Ambient AI, also known as "AI Scribes", is not difficult to develop. In fact, many systems are built in under six months, because it simply involves prompting AI engines like ChatGPT or other natural language processors. You could even try it yourself using ChatGPT and see how simple it is to generate usable output.
But here's the problem: Ambient AI is a terrible solution—for you and for medicine. And as you will learn, Praxis is a more effective, and much faster, alternative.
Let’s take a step back and ask: “Why do we chart?” After all, our predecessors fifty years ago kept notes on 3x5 cards. Charting wasn’t a burden then, and they still practiced excellent medicine.
Today, we chart:
- To think clearly at the point of care and improve clinical outcomes
- To protect ourselves against potential malpractice claims
- To comply with government and third-party audits
- To receive reimbursement for our work
Can Ambient AI help with any of these?
Quite the opposite. Ambient AI may actually increase your risk in all four areas.
Initially, it seems to document the visit accurately. It captures your dialogue with the patient, and you can always review the output, right?
Let’s look more closely.
A patient comes to your clinic complaining of severe mid-epigastric pain.
-
“Doctor, I’m dying… my stomach is killing me.”
-
“Where is the pain?” you ask.
-
“Right here,” he says, pointing to the mid-epigastric area.
Your mind immediately considers: “Peptic ulcer? Gastritis? Reflux?”
-
“How long have you had it?”
-
“Three months. It’s getting worse. I lost my job. My wife left me and took the kids. I can’t sleep. I’m going insane. My whole-body hurts. My back hurts...” and goes on like this for another ten minutes while you listen, empathize and then perform a full physical exam.
The only positive finding is mild tenderness in the mid-epigastric region—no rebound, normal bowel sounds. You prescribe a proton pump inhibitor, advise him to avoid NSAIDs, quit smoking, and possibly see a psychologist. He improves and doesn’t return.
Six months later, he visits another doctor for the worsening back pain. An MRI then reveals metastases from prostate cancer. He tells the new doctor—and eventually his lawyer: “I told my previous doctor about the back pain…”
That previous doctor is you.
Far-fetched? Not at all. In court, attorneys often cite “res ipsa loquitur”—the thing speaks for itself. In this case, “the thing” is your Ambient AI-generated note, which captured and retained the mention of back pain. That’s all it takes. The AI transcribes everything—including casual remarks—but doesn't understand what is clinically significant. And if you fail to edit out something minor that later becomes important, you’re exposed to legal liability.
In fact, to protect against disputes, AI vendors will begin storing voice recordings indefinitely if they have not already. Why? Because when a physician claims “the patient never said that,” the company needs proof to support its application. Now, everything said during the visit—jokes, personal remarks, offhand comments—is part of a permanent legal record. And because the patient gave consent to be recorded, this will be the first thing the plaintiff’s attorney will request.
The implications are obvious.
Worse, editing Ambient AI notes creates new burdens. It demands that you review every word to ensure accuracy, undermining the very efficiency it promises.
And What You Don’t Say Can Hurt You
As every seasoned clinician knows: errors of omission are far more frequent and dangerous than errors of commission. What you fail to ask, neglect to document, or forget to follow up on often causes far more harm than the mistakes you may actively make.
Ambient AI compounds this risk by focusing on the patient’s words, not the physician’s reasoning. These systems only hear what is said aloud, and only from the patient’s perspective. Critical clinical impressions, differential diagnoses, or even brief internal judgments can easily be left out—because they were never spoken.
Then, as you quickly skim the autogenerated note—assembled from a partial, passive transcript—and simply overlook a subtle phrase or symptom that seems accurate on paper but completely misrepresents your intent, your note becomes a clinical and legal liability.
Of course, during a medical encounter, two conversations take place:
- The verbal exchange with the patient
- The silent clinical reasoning within your mind
Ambient AI captures only the first. It loses your thoughts, your differentials, your internal analysis—the very elements that should form the foundation of your note. When you later review the writeup, the patient is gone, and your thoughts may have faded. The record becomes a static summary instead of a dynamic thinking tool—like a pilot’s checklist—meant to prevent errors and guide treatment, which is exactly what Praxis does.
More importantly, Ambient AI doesn’t take action. It doesn’t write prescriptions, order labs, send letters, or communicate instructions. You still have to use your mind and, once again, your fingers, to do all that manually. Ambient AI stops at transcription of the symptoms. It will not prescribe or order anything for you, even if it suggest it, you still have to use your mind and your exhausted fingers to type and click manually.
Of crucial importance, Ambient AI doesn’t help you think at the point of care. It’s not a projection of your mind. In the previous vignette, had you believed the back pain was clinically relevant, you would not only charted it, you would have addressed it! And if you didn’t think it was relevant, you wouldn’t have charted it. That simple!
“This in turn allows the defense attorney to use the record to assert the uniqueness of the patient, and why this therapy was chosen for this patient. Since a physician is not held to be a guarantor of a cure of a good result, but instead must choose an acceptable treatment based on the information available to the physician at the time the choice was made, clear documentation of what was known is usually an adequate defense….
Ambient AI doesn't think. It finds patterns and mimics them. Can we trust it not to make subtle errors? If it mistakes a dog for a cookie, the flaw is obvious. But what if it leads to a missed cancer diagnosis?
A modern EHR should protect and empower physicians—not dehumanize or constrain them.
Think of medicine like flying a commercial aircraft. Most of the time, autopilot is in control. But when things go wrong, the pilot must make quick, subtle decisions. The same applies to physicians. If we rely on automation for everything, we risk catastrophic mistakes.
Enter the Concept Processor. Like Ambient AI, it generates your notes instantly—but with several critical differences. Most importantly, it doesn’t simply transcribe a conversation; it captures your clinical reasoning as it creates your entire progress note—not just the history. The content originates from your own thought process, not from the patient or generic phrases that may not reflect your personal approach. The Praxis Concept Processor evolves with each case you see, continually learning from you. It is a comprehensive tool that ensures nothing is forgotten or overlooked, capturing your thoughts, words, and logic in your own language. This makes the review process fast, efficient, and intuitive. By aligning with your reasoning, Praxis anticipates your next steps—such as sending prescriptions, lab orders, instructions, portal messages, intake forms, and referral letters—and completes them on your behalf with your approval. And since it’s following your own logic, it’s hard to disagree with yourself. Instantly, the entire note is not only generated but also executed. You’re already seeing your next patient.
Having said this, we are currently developing a groundbreaking ambient feature—distinctly different from conventional ambient AI. Rather than transcribing your conversation, this intelligent tool listens, suggests differential diagnoses, and provides relevant background information in real time. Most importantly, it never replaces your clinical reasoning. Instead, it complements your thought process, reinforcing your decisions and gently reminding you of anything that may be missing—all while your note is being generated instantly based on your own medical logic. The syntax, the order of symptoms, and the emphasis placed on each element are entirely driven by your own clinical reasoning. The scribe simply highlights the key aspects in your own writing to streamline your charting—accelerating your workflow without interfering with your medical judgment. Then, the Concept Processor completes the entire clinical note automatically. It goes further—generating your prescriptions, ordering labs and other services, providing patient instructions, composing admission orders, guiding your staff and third parties, drafting letters to consulting and referring providers, and producing your superbills—all seamlessly with your note. None of this is generated through passive “ambient” transcription—it originates entirely from your own clinical reasoning.
“Good documentation is critical to properly defend a physician against a malpractice claim. It’s not just about recording what was done, but also showing the logical progression that led to a diagnosis or treatment…
..Praxis requires the physician to document their reasoning and refine it with every new case. By focusing on the differences between patients, the record reveals why a particular diagnosis or therapy was chosen over another…
…While the process by which a physician using PRAXIS to enter data is discoverable, that process is little different than what is now done without PRAXIS. While several other systems provide for such things as accurate and legible recording of notes and prescriptions, it is the flexibility and theory behind PRAXIS that will, in the long run, provide the best legal protection for the practicing physician."