, Proposition 65, administered by the Office of Environmental Health Hazard Assessment. The list has grown from its first publication in 1987 to roughly nine hundred chemicals. Whether the warnings change behavior is actively disputed and the law has been criticized on precisely the grounds described here.

[8] Device counts are the author’s own, taken from a photograph of a single unoccupied intensive care bay. No patient was present in the bed. That bay is not the one reproduced in this article. It is a different room in a different hospital, and it is not published here, so counting the pumps in the photograph above will not produce the same numbers. The flight deck photograph is licensed stock.

[9] Edwards Lifesciences. HemoSphere advanced monitoring platform with the Acumen suite of decision-support solutions, FDA clearance announced 3 December 2018. Edwards describes the Acumen Hypotension Prediction Index as introducing “artificial intelligence (AI) to hemodynamic monitoring through a machine learning, data-driven algorithm that indicates the likelihood of a hypotensive, or low blood pressure, event before it occurs”, with algorithms derived from more than 200,000 patient events. The reading of 42 and the paper protocol are the author’s observations from the licensed photograph. No patient is visible in the frame.

Daniel Pettus spent nearly fifty years in medical device and health IT leadership at Alaris, CareFusion and Becton Dickinson. He contributed to IHE Patient Care Device interoperability standards and holds two patents with three provisional filed. He is the inventor of AI MedAgent and the founder of Inside the Loop.

Inside the Loop // Issue 09

In the spring of 1995 I stood in the parking lot of my own company and bought a laser printer.

Diatek was selling itself off a folding table. I paid cash, carried the printer to my car, and drove home with my gut hurting. Not about the printer. About what I knew we had built, and what I knew it could do.

Two years later the Journal of Clinical Monitoring published a study of the system I had spent the back half of that decade on. Duke University Medical Center bought our ARKIVE anesthesia information management system for six cardiothoracic rooms in 1987, installed it in August 1988, and on May 25, 1992 rolled it out hospital wide to every anesthetizing location in the main and auxiliary suites. Duke had built its own system years earlier and abandoned it in 1983. Ours stuck. [1]

On the first page of that paper there is a footnote. It reads: “Arkive Information Systems, Inc. has changed corporate identity.” [1]

The product worked. The company was in the parking lot.

This is an article about why that keeps happening, and it is going to name names. Not to blame anyone. The opposite. I want to give credit to the people who taught me this business and who changed the practice of medicine for the better, and who mostly did not get rich doing it. Early entrepreneurs pay the price. That is not a complaint. It is the job description.

1985: Tom Gregory Puts Me in Charge of Software

It was 1985 and it was my first job in healthcare technology. My mentor was Thomas K. Gregory, PhD, and he was the perfect mentor at exactly the right moment.

In June of that year an investor group consolidated three San Diego medical technology companies, Diatek Inc., Neurometrics, and Diatek Medical Technology, formerly Dade Medical, into a single entity called Diatek Corp. Phil Faris was chief executive. We operated out of a 69,000 square foot plant in Sorrento Valley, and we grew into the second largest manufacturer of electronic thermometers in the country, behind our crosstown rival IVAC, with roughly forty percent of the market and about 180 employees. [2]

Neurometrics was the piece that mattered to me. The Lifescan EEG monitor was just getting started as a project and Tom handed me the software. I later hired Dave Geving, and together we extended Lifescan with an evoked potential option, which turned it from a passive monitor into something closer to a diagnostic instrument. In 1986 Tom and I published the acquisition algorithm behind it. [3]

Understand what Lifescan did. It gave the anesthesiologist a continuous, readable picture of depth of anesthesia, in an era when that was substantially a matter of judgment and experience. It was also an early indicator of potential brain injury. A signal that had not existed before, at the head of the table, while the patient was still on it.

Lifescan did not make it. Not because it did not work.

Between 1985 and 1994 there was no clear billing code for brain monitoring during anesthesia. No code, no payment. No payment, no adoption. It did not matter that it improved outcomes. It did not matter that it could catch injury early.

I was thirty something and I did not understand yet that I had just met the single most important fact in this industry. I would meet it again three more times.

ARKIVE, and What $10.5 Million Bought

I moved onto the anesthesia information system in 1991 until 1994. I have written about ARKIVE before and I will not repeat all of it here, because Issue 08 covers the machine in detail and I gave it its due there. [4] The short version for anyone landing here first.

ARKIVE was a full anesthesia information management system, not a data logger. Touch and voice sensitive workstations on articulated arms sat at the anesthesiologist’s side, wired into the anesthesia machines and the physiologic monitors, capturing inspired and end tidal gases, drugs, infusions and vital signs automatically, and assembling the anesthesia record as the case unfolded. Getting devices from a dozen manufacturers to speak to one system in 1988 was genuinely hard. We did it. I put that product into Ohio State, Duke, University Nijmegen in the Netherlands, and UCSD.

Development of ARKIVE cost an estimated $10.5 million. [2] In late 1980s dollars, for a company of 180 people, that is not a side project. That is the company betting itself on an idea.

The autopsy

Diatek’s lead backing came from Douglas J. Dayton of the Minneapolis Dayton Hudson retail family. [2] If the name is unfamiliar, the store is not. Dayton was the first president of Target, ran it through its early expansion, left Dayton Hudson in 1972, and formed a venture capital firm in 1974 that he ran until he retired in 1994. [5]

He sat on our board, and this is the part that matters. He was not a line item in a fund. He was personally responsible for the continued infusion of capital. He was the money.

In 1995 he ended the relationship for family reasons and gave us a short window to find a replacement lead. Phil Faris ran that search. It did not pan out. Dayton’s public record shows him retiring from venture capital in 1994. Faris’s record shows him as president, chief executive and director of Diatek Corporation and Subsidiaries from 1982 to 1995. Three separate timelines, none drawn from each other, all landing inside the same eighteen months.

A company with peer reviewed clinical adoption at a top academic medical center ended because one board member stopped writing checks and nobody else picked it up.

As for Duke, they never went back to paper. In October 2006, fourteen years after our hospital wide rollout, Duke University Health System implemented Dräger’s Innovian Anesthesia across more than 150 point of care locations, described at the time as the largest upgrade in Dräger Medical’s history and the most extensive Innovian installation in the world. [6]

The department that took a chance on six operating rooms in 1987 became the global flagship for a German multinational. Just not for us.

The Second Company, and the Man Who Brought Me Into It

Joseph Condurso was executive vice president and managing director at Diatek Medical and Arkive Information Systems from 1993 to 1995. [7] Where Tom Gregory was my technical mentor, Joe was my business mentor, and my recollection of those years is that he saw where this market was going well before the rest of us did.

He brought me into the formation of iMetrikus. The seed financing came from Dr. William J. Rutter, and I want to be clear about who that is, because it tells you this was not a couple of guys with a slide deck.

Bill Rutter took over an undistinguished biochemistry department at UC San Francisco in 1968 and turned it into one of the best in the world. In 1977 he and his colleagues isolated the rat insulin gene and put it into bacteria, one of the first real triumphs of genetic engineering. In 1981 he cofounded Chiron, which produced the first recombinant hepatitis B vaccine, completed the first sequencing of the HIV genome in 1984, and discovered, sequenced and cloned hepatitis C in 1987. He died on July 11, 2025. [8]

Rutter’s original vision for iMetrikus was a system to help manage HIV and AIDS for patients being cared for primarily in primary care. Look at that against his own science and it is not a coincidence, it is a straight line. Joe and the rest of the team expanded it to chronic care generally.

MetrikLink was my idea and my invention, a home connectivity hub that pulled readings off more than thirty consumer medical devices and sent them to our data center. Jae Evans led platform development. Peter Thompson extended the product schema to incorporate the UMLS public data mart, and if you want the full breakdown of what public data marts are and why that mattered in 2003, it is in Issue 08. [4] I hold two patents from that work. [9]

Why iMetrikus did not make it

Payers, both commercial insurance and CMS, looked at what we were doing and saw an experiment.

This was while our early clinical users were giving talks and publishing on how the technology was improving lives and saving time and money. A UC San Diego endocrinologist presented our remote monitoring results in an underserved diabetes population in 2003. [10] None of it moved the payment question, because the payment question was not being asked by the same people who were impressed by the results.

Startups have to move fast. Healthcare payers move at a snail’s pace, and that is being generous. A company with a two year runway cannot outlast a five year coverage debate. That is not a business failure. That is arithmetic.

So I Went Where the Runway Was Longer

After iMetrikus I made a deliberate choice, and I would make it again. I joined the larger forces in medical technology, the ones with balance sheets that can absorb the pace and the risk aversion. Alaris, then Cardinal Health, then CareFusion, then BD.

We did genuinely innovative work there. I spent those years on the fringe, building and promoting the case that clinical care improves when the devices talk to each other, and doing it without being strangled by the culture of the status quo. It still had problems with pace and with protectionism.

And it taught me the barrier I did not fully appreciate from inside a startup.

Healthcare technology that generates a wealth of data mostly does not share it, or does not share it without a substantial toll.

This is where the industry conversation usually turns into shouting. Giving the data away is not the answer either. Nobody develops anything if there is no return on developing it. That is how the business works and I am not going to pretend otherwise after spending my career inside it.

But holding it back and charging a toll for access slows the whole field to a crawl. We built the connectivity. The value was in the data, and the data sat behind a gate with a price on it. Every founder reading this already knows what I am describing, because they got the quote.

Qualcomm Life Had Every Part, and It Still Did Not Work

Qualcomm Life set out to do essentially what iMetrikus had attempted, at a scale we could never have reached, with research dollars we could not have imagined. Several people from the BD world moved over to help jump start them into healthcare.

They had all the pieces. Cellular connectivity built for home health by a company that understood cellular better than anyone. The 2net Hub, which was a modern version of what MetrikLink had been. And Capsule Tech, a hospital gateway product, brought in to bridge from the hospital to the home. On paper it is the correct architecture, and it is the architecture I had been arguing for since 2003.

I will note, because it is a matter of public record rather than an accusation, that I am a named inventor on United States patent 7,375,647 covering a method and apparatus for communicating data between a medical device and a central data repository. [9] I never tested it against anyone. I make no claim here, and I never filed one.

It did not matter. The closed, siloed structure of American healthcare never produced a revenue stream of any consequence for it. Then the Qualcomm board changed direction. Qualcomm Life was divested to a private equity firm in February 2019 and renamed. The remote monitoring business went to Philips in 2020. [11]

And If You Still Think It Is a Money Problem

Look at PatientSafe Solutions, here in San Diego.

PatientSafe was founded in 2002 as IntelliDOT and built PatientTouch, a mobile platform putting barcode medication administration, secure messaging, alerts, record data and documentation into a clinician’s hand. It is a good idea aimed at a lethal problem. Over nineteen years and eleven rounds it raised somewhere between $135 million and $143 million from about a dozen investors including TPG Biotech, the Merck Global Health Innovation Fund, HighBar Partners and Telus. Institutional money, strategic pharma money, telecom money.

In April 2021 Vocera acquired it for $36 million. [12]

Roughly $140 million in, $36 million out, nineteen years. That is not a funding problem.

Now hold that against Google Health, which launched in 2008 and was gone by 2011 because it never achieved what Google called broad impact. [13] Against Microsoft HealthVault, which lasted twelve years and shut down in November 2019 on low usage and no durable business model. [14] Against Intel, which spent nineteen years on this, launched a fifty fifty joint venture with GE in January 2011, took full control in 2016, and sold the remains to a contract research organization at the end of 2019. [15]

Diatek ended with one angel and no institutional capital. PatientSafe ended with everything Diatek never had. Google, Microsoft, Intel and Qualcomm had more money than all of them combined. If the variable were capital, those endings would not rhyme.

The Six Gates

Here is what actually stops things, in roughly the order founders underestimate them.

Gate 1. The buyer is not the user, and neither one is the payer.

Three parties, three incentive structures, one signature. The nurse who benefits does not buy. The system that buys does not get paid more for having it. The insurer that pays was never in the room.

Gate 2. No reimbursement code means no revenue.

Clinical benefit and revenue are separate questions and founders keep collapsing them into one. Lifescan is my proof. It read the brain during anesthesia and it could flag injury early, and there was no code for that, so it did not matter.

Gate 3. The sales cycle outlasts the funding round.

Enterprise hospital implementations routinely run eighteen months and longer, and slow provider sales cycles are the most cited barrier to growth among digital health companies. [16] A seed round is not eighteen months long.

Gate 4. The data is a toll booth.

Integration is not a technical problem, it is a permission problem, and permission usually has a price. Frequently it is set by someone who would rather sell the hospital its own version of you.

Gate 5. Liability lands on the clinician.

Adoption asks a physician or a nurse to accept personal professional risk for software written by strangers. This is the gate that kills pilots after they succeed.

Gate 6. The capital itself is fragile.

Healthcare payback periods break fund economics, so the patient money has historically come from individuals rather than institutions. That makes a company hostage to one person’s private circumstances. Ask me how I know.

What a Silo Actually Costs You

Gate one deserves more than a paragraph, because the abstraction hides how bad it is.

Selling a single product line into an acute care hospital, one product, meant roughly a year of interaction before anyone signed anything. Not a year of negotiation. A year of interaction. The rooms I had to satisfy included executive management, medical strategy under the chief medical officer, pharmacy under the director of pharmacy, technology under the chief information and chief technology officers, nursing, biomedical engineering and maintenance, finance under the chief financial officer, and legal and regulatory. There were more.

Every one of those departments ran its own way, with its own reason to say yes and its own reason to say no. Most were nominally aligned to an institutional strategy and the walls between them were still nearly impossible to manage. It was inefficient, and sometimes it was open infighting over whose priority came first.

This is not only my experience. It is not uncommon for a hospital value analysis committee to seat twelve to twenty four people. [17]

Consumer technology has sprinted past healthcare over the same forty years for a reason that has nothing to do with engineering talent. In consumer, one person decides. In acute care, twenty four people decide, and any one of them can end it.

How Far Behind Is Healthcare, Really

People in this industry like to say healthcare runs about twenty years behind everybody else. I have said it myself. I want to be careful with the number, because it is an aphorism rather than a measurement, and nobody has ever shown me the arithmetic underneath it.

So instead of asserting it, let me hand you one receipt and you can do the subtraction yourself.

Take the question “is this approved?” In consumer finance the card networks began answering it electronically in 1973, when the authorization system that became VisaNet went live. [24] Half a century later you tap a card in a coffee shop and have your answer before you have put your phone back in your pocket. Nobody thinks about it. Nobody markets it. It is plumbing.

Now ask the same question inside American healthcare. Is this treatment approved? That question still produces phone calls, faxes, forms, peer to peer reviews and appeals. Medicare Advantage insurers alone issued nearly 53 million prior authorization determinations in 2024, and hospitals spent roughly $43 billion in 2025 chasing money that was already owed to them. [19][20]

The federal requirement for payers to run a standards based prior authorization interface takes effect January 1, 2027. [22]

1973 and 2027. Fifty four years to get an electronic answer to the same four word question.

I am not going to tell you healthcare is exactly twenty years behind, because I cannot measure it and neither can anyone else who says it. What I will tell you is that I have been watching from inside the building since 1985, and twenty years is the number that feels right to me. Call that my estimate rather than a finding. The fifty four years is the finding.

The One Serious Attempt to Fix It

Darren Dworkin, then chief information officer at Cedars-Sinai, built the most credible bridge anyone has built. He founded the Cedars-Sinai Accelerator with Techstars, and the first class kicked off in March 2016: eleven companies, one hundred twenty thousand dollars apiece, three months inside a real health system with real clinicians and real data. [18] It has run at least ten classes since.

I retired from BD in April 2019. About a month later Darren invited me to a Demo Day and raised the idea of my serving as an industry mentor for future classes. I went. I thought the concept was brilliant and I still do.

The accelerator fixed access, which was real and worth fixing. It could not fix reimbursement, procurement, liability, or the fragility of the capital. No accelerator can. Those are properties of the market.

So Why Am I Optimistic

There is a version of this ending I refuse to write. The popular version says artificial intelligence unlocks what was previously impossible and therefore the time is finally right.

That does not survive the evidence above. Not one company in this article failed for lack of capability. Lifescan worked. ARKIVE worked. MetrikLink worked. PatientTouch worked. Google, Microsoft, Intel and Qualcomm all shipped functioning products. AI does not repeal a single one of the six gates. Reimbursement does not care how good your model is.

Here is the narrower claim, and it is the one I will defend.

Every previous wave sold the hospital a new capability. AI is the first wave aimed at the hospital’s largest expense.

Workforce is the biggest line on the hospital income statement. American hospitals spent more than one trillion dollars in 2025 paying health care workers, with workforce costs up 5.6 percent against total expense growth of 7.5 percent. [19]

Now look at what a meaningful slice of that labor is doing. Hospitals spent roughly $43 billion in 2025 trying to collect money insurers already owed them, including nearly $18 billion just overturning denials. The average hospital employed about sixty four administrative and billing staff for that work, roughly 6.5 percent of total employment. Medicare Advantage plans denied about 17 percent of initial claims and 57 percent of those denials were eventually overturned, which means most of that labor produced delay rather than a decision. Those same insurers issued nearly 53 million prior authorization determinations in 2024. [19][20]

Fifty three million determinations is not a technology gap waiting for a breakthrough. It is a labor problem waiting for automation, sitting inside the workflow rather than bolted onto the outside of it. No previous wave went near it.

What Actually Changed

Go back to the six gates and ask a different question. Not whether they are gone. Ask which ones you have to pass through before anyone can find out whether the thing works.

Every product in this article had to get inside the building. Lifescan was a monitor at the head of the table. ARKIVE was a workstation on an arm in an operating room. MetrikLink was a box in a living room. PatientTouch was a device in a nurse’s hand. Hardware, on the network, inside the perimeter. Security review, biomedical engineering, an asset tag, a maintenance plan, an integration project, the value analysis committee, and a department that had to agree to staff it. Every gate cleared before the first result.

The reasoning layer is not in the building. For the most part it sits outside the gates.

It does not get installed on the hospital network. It does not sit on a cart. It needs no asset tag and no preventive maintenance schedule, and nobody has to be hired to push it down a hallway. That does not make the gates disappear. It moves most of them from before the evaluation to after it, and that reordering is the whole difference.

What Still Has to Be Solved

The remaining barrier is inbound, and it is gate four. Getting the data out of the hospital still costs money. That toll booth has not been dismantled and I am not going to pretend it has.

Regulation may finally do some of the work. Certified electronic health records have been required to expose standardized FHIR application programming interfaces since January 1, 2023. [21] Under the CMS Interoperability and Prior Authorization Final Rule, published in February 2024, affected payers including Medicare Advantage organizations, Medicaid and CHIP programs and managed care plans, and qualified health plans on the federal exchanges must run four production FHIR interfaces by January 1, 2027, one of them a prior authorization interface, with operational provisions phasing in from January 2026. [22]

Read that against the numbers above. The federal government has now placed a dated, standards based interface requirement directly on top of the largest pile of wasted administrative labor in American healthcare.

Now the honest caveat, because this is the part optimists skip. Information blocking has been prohibited since the Cures Act, and as of the most recent assessment I could find there had been no federal enforcement of those rules at all. The claims were surfacing in commercial disputes rather than in penalties. [23] A mandate with a date is not a mandate with teeth, and the rulemaking here keeps moving in both directions. Budget for the interface cost. Do not assume Washington will delete it for you.

So Can It Work Now

Here is the practical version, and it is why I am spending my retirement on this instead of on a boat.

The cost to trial an AI reasoning layer in a non production environment is a rounding error against the cost of installing a product on a hospital network. No cart, no asset tag, no downtime window, no go live weekend, no training the night shift. You run it against a shadow feed or a de-identified extract, and you find out.

For years the only way to learn whether something worked was to put it in the room first. Now you can find out first and negotiate second. That is the inversion, and it is the first genuinely new thing I have seen in this industry since 1988.

Which brings me back to Darren.

An accelerator embedded in a real health system was the right idea in 2016 and it was constrained by the same physics as everything else. Everything it incubated eventually had to go through the doors. But an AI reasoning layer that can be evaluated outside the production network is exactly the kind of thing a program like that can now test at a speed it could never manage with hardware. Real clinical questions, real data, real clinicians, and a trial that does not require a go live weekend or a capital request. If I were still building, that is where I would want to be standing.

Can the Culture Change

I have been asking myself that for a very long time and I want to answer it honestly, which means answering it in two parts.

The culture will not change because we ask it to. It never has. Silos exist because the incentives inside them are real, and every one of those twenty four people has a legitimate reason for the veto they hold. Telling them to be less cautious is not a strategy, it is a complaint.

But cultures do change when the economics change underneath them, and they change fast when they do. That is the whole history of this industry and it is the whole history of every other one. The reason I am hopeful is not that AI is impressive. It is that for the first time the technology, the economics and the entry path are pointed the same direction, and the cost of finding out has collapsed.

None of this excuses founders from the work. Every failure in this article was engineered competently and defeated structurally, which means the scarce skill is not engineering. It is comprehensive knowledge of the space. Who signs, who pays, who is liable, which code applies, which committee can kill it, and which department will quietly refuse to staff it. That is not a caveat on top of the job. That is the job.

Nobody in this article failed for lack of talent. Tom Gregory, Bill Rutter, Phil Faris, Joe Condurso, Dave Geving, Jae Evans, Peter Thompson. They were right, and they were early, and being early is indistinguishable from being wrong right up until the moment it is not.

Some of what we built became ordinary. Duke never went back to paper. Every anesthesia department in the country runs an information management system now. Home monitoring is not experimental anymore, it is a product category. We were not wrong. We were early, and the bill for being early came to us instead of to the people who arrived later and got it right.

That is a fair trade, and I would take it again.

The structure has shifted in one specific place, for one specific reason, and the cost of testing that claim has collapsed. I have been wrong about timing before. I have a laser printer that proves it.

Worth a try. After watching better ideas than mine die in that parking lot, that is not a small sentence. It is the boldest one I have.

About me

Daniel Pettus is the founder of Inside the Loop and the inventor of AI MedAgent, a proof of concept for continuous, AI driven medication oversight, free and public at aimedagent.net. He spent forty years in medical device and health IT leadership at Alaris, CareFusion, and BD, co-founded iMetrikus in 1998, and contributed to IHE Patient Care Device interoperability standards. He holds two issued patents and has filed three provisional patents on the continuous reasoning method behind AI MedAgent. In November 2026 he will present “The dark cockpit: AI sits at the core of care, not at its edge” at AI.Care 2026 in Melbourne.

This newsletter is now read inside the companies that can build what it describes: the cloud platforms, the medication and device leaders, and the investors deciding where AI belongs in clinical care. He is available to industry, acute care organizations, and investors in an advisory capacity, as a subject matter expert in applying AI to proactive clinical care and medical device integration. The public data was never the hard part. The reasoning was, and that is the work he has spent a career on.

References

[1] Coleman RL, Stanley T III, Gilbert WC, Sanderson IC, Moyer GA, Sibert KS, Reves JG. The implementation and acceptance of an intra-operative anesthesia information management system. Journal of Clinical Monitoring 1997;13:121-128. Corporate identity footnote, p. 121.

[2] Contemporaneous San Diego business press on the June 1985 consolidation of Diatek Inc., Neurometrics and Diatek Medical Technology (formerly Dade Medical) into Diatek Corp., including market position, employment, the Sorrento Valley facility, ARKIVE development cost and Douglas Dayton’s backing.

[3] Gregory T, Pettus DC. An electroencephalographic processing algorithm specifically intended for analysis of cerebral electrical activity. Journal of Clinical Monitoring and Computing, 1986.

[4] Pettus D. The Data Was Public. The Reasoning Wasn’t. Inside the Loop, Issue 08. Full detail on ARKIVE, MediCompass and the UMLS public data mart.

[5] Yardley W. Douglas J. Dayton, First President of Target, Dies at 88. New York Times, July 10, 2013. See also Star Tribune, July 8, 2013.

[6] Anesthesia Information Management System Implemented at Duke. Imaging Technology News, October 17, 2006.

[7] Public professional record of Joseph Condurso: executive vice president and managing director, Diatek Medical and Arkive Information Systems, 1993-1995; cofounder and chief operating officer, iMetrikus; president and chief executive officer, PatientSafe Solutions.

[8] William J. Rutter (1927-2025), University of California San Francisco. See UCSF, Remembering William J. Rutter, PhD, and A History of UCSF biography.

[9] United States Patent 7,375,647, Method and apparatus for communicating data between a medical device and a central data repository. See also United States Patent 8,010,717.

[10] Bailey TS. Remote monitoring with MetrikLink improves diabetes care in an ethnic underserved population. Healthcare Across Borders, September 2003.

[11] Francisco Partners Acquires Qualcomm Life. Francisco Partners, February 2019. Divestiture of the 2net remote patient monitoring business to Philips Healthcare, 2020.

[12] Vocera to Acquire PatientSafe Solutions. BusinessWire, April 29, 2021. Funding history per PitchBook and Tracxn company profiles.

[13] Google Shuts Down Medical Records And Health Data Platform. TechCrunch, June 24, 2011.

[14] Microsoft HealthVault is officially shutting down in November. MedCity News, April 2019. Service ended November 20, 2019.

[15] Intel, GE launch telehealth joint venture. MobiHealthNews, January 2011. Intel Corporation Form 10-Q, FY2016. FTC early termination notice, December 17, 2019.

[16] Digital health growth hit by slow provider sales cycles. Modern Healthcare annual Digital Health Survey. See also Rock Health, Streamlining enterprise sales in digital health.

[17] Inside the Hospital Value Analysis Committee. Medical Product Outsourcing, March 2023.

[18] Launch of the Techstars Healthcare Accelerator in partnership with Cedars-Sinai. Cedars-Sinai Accelerator, April 2016. See also Los Angeles Business Journal, April 5, 2016.

[19] Costs of Caring: Challenges Facing America’s Hospitals as They Care for Patients in 2026. American Hospital Association, March 2026.

[20] Medicare Advantage Insurers Made 53 Million Prior Authorizations in 2024. KFF, January 28, 2026, as cited in AHA Costs of Caring, March 2026.

[21] Application Programming Interfaces, ONC/ASTP HealthIT.gov certification companion guide. Standardized FHIR requirement for certified health IT effective January 1, 2023.

[22] CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F), Centers for Medicare & Medicaid Services. Federal Register, February 8, 2024.

[23] End of Year Regulations on Interoperability. Crowell & Moring client alert, December 2024, noting the absence of federal enforcement of information blocking rules to date.

[24] Visa, Inc. history. National BankAmericard launched BASE I, the first electronic authorization system and the precursor to VisaNet, in 1973. See Britannica Money, Visa, Inc.

Every drug label the FDA has ever cleared is sitting on a public server right now. Free, structured, machine readable. So is a normalized map of nearly every medication name used in American medicine, built and kept current by the National Library of Medicine. So is the adverse event history of the country. You paid for all of it with your taxes, and for most of my career almost none of it was read while a patient was still in the bed.

I know how far ahead of the tools this idea can run, because I tried to build it more than twenty years ago, with SQL scripts and a room full of servers, and it was not enough.

I want to talk about the public data marts that have been hiding in plain sight, why I reached into them long before it was fashionable, and why an AI reasoning layer finally changes what they are for.

A Confession: I Tried to Build This in 2003

Long before I was connecting infusion pumps to the medical record [1], I helped build the Diatek Arkive, and I want to give that machine its due, because it was more than a data integrator. It was a full Anesthesia Information Management System, an electronic medical record built for the anesthesiologist and wired directly into the room. Touch screen workstations sat at the clinician’s side, integrated with the anesthesia machines and the physiologic monitors, capturing end tidal and inspired gases, drugs and infusions, and vital signs automatically, and assembling the anesthesia record as the case unfolded. [2] Commercial adoption began in 1988. Getting devices from a dozen manufacturers to speak to one system was hard in 1988, and the result was one of the first electronic records a specialty actually trusted.

Years later, at iMetrikus, I took the same idea home. I led the design of a device we called MetrikLink, a small gateway hub that sat in a patient’s house and captured readings from more than thirty home monitoring devices, glucose meters, blood pressure cuffs, scales, and sent them to our data center over a phone line or the internet. I made many trips to China, where we built it, because the only way the math of telehealth works is if the hardware is cheap. I wrote the patents, and I am named as co-inventor on United States patent 7,375,647. [3]

MetrikLink was only the plumbing. MediCompass was the system, and it came in two parts. [4] A home version on the patient’s desktop, paired with the MetrikLink hub to capture the device data, and a clinician office version that pulled the home data together and presented what mattered. Here is the part I am proudest of, because we got the philosophy right before we had the tools to deliver it. We ran what I called a dark cockpit. The term is borrowed from aviation, where Airbus decided in the 1980s that a cockpit should stay dark and quiet until something actually needs the pilot’s attention. [5] We did the same. The architecture watched everything and said nothing, until a value drifted out of range or a pattern turned worrisome. We stopped pushing information that was not actionable, because information that is not actionable is just noise wearing a lab coat.

The secret was in the data center, in the database, and this is where public data enters my story for the first time. We built our analytics on an OLAP database, and inside it we incorporated the UMLS Metathesaurus, the National Library of Medicine’s map that links medical vocabularies by meaning. [6] The NLM built the UMLS for exactly the problem we had, to let a computer recognize that a hundred different vocabularies are often saying the same thing, and to connect them. We did not download it with a click. We ordered it from the NIH as a stack of software CDs, at no cost, and installed it part by part. The process was complex. The result was worth it. With a stack of SQL scripts we resolved free text, whether it came from the patient at home or from the clinician’s office, into a Concept Unique Identifier, a CUI, the Metathesaurus’s stable code for a single medical idea. Once you have CUIs you can look for co-occurrence, the concepts that keep showing up together for one patient, and you can point the clinician toward the right literature in PubMed before a problem becomes an event. The academic world was working the same seam. There is a careful 2012 paper on using UMLS concept identifiers and co-occurrence to connect clinical text to the medical literature. [7] We were doing it live, in a shipping product, for real patients, back in 2003.

It was crude. It was a ton of code. Installing and using the UMLS was miserable work, and I will not romanticize it. But it was a real attempt to get past the number on the glucose meter and say something about the patient’s journey, using as much data as we could reach. No one else was reaching into a public data mart to reason about a remote patient inside a commercial telehealth product then. We were early, and we were alone.

So why have you never heard of any of this. Because the company ran out of money. That is the whole autopsy. The model was skewed toward acute care, the hospital, where the urgency and the dollars live. Managing a patient at home is a completely different animal, with different economics and a much longer payback. Anyone who wants the cautionary version of that story can follow what happened to Qualcomm Life. That is another article, and I will write it.

The point I want to carry forward is this. The instinct to build on public, authoritative, taxpayer funded data was right in 2003, and the instinct to stay dark until something is actionable was right too. What I did not have, what nobody had, was a way to make the reasoning itself intelligent. I had SQL. What I needed was a colleague who could read.

What a Data Mart Actually Is, and Why You Already Own These

Read more

INSIDE THE LOOP · SUPPLEMENT TO ISSUE 07

The laboratory never loses. The models get sharper, the reasoning improves, the science marches. Implementation is where good ideas go to die, and lately the cause of death is written on an invoice.

In Issue 07 I argued that the problem in patient safety is not the people and not the data, it is a design that waits for the next scheduled look. This is the companion piece. Suppose the design problem is solved and the reasoning works. What actually stops it from reaching a patient? I went looking for the honest answer, and while I was looking, a German hospital-IT executive published the same argument in the same week, from four thousand miles away, about a health system nothing like ours [7]. When two people who have never met describe the same wall from opposite sides of an ocean, it is worth asking what the wall is made of.

What the machine can already do, and what it cannot yet

Be honest about what the reasoning can and cannot do. The honesty is the whole credibility.

Start with a claim I can defend in both directions. Machine reasoning is improving quickly, and it is not a doctor in a box. Both are true, and anyone who tells you only one of them is selling something. On structured medical knowledge, frontier models now score in the low nineties on licensing-exam reasoning, better than most people expect [9]. On the hardest open-ended diagnostic cases, the same models still fall to around half right, which is nowhere near good enough to trust alone [10]. So the sober read is this: the technology is already strong at bounded, well-posed questions, still weak at open-ended ones, and getting measurably better at both every few months.

That is exactly why my own demonstration does not ask the model to diagnose. It asks a narrower and more answerable question. Given this patient’s trajectory, the orders already written, and the evidence around them, is this permitted action the right action right now. That is a bounded problem the reasoning handles well today, and it is the kind of problem that improves fastest as the models improve. I am not claiming a breakthrough in judgment. I am claiming that a capability which did not exist five years ago now exists, it is cheap, and almost nobody is testing it against live clinical logic. We are among the first who actually are, on real public data rather than a slide.

Two prices, one of them quietly enormous

The thinking is nearly free. The permission to think is not.

So if the science is improving, what is the wall? Money, but not the kind you would guess. I priced both ends of this in a companion research note. A single reasoning pass over a patient, the actual thinking, costs somewhere between one and ten cents depending on the model, and less once the patient’s context is cached and reused [1]. The other end, the plumbing that lets that reasoning see a real patient and act, is where the zeros pile up. A read-only feed from a major electronic health record starts around fifteen thousand dollars. A live, two-way interface that can take in vitals and send back an order runs one hundred fifty thousand to three hundred thousand dollars and most of a year to build [2]. A mid-sized system running fifty such interfaces reported more than six hundred thousand dollars a year just to keep them alive [3]. The thinking is nearly free. The permission to think is not.

A price, or a toll

There is a line between charging for work and charging for passage. Two countries crossed it the same week.

Here is where the German commentary matters, because it is not my voice and it is not my market. Bastian Stockhausen, who runs a hospital-IT services company in Germany, called the mechanism by its financial name: Cause of Death, EBIT [7]. His argument is the one I have been circling. An interface is a bottleneck through which all the information must pass, and large incumbent vendors price that passage far above what it costs to build or maintain, not for technical reasons but strategic ones. He gives a real example: a vendor asked for more money to supply a data server by itself than it had quoted for that same server bundled with its own patient portal, because the hospital had chosen a competitor’s portal. The partial service cost more than the whole package. As he puts it, that price has nothing to do with cost. It is the exercise of market power to punish a free purchasing decision.

His analogy is better than mine, so I will borrow it. Roughly a fifth of the world’s oil passes through the Strait of Hormuz, a channel a few kilometers wide. Whoever controls that passage does not need to own the oil. Controlling the transit is enough to move prices, and the bill is never paid by the powerful at the bottleneck, it is paid by the consumer at the pump. An interface in a hospital is the same shape. What is a consumer at the pump is a patient at the bedside. And both Stockhausen and I are careful about the same line, because it is the line that keeps this honest: genuine integration work deserves a real profit, and no serious person argues otherwise. The toll is not profit. The toll is the part of the price that exists to exclude a competitor rather than to deliver a service.

Even the law gets gamed

Germany made open interfaces mandatory. Guess what happened to the price.

You might assume a legal mandate for open interfaces settles this. Germany tried exactly that. Under section 373 of its social code, the agency Gematik was directed to define binding, standardized, modern interfaces for hospital systems, and compliance is mandatory for manufacturers [8]. Open data exchange there is not a favor. It is the law. And here is the sentence Stockhausen reports hearing from inside the industry, which I cannot improve on: if we are legally obliged to do so, we will just make it even pricier [7]. The mandate was not read as an instruction to open up. It was read as a new place to collect at the bottleneck.

We have the American version of the same lesson. Our Cures Act rules require certified records to expose a standardized, modern interface without special effort, and they bar fees designed to shut out competing apps [5]. A national exchange framework is live and expanding [6]. That genuinely lowers the cost of reading standardized data, and it is real progress. But it does not yet deliver a free, real-time, write-capable production feed, so the last mile stays tolled. Two continents, two mandates, one workaround: when you cannot refuse to open the gate, you raise the price of walking through it.

The four-billion-dollar anchor

A sunk cost that large stops being an investment and starts being an instinct.

One more number, because it reframes the incentive. Kaiser Permanente spent an estimated four billion dollars standing up its electronic record platform, at the time the largest privately funded record in the world [4]. I am not mocking that figure. It bought real capability, and running a system at that scale is genuinely hard. I raise it because a sunk cost that large changes behavior. It creates a powerful instinct to protect the investment, to keep spending to keep the legacy alive, and to treat every new connection as a toll that helps recover what was spent. That instinct is human and it is understandable. It is also, step by step, how an industry talks itself into spending more and more each year to stand exactly still.

Meanwhile, the patient walked out of the building

While hospitals argue over the price of a feed, the data got up and left.

Now the part the toll collectors are not pricing in. The continuous physiologic stream that hospitals guard behind expensive feeds is increasingly being generated somewhere else entirely: on the patient, at home, on hardware they bought themselves. In 2024 the Food and Drug Administration cleared the first over-the-counter continuous glucose monitors, Dexcom’s Stelo and Abbott’s Lingo, sold without a prescription for roughly the price of a phone case a month [11]. The wearable medical device market was about fifty-four billion dollars in 2025 and is growing at double-digit rates, with home health already its largest segment [12]. Remote patient monitoring is a market in the tens of billions and roughly doubling across the decade [12].

Read that against the interface problem and the picture turns. A reasoning layer runs over a consumer data stream for pennies, and that stream does not sit behind a forty-thousand-dollar gate, because the patient owns it. If the walled systems keep the drawbridge up, the data gravity and the innovation do not wait politely at the moat. They move to the wrist. The real risk to legacy healthcare is not dramatic disruption in a boardroom. It is quieter and more humiliating: getting routed around while spending more every year to defend a gate the world is learning to walk past. The reasoning that a hospital cannot afford to connect to its own record will run, cheaply, on the monitor a patient bought on the internet. That should terrify the toll collectors, and it should encourage the rest of us.

What this is actually asking for

This is not a demand to work for free. It is a demand to stop paying at a booth that produces nothing.

So let me be plain about the goal of this piece, because it is not merely to describe the weather. It is a call to stop treating the toll as a fact of nature. The ask is narrow and it is fair. Price an interface like the service it is, not like a checkpoint. Honor the open-standard mandates in their spirit and not just their letter. And let an integrated system, or any willing partner, test a promising method on data it already owns without a six-figure tax at the door. None of that asks a vendor to abandon profit. Stockhausen and I agree on that completely. It asks only that the price reflect work performed rather than passage permitted.

Bottlenecks do not fall because the gatekeeper has a change of heart. They fall when enough people downstream, the buyers, the clinicians, the hospital IT leaders, and yes the readers of a newsletter like this one, refuse to keep pretending the toll is a cost. Say the quiet part out loud often enough and charging the toll becomes an embarrassment. That is the grassroots version of progress, and it is the version available to people like us who have no boardroom and no leverage except being right and being loud.

It is not only greed, and the regulator is not the enemy

Hand the skeptic their due, and the argument gets stronger, not weaker.

I would be a poor journalist if I let greed carry the whole blame, so here is the fair accounting, and it holds a surprise. Money is not the only thing slowing this down. Someone has to own the liability when a recommendation is wrong, clinicians have to trust it, and governance committees have to approve it. All real. But the regulatory picture is lighter than most people assume, and it just got lighter. Under the 21st Century Cures Act, clinical decision support that lets a clinician independently review the basis of its advice, and does not ask the clinician to rely on it primarily, is generally not regulated as a medical device [13]. In January 2026 the Food and Drug Administration revised its guidance to cut red tape further, extending enforcement discretion even to tools that offer a single recommendation when only one is clinically appropriate, while leaning harder on one thing in particular: transparency about the data, the logic, and how a recommendation was reached, especially for tools driven by artificial intelligence [14].

Sit with that, because it describes the design I already chose. My demonstration is advisory. A physician accepts, adjusts, or rejects at every cycle, and the consult note exists precisely so the basis can be reviewed: the traceable reasons, the supporting evidence, and an honest statement of what the model could not see. Those are not decorations. They are the exact properties that keep transparent, human-in-the-loop support in the lighter-touch lane the FDA just widened. So I will not pretend a regulatory wall is the thing stopping this, because for advisory reasoning it largely is not.

I want to be very clear about tone here, because it matters. I am not anti-regulation and I am not anti-FDA. My proposed method needs oversight, and it should have it. A regulator asks the questions an inventor is too close to see, and proper oversight is not a tax on a good idea, it is part of what makes the idea safe enough to trust. I want the agency in the room, as a partner rather than an adversary. My one honest frustration is timing, not authority. Regulators move at the speed of process, and this technology moves at the speed of a software release. By the time guidance catches up to one generation of a tool, the market has morphed into the next. The answer to that is not less oversight. It is oversight that engages earlier, iterates faster, and treats a builder as a collaborator. The January revision is a sign the agency knows this. I want more of it, sooner.

And where the wall is real, I will say so plainly, because a careful reader deserves the whole map. Two features of this work can pull it back under full device regulation. Reasoning over a continuous signal from a monitor is treated differently, in the agency’s eyes, from reasoning over a lab value a clinician could look up, and my method watches trajectories built from device data. And the moment the loop actually closes, when the reasoning acts instead of advises, it is a device, full stop, and it should be. None of that is a reason to stop. It is a precise map of where the honest review belongs. But notice what none of it is: not one of those obstacles is a reason the plumbing should cost a quarter of a million dollars before the first patient is ever watched.

The cost of doing nothing

Inaction is not the neutral choice. It is the expensive one, and it is not paid in dollars alone.

Every argument in this piece has an alternative, and it is the one we are already living: change nothing. So price that honestly too, because doing nothing is neither free nor safe. If the current course holds, the trajectory is not in dispute. National health spending reached eighteen percent of the American economy in 2024, about five point three trillion dollars, and the government’s own actuaries project it to climb past twenty percent within the decade [15]. That is one dollar in five. The outcomes that money buys have barely moved, the avoidable harm I documented in Issue 07 keeps recurring at rates that have not fallen in a generation, and the population that must be cared for grows older every year, which only raises the demand and the bill. Hold the line exactly where it is and look at what you have chosen: prices that keep rising, outcomes that keep flatlining, and a widening gap between what the system costs and what it delivers. Inaction has a body count and it has an invoice. We simply do not print them on the same page, so we let ourselves believe they are not there.

Where I land

AI reasoning will get better. That is a fact, not a hope. The laboratory will not be stopped. The implementation might be, and if it is, the cause of death will not appear in any journal, because greed does not file a case report. But the way through is already arriving from three directions at once. The reasoning is getting cheaper and stronger. The standards are being pried open, slowly, on two continents, and the regulator has started clearing a path rather than blocking one. And the data is migrating onto devices the patient already owns, where no toll can reach it. The machine can already think well enough to help. The open question is whether we will let it, or spend another decade paying at the booth while the patient, wearing tomorrow’s monitor on her wrist, walks straight past us and out the door.

One last thing, from me

I will end closer to home than I usually allow. I have decided that aimedagent.net is finished, at least as a thing I keep building. It is live, it works, and it proves what I set out to prove. If someone finds a bug, tell me and I will fix it, but the demonstration has done its job. This argument was never going to be won in my code. It will be won, or lost, by whether the rest of us insist on it.

I can afford to be patient about that, and I want to be honest about why. Like a lot of families, mine has quietly taken on more of its own healthcare, because the system we pay so much for makes us wait months for an appointment and then routes that appointment through an approval process built to contain cost rather than to help. That experience changed me. I wear an Oura ring every day and a new Apple Watch, and I do a careful amount of my own AI-assisted research on my health, not to replace my doctors but because I have learned I have to meet them halfway and walk in prepared. I am old enough to know how to work this system. I even pay my primary a monthly fee, out of my own pocket, for nothing more than a better place in line if I need a consult. It is not cheap, and I can manage it.

What I cannot stop thinking about is that my children will manage it less easily, and my children’s children may not even know the system was once meant to work differently. They will not have my inside knowledge, and most of them will not have the money to buy their way up a waitlist. If we do not find a way, or make a way, to lower the barriers this whole piece has been about, they will inherit a thinner, slower, more expensive version of care and be told it is normal. For some of them that difference will be measured in comfort. For a few, it will be measured in years. That is not an inheritance I am willing to hand down without a fight.

So I am going to step away from this project for a while and give some other work its turn. The passion has not dimmed, only redirected, and I will keep publishing here, because the writing may be the part that matters most now. The machine can already think well enough to help. Whether we let it, and whether we finally tear down the toll booth standing between it and the people it could serve, is up to far more of us than me. I have said my piece. Now I am asking you to say yours.

The lab never loses. Let us stop making the toll the reason our children do.

References

[1] Anthropic. Claude model and API pricing, accessed July 2026. Per-pass cost is an estimate from published token rates, detailed in the companion research brief. https://platform.claude.com/docs/en/about-claude/pricing

[2] Arkenea. Epic EHR Integration: costs, APIs, and best practices, 2026. Read-only and bidirectional interface cost ranges. https://arkenea.com/blog/integrating-healthcare-app-with-epic-ehr/

[3] Nirmitee.io. Mirth Connect interface build cost and pricing matrix, 2026. Multi-interface program and annual maintenance figures. https://nirmitee.io/blog/mirth-connect-interface-build-cost-pricing-matrix-2026/

[4] InfoWorld. How Kaiser bet 4 billion dollars on electronic health records, and won. Kaiser Permanente HealthConnect implementation cost. https://www.infoworld.com/article/2272927/how-kaiser-bet-4-billion-on-electronic-health-records-and-won.html

[5] Office of the National Coordinator for Health IT. ONC Cures Act Final Rule: standardized FHIR API and information blocking. https://www.healthit.gov/topic/oncs-cures-act-final-rule

[6] Office of the National Coordinator for Health IT. Trusted Exchange Framework and Common Agreement (TEFCA), status 2025-2026. https://www.healthit.gov/topic/interoperability/policy/trusted-exchange-framework-and-common-agreement-tefca

[7] Stockhausen B. Cause of Death EBIT: Expensive interfaces block healthcare digitalization. heise online, 17 July 2026. Commentary; includes the FHIR-server pricing example, the Strait of Hormuz analogy, and the reported industry quote. https://www.heise.de/-11368331

[8] gematik / ISiK (Informationstechnische Systeme im Krankenhaus), Section 373 SGB V. Binding, FHIR-based hospital interfaces mandated for manufacturers; Basis Stufe 3 deadline July 2025. https://kodjin.com/blog/isik-compliance/

[9] Benchmarking large language models on the United States Medical Licensing Examination for clinical reasoning, 2025. Frontier models reach the low nineties on structured licensing-exam reasoning. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12796295/

[10] DiagnosisArena: Benchmarking Diagnostic Reasoning for Large Language Models, 2025. Top reasoning models reach roughly half accuracy on the hardest open-ended diagnostic cases. https://arxiv.org/abs/2505.14107

[11] MedTech Dive and CNBC, 2024. FDA clears first over-the-counter continuous glucose monitors: Dexcom Stelo (March 2024) and Abbott Lingo (June 2024). https://www.medtechdive.com/news/dexcom-sells-stelo-over-the-counter-cgm/725310/

[12] Fortune Business Insights and Grand View Research, 2025-2026. Wearable medical device market near 54 billion dollars in 2025 with home health the largest segment; remote patient monitoring in the tens of billions and growing at double-digit rates. https://www.fortunebusinessinsights.com/industry-reports/wearable-medical-devices-market-101070

[13] U.S. Food and Drug Administration. Clinical Decision Support Software, guidance for industry; non-device CDS criteria under Section 520(o)(1)(E) of the FD&C Act, added by the 21st Century Cures Act. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software

[14] U.S. Food and Drug Administration, revised Clinical Decision Support Software guidance, January 2026; enforcement discretion extended to single-recommendation tools and heightened transparency expectations for AI-driven CDS. Summary: Covington & Burling, Five Key Takeaways, January 2026. https://www.cov.com/en/news-and-insights/insights/2026/01/5-key-takeaways-from-fdas-revised-clinical-decision-support-cds-software-guidance

[15] Centers for Medicare & Medicaid Services, National Health Expenditure data and 2024-2033 projections; health spending was 18.0 percent of GDP in 2024 (about 5.3 trillion dollars) and is projected to exceed 20 percent within the decade. https://www.cms.gov/data-research/statistics-trends-and-reports/national-health-expenditure-data/nhe-fact-sheet

About the author

Daniel Pettus is the founder of Inside the Loop and the inventor of AI MedAgent, a proof-of-concept closed-loop medication management architecture. He spent forty years in medical device and health IT leadership at Alaris, CareFusion, and BD. He co-founded iMetrikus in 1998 and contributed to IHE Patient Care Device interoperability standards. He holds two patents with a third provisional filed.

A monitor watched a patient get worse for four hours and did exactly what we built it to do. That is not a malfunction. That is the design.

The glucose started climbing a little after two in the morning. The next scheduled check was at six. For four hours the trend sat in the record, correct and complete, and unread. When the nurse drew the six o’clock stick, the number was almost exactly where the earlier trajectory said it would be. The team did what good teams do. They corrected it, they charted it, they moved on to the next room.

Nobody failed. No one was careless, no policy was broken, no device malfunctioned. The monitor worked. The record worked. The people worked, and they worked hard. The system did precisely what it was designed to do, which was to wait for the next scheduled look and then react to whatever it found. That is the problem. Not the people, the design. And it is a problem I have spent the better part of a year trying to say out loud in a way that lands.

For thirty years I helped build machines that sound alarms. Faster alarms, smarter alarms, medication orders sent straight into the pump and into the chart. We got very good at telling a clinician that something is already wrong. We never taught the machine to reason about what to do next. An alarm is a threshold. It waits for a number to cross a line and then it shouts. It does not know where the number was ten minutes ago, or where it is heading, or what the other twelve numbers around it are doing. That is the line I want to cross: from a machine that notices to a machine that reasons.

Why me, honestly

I have no product to sell and no job to protect. That is the only reason I can afford to tell you the truth instead of a pitch.

People ask why me, and I owe you an honest answer rather than a flattering one. I do not have a tidy story. I am a retired medical device executive who cannot let go of a single idea, and that is most of it.

The rest is a pattern I only see clearly looking backward. In 1986 I worked on a processed EEG that gave anesthesia clinicians an early read on brain function, a continuous signal where before there was a guess. In 2003 we built MediCompass on the national medical terminology system, reasoning over public data to give clinicians and patients a clearer picture of therapy. Every decade or so I have made the same bet in a new form: take a signal that already exists, add a better way of reasoning over it, and try to catch harm before it lands. Some of those bets became ordinary. So when I say I do not have a clean answer for why me, the closest true one is that I have done this four or five times before it was obvious, and I am doing it again. I am retired, which means there is no product to sell and no job to protect. That is not a small thing. It is the only reason I can afford to be this blunt.

What we are actually buying

We spend twice what the rest of the wealthy world spends, and our patients live shorter lives for it. Read that sentence again.

I have made the money argument in these pages before, so I will not re-litigate it in full. The short version, updated with this year’s figures: in 2024 the United States spent an estimated 14,885 dollars per person on health care, against an average of 7,371 dollars across comparable wealthy nations. We spend roughly twice what our peers spend [1]. For that premium the Commonwealth Fund’s 2024 comparison placed the United States dead last among ten high-income countries on overall performance, and last specifically on health outcomes, with Americans projected to live more than four years less than the ten-country average [2].

Let me put a sharper edge on the word outcome, because it gets used loosely. I mean three things a patient or a family would actually recognize. How long you live. Whether you have to come back. Whether the treatment itself hurts you. Take the second one: nearly fifteen of every hundred Medicare patients discharged from a hospital are back inside thirty days, a rate that has barely moved despite a decade of federal penalties aimed straight at it [3]. Take the third: adverse drug events send more than 1.5 million Americans to the emergency department every year, and nearly 500,000 of those visits end in a hospital admission [4].

Here I have to be a careful journalist rather than a passionate one, because this is exactly where our field oversells. You may have read that medical error is the third leading cause of death in the United States, a figure of around 250,000 deaths a year. That number comes from a 2016 analysis, and it is genuinely contested [5]. More conservative work puts avoidable inpatient deaths closer to 26,000 a year [6]. I am not going to hand a skeptic an easy reason to dismiss the rest of what I say by pretending the high number is settled. The honest range runs from tens of thousands to a couple hundred thousand, the measurement is hard, and even the low end is a jetliner falling out of the sky every week or two. The point does not need the biggest number. It needs the trend line, which has been nearly flat for twenty-five years while the spending line has not.

Same data, same math, same result

Feed the same numbers through the same logic for twenty-five years, then act surprised at the flat line. That is roughly what we have done.

Here is my actual thesis, and it is a testable one, not a proven one. We have spent a generation and a great deal of money pointing better and better technology at this problem, and we have looked at the same data with the same kind of logic the whole time. A threshold. A rule. An algorithm that maps one measured variable to one response. When you feed the same data through the same kind of math, you should not be shocked to keep getting the same kind of result.

Something changed recently, and it changed fast. A large language model does not reason the way a control algorithm reasons. A rule asks a single question: has this one value crossed its line. A model can hold the whole picture at once, the trajectory of the glucose and the vasopressor already running and the potassium drawn an hour ago and the drug’s known adverse-event signal, and weigh them together in plain clinical logic, the way a good clinician does on rounds. It is not magic and it is not a diagnosis. It is a different kind of reading of the same stream we already collect.

And there is more of that stream sitting in the open than most people building in this space seem to use. My demonstration draws on four public federal data marts. When I went looking for what else is freely accessible, the answer was a lot. The openFDA service alone exposes far more than the adverse-event feed everyone cites: structured drug labeling, the recall and enforcement database, the National Drug Code directory, and the device adverse-event record known as MAUDE, all as free JSON with no login [7]. The Centers for Medicare and Medicaid Services publishes hospital-level readmission, mortality, and complication data through its Provider Data Catalog, the machinery behind Care Compare, as open downloads and interfaces [8]. This is the raw material of patient safety, already public, already machine readable, and mostly unread by the systems that could act on it. Reasoning is the part we were missing, not data.

How it actually works

The data to catch this kind of harm is already public, already free, already machine readable. The part we never built was the reasoning.

The part that makes it an invention rather than a faster alarm is what happens when you let the reasoning close the loop instead of only advising. Every pass returns one of three decisions. Act, when a change is warranted and fits inside the bounds the clinician ordered. Hold, when the right move is to do nothing and watch. Escalate, when a change is warranted but the controller cannot properly deliver it, because the ordered envelope is spent or the correct response is one it is not authorized to make. That third decision is the whole ballgame. A feedback controller sitting at 93 percent of its ceiling computes the next increment and sends it, every time, because that is all it can do. Asked in one run to nudge a vasopressor two units below its ordered maximum against a still-falling pressure, the reasoning engine declined. Its logic: the step is permitted but not enough, and spending the last authorized increment would exhaust its authority while delaying the clinician who actually needs to be called. An action is not automatically the right action. No fixed algorithm can hold that distinction, because no fixed algorithm has a concept of its own authority as something finite. That is not a rule behaving. That is reasoning.

In FSD mode the system does not stop at a recommendation. It reasons over the live stream and decides: it issues a medication order and drives it through ordering, pharmacy, dispensing, and the administration record, and it changes an infusion rate by commanding the pump directly. It also decides when not to act, and when the right action is one it is not authorized to take, in which case it hands the decision back. The clinical team is aware and supervising throughout, exactly as a driver supervises an autonomous vehicle. This is a demonstration of a future capability. It is not cleared for clinical use

A reason

Tens of thousands of avoidable deaths a year is not a number to manage down by a percentage point. It is a design we chose. Designs can be un-chosen.

I am not going to dare anyone to try this. Daring is cheap and this subject is not. I will make a quieter case. We have spent twenty-five years and a fortune defending a design that waits for the next scheduled look, and the outcomes that matter to patients have barely moved. A genuinely new kind of reasoning arrived in the last few years, it runs over data we already collect and already publish, and for the first time it can weigh a whole patient the way a clinician does rather than watching a single line for a threshold. That might move the needle. It might not. I believe it is worth finding out, and I believe protecting the method while we find out is the sensible thing any diligent person would do, which is why the approach sits under three provisional patents rather than in a slide deck.

For me there is no product at the end of this, and there never was. If there is a personal payoff, it is narrow and honest: helping people who want to build clinical solutions on AI reasoning and public data do it well and do it safely. That is the whole ambition.

If you have read this far, I have three asks. Argue with me in the comments, because I read every one and the sharp disagreements are where this gets better. Send it to the one person you know who should see it. And if you want to see the method reason over live federal data for yourself, go to aimedagent.net and request an access code.

Request Access Code

Nobody failed. Let us build something that does not need them to.

References

[1] Peterson-KFF Health System Tracker. How does health spending in the U.S. compare to other countries? 2024 per-capita spending: United States 14,885 dollars; comparable-country average 7,371 dollars. https://www.healthsystemtracker.org/chart-collection/health-spending-u-s-compare-countries/

[2] The Commonwealth Fund. Mirror, Mirror 2024: A Portrait of the Failing U.S. Health System. United States ranks last overall and last on health outcomes among ten high-income countries. September 19, 2024. https://www.commonwealthfund.org/publications/fund-reports/2024/sep/mirror-mirror-2024

[3] Centers for Medicare & Medicaid Services. Hospital Readmissions Reduction Program (HRRP). 30-day risk-standardized unplanned readmission measures; U.S. average readmission rate approximately 14.7 percent. https://www.cms.gov/medicare/payment/prospective-payment-systems/acute-inpatient-pps/hospital-readmissions-reduction-program-hrrp

[4] Centers for Disease Control and Prevention, Medication Safety (FastStats); Shehab N, et al. US Emergency Department Visits for Outpatient Adverse Drug Events, 2013-2014. JAMA. 2016;316(20):2115-2125. More than 1.5 million ED visits annually; approximately 27 percent result in hospitalization. https://www.cdc.gov/medication-safety/data-research/facts-stats/index.html

[5] Makary MA, Daniel M. Medical error, the third leading cause of death in the US. BMJ. 2016;353:i2139. Estimate of approximately 251,000 deaths; methodology subsequently contested. https://pubmed.ncbi.nlm.nih.gov/27143499/

[6] Agency for Healthcare Research and Quality, PSNet. Measuring and Responding to Deaths From Medical Errors. More conservative estimates place avoidable inpatient deaths near 26,000 per year. https://psnet.ahrq.gov/perspective/measuring-and-responding-deaths-medical-errors

[7] openFDA, U.S. Food and Drug Administration. Drug and device API endpoints: adverse events, structured product labeling, recall enforcement, NDC directory, and device MAUDE. Free public JSON, no authentication. https://open.fda.gov/apis/

[8] Centers for Medicare & Medicaid Services. Provider Data Catalog and Care Compare. Hospital-level readmission, mortality, and complication measures published as open data. https://data.cms.gov/provider-data/

[9] AI MedAgent Technical Datasheet. Reference architecture, data flow, and governance for the advisory method. aimedagent.net. https://aimedagent.net

Daniel Pettus is the founder of Inside the Loop and the inventor of AI MedAgent, a proof-of-concept closed-loop medication management architecture. He spent nearly forty years in medical device and health IT leadership at Alaris, CareFusion, and BD. He co-founded iMetrikus in 1998 and contributed to IHE Patient Care Device interoperability standards. AIMedAgent methods are patent pending.

A neighbor told me a story last month that I have not been able to shake. He drove the family Tesla on Full Self Driving from Murrieta to a relative’s home in the San Fernando Valley, and the car handled all of it. Point to point and back. Merging, lane changes, the grind of Los Angeles traffic that rattles seasoned human drivers, and finally the surface streets on the far side. His hands stayed in his lap. He was supervising a machine that was driving better than he would have, and by his account it was, frankly, brilliant.

The most advanced piece of autonomous technology in his day was not in any hospital. It was under his seat.

I heard that story and thought about my own field. Then, a few days later, I lived the same lesson myself. I was in Hollywood with a friend visiting the Petersen Automotive Museum, of all places, a shrine to the machine we are now teaching to drive itself. When we finished, we took a self-driving Waymo back to our hotel through the busy streets of Hollywood. No safety driver. No hands on anything. No issues. Two ordinary Southern California errands, both chauffeured by software, and neither one remarkable to anyone but me.

I have spent nearly fifty years building the machines that deliver medicine. In 1988 I helped connect every monitor, ventilator, and infusion pump in an operating room into a single automated record, at a time when there were no standards to connect anything to anything. I know this field down to the wire. So what follows is not a technologist’s daydream. It is a builder’s frustration. We will let a car drive itself across four lanes of freeway, with strangers in the next lane, but we will not let the pump beside a patient’s bed adjust its own rate, even where the science to do that safely has existed for most of my career.

The Uncomfortable Part, Said Out Loud

Closed-loop medicine already works. It is in safe, routine use, today, almost everywhere except the United States.

The evidence is not thin. Target-controlled infusion, in which the system manages an anesthetic drug to a chosen blood or effect-site concentration rather than to a fixed rate, was first approved in Europe in 1996. It is now available in more than ninety countries and has been used to deliver anesthesia and sedation to well over twenty million patients.[1] The one notable holdout is here. The FDA has never cleared a target-controlled infusion device for routine use, so in the United States the technology survives only inside research studies.[2] The reasons are regulatory, not clinical. The safety record abroad is long and well documented.[3]

And lest closed-loop control sound exotic, remember that we already trust it here with one of the most dangerous drugs we give. The artificial pancreas, an insulin pump that reads a continuous glucose sensor and adjusts its own delivery, is FDA-cleared and sold in American pharmacies.[4] Hundreds of thousands of people wear one right now. A machine is titrating insulin, unsupervised, on the body of a child asleep in the next room, and we consider that unremarkable. We have already crossed the line we claim we are afraid to cross. We just crossed it quietly, one disease at a time.

The Real Objection Is Not Safety. It Is Habit.

When I raise this, the reflex answer is safety. It is always safety. And I understand the instinct, because I have spent a career respecting it. But look honestly at what the objection actually protects. We accept a closed loop for anesthesia everywhere but here. We accept a closed loop for insulin even here. What we will not accept is the idea, in general, that a machine should decide the next dose. The line we defend is not a line of evidence. It is a line of comfort. It moves, disease by disease, country by country, and it moves in exactly one direction.

Elon Musk drew the same fire with Tesla. Autopilot was a nice-to-have, a lane-keeper for the freeway, and no one lost sleep over it. Full Self Driving was different. It was a claim that the machine should hold the wheel, and it provoked genuine controversy, some of it deserved and some of it pure reflex. But the rest of the automobile industry did not respond by writing op-eds about why it could never be done. They responded by scrambling to catch up, because they understood that the nice-to-have was becoming the must-have, and that the company willing to say so out loud would define the category.

I think medicine is standing exactly where the car industry stood a decade ago, and telling itself exactly the same comforting story.

What I Built, and Why the IV Pump Is Only the Doorway

Read more

Before I was a software engineer. Before infusion pumps and IHE standards and forty years in medical device connectivity, I was a kid in a high school darkroom, learning to develop film by feel because the lights had to stay off.

That is where I met Myra. 1968. She was fifteen. I was learning lighting setups, main light, fill light, hair light, on a Mamiya/Sekor camera loaded with Kodak Tri-X. We did not know it yet, but the two of us had just enrolled, tuition free, in one of the cleanest case studies in market disruption this country has ever produced. The syllabus turned out to be brutal. It took thirty years to play out, and I was there for the whole thing, on both sides of it.

This issue is a departure. No EHR data. No HITECH spending figures. Just a story, because I think it explains something about AI that a citation cannot. Consider it a summer break from the footnotes. I left a few in anyway. Old habits.

A Camera Develops a Mind of Its Own

Here is the part most people get wrong about Kodak. They think the company missed digital photography. It did not. Kodak invented it.

In 1975, a Kodak engineer named Steve Sasson built the first digital camera in the company’s own labs. It was the size of a toaster and took 23 seconds to capture one blurry black-and-white image. Today the camera in your doorbell does better while you are still fumbling for your keys. His own management’s reaction to the toaster, by his own account: cute, but do not tell anyone about it. [1]

Eleven years later, in 1986, Kodak went much further. Company scientists built the world’s first megapixel sensor, a fingernail-sized chip that crossed the one-million-pixel line and put a genuinely filmless future within reach. Kodak did not just dabble in digital. Kodak invented it, led it, and then filed it under do not disturb. [2]

Then Kodak’s own internal research told them exactly what was coming. In 1981, the company commissioned a study to model the adoption curve for digital photography against film. The findings: digital had the technical capability to fully replace the film business, and Kodak had roughly ten years to prepare for it. [3]

Ten years of advance warning, from their own research, after their own engineers had already built the technology twice.

What did Kodak do with the warning? They spent it propping up film. The CEO Kodak hired away from Motorola in 1993 to drag it into the digital era, George Fisher, summed up the mindset he inherited in one line to the New York Times: Kodak regarded digital photography as the enemy, an evil juggernaut that would kill the business that had fueled the company’s profits for decades. Not a competitor. The enemy. [4]

They were not behind. They were ahead, by a decade, with the warning in hand, and they chose to fight the thing they had invented rather than build toward it.

By 2003, Kodak still employed roughly 64,000 people. By 2011, that number had fallen to around 17,000. [6] In January 2012, the company filed for Chapter 11 bankruptcy. [5] The 132-year-old company that put a camera in every American household did not lose to a competitor. It lost to its own decision to defend a workflow instead of reasoning about what its own invention actually meant.

Kodak is not the only ghost in this story. While Myra and I were building our photography business, another familiar name was making nearly the identical mistake in a completely different industry.

Borders Books was, at its peak, the second-largest bookstore chain in America. Big stores, deep inventory, a genuinely superior in-store experience. In 2001, when the internet had already started to look like a real threat to book retail, Borders made a decision that still reads as almost unbelievable: it outsourced its entire online sales operation to a young company called Amazon. As one retail analyst later put it, that was less like building a defense and more like handing the keys directly to a competitor. [7]

Borders kept its physical stores beautiful, kept its inventory deep, and let the company that would eventually destroy it run its website for years.

In July 2011, Borders announced it would liquidate. Almost 11,000 employees lost their jobs. [8] The CEO’s farewell statement cited the rapidly changing book industry, the e-reader revolution, and a turbulent economy. All true. None of it was the real reason. The real reason sat one click away on every customer’s screen, a company Borders itself had pointed them toward.

If you had put $1,000 into Amazon on the day of its 1997 IPO and never touched it, that investment would be worth more than two million dollars today, and closer to three by some calculations, depending on the entry price you use and the day you check. [9] Borders is worth zero. Both companies saw the same wave coming. One built the surfboard. The other handed someone else the surfboard, kept selling beach towels, and wondered what happened.

What This Has to Do With a Wedding Photographer From Murrieta

Here is where the story turns personal.

Read more

You have heard the number from me before. The United States spends about $14,885 per person on healthcare, roughly twice what other wealthy nations spend, and the Commonwealth Fund still ranks us last among them on outcomes. [1] I have said this before. It is worth repeating, because nothing about it has changed. We pay the most. We get among the worst results.

So let me get past the number and to the point.

What 40 years taught me, and where I now think I was wrong

I spent four decades in this industry. For most of it, I believed the same value proposition everyone around me believed: more technology means a safer, better patient experience. I sold it. I built it. I believed it.

I am retired now, with no company to protect and no quota to make. And I have come to believe something close to the opposite. We may be solving the wrong problem with solutions that cost a fortune and return very little where it actually counts, in the patient’s experience and possibly their outcome. We spend double and get little back. At some point the honest thing is to ask whether the approach itself, not the effort, is the problem.

Then I started using AI. Not as a chatbot. As a development and research partner. And it changed how I see the whole thing.

The idea

Here is what I invented. A method with two moves, and both run the same direction, forward instead of back.

The first move inserts AI across the entire patient encounter. Continuous reasoning by a large language model, reading the data stream the whole time, looking for the chance to intervene early, before the crisis.

The second move inserts AI into the moment therapy changes. When a new or modified medication order is placed, the AI follows the clinical workflow, the ordering, the pharmacy, the dispensing, and reasons at each step, speaking up when something is wrong.

Both are a reversal of how we work today. Instead of waiting for a threshold to be crossed and then reacting, the method manages the patient and the clinical process proactively, with alerts and stops that fire before the retroactive scramble begins. That matters, because more than 75% of medication errors happen at the prescribing and administration stages, [2] and preventable drug events still contribute to tens of thousands of hospital deaths and tens of billions in cost every year in this country. [3]

I will go one step further, because I am retired and can. I believe this path leads, eventually, to semiautonomous care. I know that word is unsettling. So was a car with no driver, right up until the moment a robotaxi picked someone up and it became ordinary. [4] The direction of travel is the same.

I believed in this strongly enough to file a provisional patent. I think the method is unique, and I think it has merit. The method is called AI MedAgent.

How do I prove it works

Let me be crystal clear about what this is not. I am not building a product. I am not starting a company. As one retired person, that is simply not achievable, and I have no interest in pretending otherwise.

That does not mean I cannot show the functionality and the potential. It just means I show it a different way. I have been working with AI to build a working model and demonstration. Like any complex project, it had its false starts and its failures. I flushed an early version and started over. It is now at a place where it demonstrates well.

Here is what it does, and here is exactly what is real and what is not.

It simulates the hospital data feed coming from the EHR and the interface engines. That part is simulated. I am not a hospital and I am not an EHR company, and I will not pretend to be either.

That feed is then sent to Claude AI in the cloud, where the reasoning is real, no simulator. The AI reasons over the live feed against four public federal data sources: the NIH RxNorm and RxClass drug classification services, the FDA’s adverse event reporting system (openFDA FAERS), the FDA’s DailyMed structured product labeling, and ClinicalTrials.gov. [5] When it detects an issue, it responds in two parts. Part A is the alert. Part B is the reasoned suggestion for what to do about it. The demonstration shows you both.

In the real world, those alerts would need a home, most likely inside the EHR. And a future version could adjust the IV pump or the ventilator on its own. That is a conversation for another day.

The demonstration is available to a limited few, and there is a plain reason for that. As a single developer, I have only so many AI tokens to share.

So here is the deal. If you believe there is value in reviewing what I have built, click the button below to reach my request form. If accepted, I will send you the access code and the website, and I read every request myself and reply within one to three business days.

Request access to AI MedAgent

I spent a career watching good ideas die from apathy, caution, and turf. I would rather hand you a working idea and be told exactly why I am wrong than sit on it and never ask. The worst thing I could do with this is keep it to myself.

So I am not going to.

Dan

Sources

[1] Peterson-KFF Health System Tracker, US health spending versus comparable countries, 2024 (about $14,885 per capita, roughly twice the comparable-country average); Commonwealth Fund, Mirror Mirror 2024, which ranked the US last among ten high-income nations on health system performance.

[2] Reported in the clinical literature that more than 75% of medication errors occur at the prescribing or administration stages.

[3] StatPearls / NCBI: preventable adverse drug events are associated with an estimated 44,000 to 98,000 US hospital deaths annually and roughly $37.6 to $50 billion in added healthcare costs, disability, and lost productivity.

[4] Autonomous robotaxi services (for example Waymo) now operate as paid, driverless rides in multiple US cities, a capability widely considered science fiction until recently.

[5] Federal data sources queried live by the demonstration: NIH RxNorm / RxClass, FDA openFDA FAERS, FDA DailyMed, and ClinicalTrials.gov.

That’s not a confession. That’s how the work gets done. A bench prototype exists to prove a concept can behave the way you think it will. It is not required to survive contact with a real patient, a real nurse’s workflow, or a real hospital’s security review. The actual discipline of device engineering, the part that takes years and the part nobody puts in a demo video, is the distance between “it worked on the bench” and “it’s cleared for clinical use.” I spent most of four decades living in that distance.

This week I built something that drifted toward looking finished before it was, and then I pulled it back.

The Bench

I’d been building an interactive demonstration of AI MedAgent’s reasoning method: an AI layer evaluating a post-CABG ICU patient and recommending a medication change. Partway through, I pushed further and wired a version of it to genuinely query live federal data sources, deployed it, tested it. It worked in pieces and then it didn’t, the way early integration work always does. So I made the call any device engineer makes at that stage: pull it off the bench, be explicit about what’s a concept demonstration and what isn’t, and don’t let anyone downstream mistake a prototype for a finished product. The site now says plainly what’s illustrative and what’s a scripted example. That’s not a retreat. It’s the same rule I’d have insisted on if this were an infusion pump instead of a website.

The Bedside

Here’s the part that actually matters, and it’s bigger than my website.

That gap, between a demo that looks transformative and a system that survives production, is the entire history of health IT for the last twenty-five years. CPOE was supposed to eliminate medication errors. It caught some and introduced alert fatigue that clinicians now click through by reflex. Interoperability initiatives were supposed to let a patient’s data follow them. Data followed the patient into more fields to populate, not fewer decisions for a clinician to make alone. The Surgeon General’s own workforce advisory puts nurse documentation time at close to 40 percent of a shift, not because nurses got slower, but because the systems that were supposed to save time moved the paperwork instead of removing it. None of this was fake. All of it worked on the bench. Most of it never closed the distance to the bedside, and the industry kept shipping anyway, because shipping the demo felt like progress, and admitting the gap didn’t.

AI is about to do this at a scale and a speed nothing before it has matched, because nothing before it has ever been this easy to make look finished. A model that reasons fluently about a clinical scenario is, on a bench, indistinguishable from one doing it for real. The distance to the bedside hasn’t gotten shorter. It’s gotten harder to see.

The Math

Here’s why that distance matters more now than it ever has.

By 2030, for the first time in this country’s history, there will be more Americans over 65 than there are children. By 2050, nearly one in four of us will be 65 or older, up from 17 percent today. That’s not a forecast anyone’s arguing about. It’s already baked into who’s alive right now and getting older on schedule.

Per-person healthcare spending on someone 65 and older already runs more than five times what we spend on a child, and over double what we spend on a working-age adult, by CMS’s own numbers. National health spending crossed $5.7 trillion in 2025, up 7.3 percent in a single year, and federal actuaries project the share of the economy it consumes will climb from 18 percent today past 20 percent within the decade. Spending is growing faster than the economy that has to fund it. Layer a population that’s aging into the group that costs the most onto a spending curve already outrunning GDP, and you don’t get a slow decline. You get a curve that bends toward collapse, on a schedule that’s already written.

You cannot staff, document, or bill your way out of that math. The only lever big enough to matter is spending the same dollar and getting more outcome for it, which means the reasoning has to get smarter before the patient gets sicker, not after. That’s the whole premise of everything I write here. It’s also, not coincidentally, why I keep building demonstrations of a reasoning layer instead of another dashboard.

The Garage

Am I the person who fixes this. No. I’m not twenty-five with nothing to lose and a garage. What forty years of watching this exact failure mode, demo outruns deployment, promise outruns proof, across every wave of health IT connectivity I’ve touched, closed-loop infusion integration included, actually buys you isn’t the ability to build the fix alone. It’s the ability to recognize the pattern fast, including, this week, in my own work, before anyone downstream had a reason to trust something that wasn’t ready yet.

But recognizing a pattern isn’t the same as accepting it. This is fixable. Not by patching what’s already there, another dashboard, another point solution wired onto the same workflow, the kind of incremental thinking that’s given us a $5.7 trillion system with outcomes that haven’t moved in step. Fixing this takes the other kind of thinking. Vision. A real appetite for risk. The energy to look at something everyone agrees is broken and refuse the easy version of solving it.

AI MedAgent, at the concept stage it’s honestly at right now, is my attempt at that different lens: a reasoning workflow, not a chat window, built to weigh outcome against cost the way a good clinician does, faster than any clinician has time for. Whether it or something like it turns out to be the actual answer, I can’t promise you. I can tell you it’s the right shape of question.

Which raises the one I don’t have an answer to. Are we putting our money where that kind of thinking happens? Healthcare has never been friendly to a twenty-five-year-old with nothing to lose and a garage, too regulated, too slow, too expensive to fail fast in. But reasoning that’s cheap and capable enough to sit inside a clinical workflow instead of a chat window is genuinely new. It wasn’t available the last time anyone tried to fix this. Whether that’s enough to break twenty-five years of pattern, I don’t know. Nobody does, not really, not until somebody with a different lens than mine, in a garage or a lab or a resident’s call room somewhere, decides this is the problem worth staking everything on.

I still believe someone fixes this. I just believe, forty years in, it’s more likely to be someone who hasn’t spent forty years being told how healthcare works.

By the time you finish reading this sentence, somewhere in a U.S. hospital, a preventable medication error is happening. The statistics say roughly one person will die from one before you get to the end of this article. You paid $14,885 per person last year for this. Switzerland, the second-highest spender in the world, paid $9,963. Japan, with better outcomes, paid $5,200.

We are getting robbed. And the thief is us.

I retired from Becton Dickinson in 2019. Not because I gave up. Because I was tired of pushing a very heavy rock up a very resistant hill, and I needed to breathe.

I spent forty years inside the machine. Alaris. CareFusion. BD. I helped connect the first generation of hospital infusion pumps to the electronic medical record. I co-authored the standards that made device-to-EHR integration possible. I held the patents. I fought the internal battles. I watched the external ones. I know exactly what this industry looks like from both the inside of the executive suite and the inside of the ICU.

I know what it costs when we get it wrong.

And then, after five years of watching from a distance and growing quietly furious, I came back. Not to a job. To a problem I cannot let go of.

That is what this issue is about

The Number That Should Keep Every Hospital Executive Up at Night

Between 44,000 and 98,000 Americans die in hospitals each year from preventable medical errors. That is the peer-reviewed range, established by the IOM’s landmark “To Err Is Human” report [1] and consistently validated by subsequent research including work from Johns Hopkins [2] and StatPearls in 2024 [3]. The upper estimate exceeds annual deaths from motor vehicle accidents.

Take the midpoint. Say 70,000 people per year. That is 192 people per day. Eight per hour. One roughly every 7.5 minutes.

You just read three sentences. In that time, statistically, someone in a U.S. hospital died from a mistake that did not have to happen.

We spend more money per person on healthcare than any other country on earth. More than Switzerland. More than Germany, Sweden, Norway, France, or Japan. The Peterson-KFF Health System Tracker confirmed it again in March 2026 [4]: the U.S. spent an estimated $14,885 per capita on healthcare in 2024. Switzerland, in second place, spent $9,963. The average for comparable wealthy OECD nations: $7,371.

We spend twice the average of our peers. We have the lowest life expectancy among those same peers, 79 years, 3.7 years below the comparable country average of 82.7 [5]. The Commonwealth Fund’s Mirror, Mirror 2024 cross-national comparison ranked the U.S. last among ten high-income countries on health outcomes specifically [6], a distinction that has not changed meaningfully in over a decade.

This is the deal we are getting.

The $30 Billion Bet That Mostly Moved Safety Scores

In 2009, as part of the American Recovery and Reinvestment Act, Congress passed the HITECH Act and set aside what the Congressional Budget Office estimated at $30 billion for Medicare and Medicaid Electronic Health Record incentive payments from 2011 through 2019 [7]. The objective had a name: Meaningful Use.

The logic was sound. U.S. hospitals were scandalously behind the rest of the developed world in digitizing patient records. At the time HITECH passed, 90 percent of hospitals did not have EHRs meeting federal requirements. The industry was roughly a decade behind comparable sectors in technology adoption.

The money worked, in the narrow sense. By 2015, nearly 78 percent of office-based physicians had certified EHRs. Hospital adoption accelerated dramatically. Epic, Cerner, and a handful of other vendors built what became the central nervous system of modern American hospital operations.

What the money did not do, and this is the argument nobody in the EHR industry wants to have, is move the outcomes needle in a meaningful way.

A 2016 analysis published in the journal Medicine, comparing outcomes at hospitals with high EHR adoption to those with low adoption, found that the association between EHR adoption and improved mortality, readmission, and complications did not hold up once other patient and hospital factors were accounted for [8]. The EHR improved documentation. It reduced specific, measurable categories of error. It connected clinical data that had been trapped in paper silos. It did everything it was architect to do.

It was architect to improve documentation. Not outcomes.

There is a difference. An important one.

Read more

Issue 03 // Inside the Loop

New York City, 1948. A research library on the Upper West Side. Seven professors. Nine years of cataloguing music without once listening to it.

San Diego, 2025. A retired executive with a retired executive’s amount of free time. One 1948 Howard Hawks film, watched once, then again, then again. And a question that would not let go.

I want to tell you about a play I wrote.

I know. This newsletter is about artificial intelligence, healthcare outcomes, and the decisions that matter. We have covered a Duke University cafeteria in 1989 and an insurance executive who said the quiet part out loud in 1998. We are building toward a conversation about EMR adoption, diversion detection, and the architecture that might finally close the loop.

This issue is about a jazz play set in 1948. Stay with me.

The 1948 Samuel Goldwyn film A Song Is Born is one of the most unusual artifacts in the history of American cinema. Howard Hawks directed it. Billy Wilder wrote the original story. And Goldwyn assembled on a single Hollywood soundstage what may be the greatest concentration of jazz talent ever captured on film: Benny Goodman, Louis Armstrong, Tommy Dorsey, Lionel Hampton, Charlie Barnet, Mel Powell. Together. At once.

The premise is pure screwball comedy. Seven academics have spent nine years compiling the world’s most ambitious encyclopedia of music. They know every date, every tempo, every published fact about American music. They have never once listened to it. Then a nightclub singer arrives, running from a gangster who needs her not to testify against him. She needs somewhere to disappear. Two days in a library change everything.

I watched it once and could not stop thinking about it.

Not because of the comedy, though the comedy holds up. Not because of the music, though the music is extraordinary. Because of the central argument, which the film makes almost by accident and which I recognized immediately.

The professors know everything about music. They understand nothing about it.

I have spent fifty years watching the same mistake in healthcare. We know everything about our patients. We understand very little.

Read more

San Diego. 1998. A conference room full of insurance executives. And a sentence I have never forgotten.

It started, as most things do, with a problem nobody wanted to admit existed.

The year was 1998. I had spent the previous decade inside operating rooms, building systems that captured data automatically from anesthesia monitors, ventilators, and IV pumps. ARKIVE had taught me one irreducible truth: the moment you stop asking humans to manually transcribe data and let the machines talk directly to each other, everything gets more accurate, more timely, and more useful.

The question I kept asking myself was simple. If we could do this in the OR, why couldn’t we do it at home?

Chronic disease patients (diabetics, cardiac patients, COPD sufferers) were generating diagnostic data every single day. Blood glucose readings. Blood pressure. Weight. Respiratory function. All of it measured, all of it clinically meaningful, almost none of it making it back to the physician between quarterly office visits. The patient drove to the clinic. The doctor reviewed three months of data in twelve minutes. Adjustments were made based on a snapshot, not a story.

We thought we could change that. So we built MetrikLink.

The concept was straightforward, which is not the same as saying it was simple.

MetrikLink was a home health connectivity hub, a small device that sat in the patient’s home and daisy-chained from the telephone line. No cell phones yet. This was 1998. You worked with what existed. We designed specialized adaptors for blood glucose monitors, blood pressure cuffs, thermometers, and scales. We added our own device: a spirometer called AirWatch that measured respiratory function for COPD and asthma patients.

I had spent enough time in China by then to know where to manufacture it affordably. We needed the

cost to be almost nothing because our model was radical for its time: free to the patient. Free to the physician’s office. No subscription. No hardware fee. No per-reading charge.

MetrikLink would upload the day’s diagnostic readings with a single button press. One button. An LED confirmed the transfer was complete. No screen. No menu. No instructions to follow. The whole transaction took seconds. That was intentional. We knew that complexity was the enemy of compliance, and compliance was everything in chronic disease management.

On the clinical side, we built a web portal called MediCompass. Physician offices and hospital clinics logged in and saw their patient population at a glance. Here is where we brought in a concept I had been thinking about since my aerospace days: the dark cockpit.

In aviation, a quiet cockpit is a safe cockpit. Pilots are not bombarded with constant status updates on systems that are performing normally. Instruments only demand attention when something is wrong. We applied the same logic to chronic disease management. MediCompass only surfaced a patient alert when a reading fell outside the bounds of their individual therapy plan. No alert meant the patient was on track. The clinic’s attention was reserved for the patients who actually needed it.

Patients had their own version of MediCompass, designed for home use rather than clinical workflow, where they could see their own trends and understand how their daily choices were affecting their numbers.

The technology worked. The clinical logic was sound. The business model was designed for scale.

All we needed was someone to pay for it.

Read more

Duke University Medical Center. June 1989. A Saturday morning I never forgot.

It was early. The cafeteria was half empty. I was sitting with a cup of coffee when Bob dropped into the chair across from me looking like he hadn’t slept.

Bob was a senior anesthesia resident at Duke. Brilliant, serious, the kind of physician who was going to be somebody. But that morning he looked beaten. I asked him what was wrong.

He explained his weekend ritual. Every Saturday, he had to manually calculate how much time every third-year anesthesia resident had spent in each surgical specialty over the past week. That meant pulling paper anesthesia records one by one. Finding the start time. Finding the stop time. Identifying which resident was on that case. Determining which specialty. Tallying the hours. Then doing it all again for every resident, every case, every specialty. For the entire academic year.

Hours. Every Saturday. All year long.

I listened. And then something clicked.

“Bob,” I said, “I think that information is already sitting in our database.”

I need to back up.

At the time I was working for a company called DIATEK. We built technology for anesthesiologists, including a product called ARKIVE: an anesthesia information management system I had helped architect and build. The concept was ahead of its time. Capture data automatically from OR monitors, ventilators, IV pumps, blood bypass machines. Everything in the operating room that generated a signal. Then build the anesthesia record from that data automatically.

No more hand-charting. No more transcription errors. The anesthesiologist’s hands stayed where they belonged. On the patient, not on a clipboard.

ARKIVE ran on a database engine called Paradox. Windows DOS environment. The year was 1989.

The data Bob needed was all there: case start times, stop times, resident assignments, specialty areas. Captured automatically by ARKIVE on every single case. Sitting in that database the whole time. Nobody had thought to ask it the question Bob was answering by hand every weekend.

I sat down with him right there in the cafeteria. We wrote a Paradox query together. It took maybe an hour.

It ran in about ten seconds.

Bob stared at the screen. Every resident. Every case. Every specialty. Every hour. Calculated perfectly. Instantly.

He looked at me.

“You gave me my weekends back. I truly owe you.”

He didn’t owe me anything.

But that hour, that ten-second query on a Saturday morning in 1989, taught me something I have spent the next 36 years trying to apply at scale.

The most valuable use of clinical data is almost never the use it was collected for.

The ARKIVE data was collected to build the anesthesia record. That was its purpose. That was what justified its existence to every administrator and department chair who approved the budget for it.

But sitting inside that same data, quietly, patiently, invisibly, was the answer to a question nobody had thought to ask. A question that was consuming hours of a brilliant physician’s weekend every single week.

The data was already there. It just needed someone to ask it the right question.

Thirty-six years later I keep watching the same story repeat itself at a larger scale.

We deploy an EMR. We spend $40 million. We collect more clinical data than any previous generation of physicians could have imagined. And then we use it to document the same things we used to document on paper, only faster.

We deploy BCMA. We scan every medication at the bedside. We collect a perfect digital record of every drug that reaches every patient. And then we use it to verify the right drug reached the right patient. Nothing more.

We deploy machine learning on our automated dispensing cabinets. We score every transaction against a behavioral baseline. We collect a perfect digital record of every controlled substance that moves through the facility. And then we use it to flag statistical outliers with no access to the clinical record. Seventy-seven percent false positive rate. Nursing leadership furious. Good nurses under investigation.

Each time, the data was already there. Each time, we only asked it one question. The question the system was designed to answer.

Nobody asked what else was in it.

Here is what I believe, after five decades inside the companies that built these systems:

The problem was never the technology. The EMR worked. BCMA worked. The ML diversion platform worked. Every system performed exactly as designed.

The problem was the architecture. Each system was optimized for a single data source. None of them could reason across the full picture simultaneously. None of them could read the clinical story as a single coherent narrative and ask: does this make sense?

The physician order. The pharmacist verification. The cabinet transaction. The pump record. The patient vitals. All of it together. In real time.

That question, does this make sense?, is what Bob needed answered on those Saturday mornings. The data to answer it was right there in ARKIVE. He just needed someone to write the query.

Read more