will AI replace respiratory therapists?

Your hands are in the work in a way AI can't touch. You're placing endotracheal tubes, adjusting ventilator settings in real time, and performing manual chest physiotherapy on patients who can't breathe on their own. Those aren't tasks you can hand off to software. When a patient's oxygen saturation drops at 3am, you're the one physically in the room making the call.

Out of 22 tasks analysed using O*NET task data, every single one came back at 0% AI penetration. That's not a rounding error. Tasks like emergency resuscitation, arterial blood gas interpretation, and real-time physiological monitoring require you to read a patient's colour, their effort to breathe, the sound of their lungs. No model running on a server somewhere can do that.

You also work inside a clinical team where your judgment carries weight. You're relaying blood gas results to physicians, flagging adverse reactions, and adjusting treatment when a patient's response isn't what the protocol predicted. That's judgment built from training and patient contact. It's the kind of thing that takes years to develop and that a machine can't replicate by reading discharge summaries.

Respiratory care is almost entirely physical and real-time. AI tools that work well in other healthcare settings, like documentation assistants that transcribe clinic notes, have very little to do in an ICU where your hands are busy with a ventilator circuit and your eyes are on the monitor. The work doesn't happen at a keyboard.

There's also a liability gap that matters here. AI tools can suggest ventilator settings based on a patient's weight and diagnosis, but if those settings cause barotrauma, a human clinician signed off on the care. Hospitals aren't in a rush to let algorithmic recommendations run unsupervised on patients who are already critically ill. The accountability sits with you, and the legal and clinical systems are built around that.

And then there's the physical inspection problem. You're checking equipment for water accumulation in tubing, testing for leaks, listening for circuit faults. A camera-based system can't do that reliably in a clinical environment with variable lighting, movement, and humidity. AI can't smell a burnt motor or feel a tube that's kinked inside a patient. That's not a temporary limitation waiting for the next model update. That's a structural constraint.

Let's be honest about what AI actually does in respiratory therapy today: not much. There's no tool in this field with meaningful penetration into core clinical tasks. That's what a 0% exposure score means in practice.

Where AI does show up, it's in the periphery. Sepsis and early deterioration alert systems like those built into Epic and Cerner flag patients whose vitals suggest worsening respiratory status. You'll see these alerts in some hospitals already. They don't replace your assessment. They just surface a name on a list sooner. Nuance DAX Copilot can help physicians draft clinical notes faster, and some respiratory therapists in larger systems use similar ambient documentation tools for their own charting. But the core assessment, treatment, and equipment management? That remains entirely yours.

There's also some AI-assisted analysis appearing in pulmonary function testing software. Tools built into systems like ndd EasyOne and CareFusion's spirometry platforms can flag patterns in lung function data. But you're still the one running the test, coaching the patient through the manoeuvre, and deciding whether the result is valid or if the patient's effort was suboptimal. The software gives you a number. You decide what it means for this patient.

Your day hasn't changed much structurally. You're still doing rounds, responding to ventilator alarms, drawing arterial blood gases, and adjusting therapy based on what you find. The sequence hasn't been reshuffled by AI the way it has in, say, radiology or pathology.

What has shifted slightly is the administrative layer. In hospitals using integrated EHR systems, some charting is faster because flowsheet data pulls automatically from ventilators and monitors. You spend a little less time transcribing numbers and a little more time on the patient. But that's the result of device integration, not AI specifically, and it's been happening for a decade.

What hasn't changed at all is the direct patient contact time. Emergency response, ventilator management, patient education on inhaler technique, weaning protocols: these take the same amount of your time they always did. If anything, the 12.1% projected growth rate suggests you'll be doing more of it, not less, as the population ages and demand for pulmonary and critical care services goes up.

The BLS projects respiratory therapy to grow at 12.1% through 2034, adding roughly 8,800 openings a year against a current workforce of 139,600. That's faster than most healthcare occupations and well above the 4% average for all jobs. The growth is demand-driven, not a sign that AI is being used to stretch a shrinking headcount.

The demand comes from two directions. First, the US population is ageing, and older adults have higher rates of COPD, asthma, sleep apnoea, and respiratory complications from heart disease. More patients, more need for therapists. Second, the COVID-19 pandemic exposed how thin respiratory therapy staffing was in most hospital systems. Hospitals have been rebuilding that capacity since 2021, and that rebuilding is still ongoing in many regions.

The 0% AI exposure score here isn't just a curiosity. It tells you that AI isn't a factor in the supply-demand calculation at all. Demand for respiratory therapists is rising, the pipeline of new graduates doesn't fully cover projected openings according to the American Association for Respiratory Care, and there's no AI substitute on the horizon. That's a genuinely strong labour market position.

The 0% exposure score is unlikely to move much in the next five years. The bottleneck isn't data or computation. It's physical presence, real-time clinical judgment, and hands-on equipment management. For AI to make serious inroads here, it would need to arrive in the form of robotics capable of placing airways, adjusting physical equipment, and responding to unpredictable patient deterioration. That's not close.

In 10 years, you'll probably see more AI-assisted decision support in ventilator management, where algorithms suggest weaning protocols or flag patients ready for extubation. Some of that is already in early trials. But 'suggests a protocol' is very different from 'replaces the therapist running the protocol.' The human accountability layer isn't going away, and hospitals have strong legal and clinical reasons to keep it. Your exposure score is more likely to stay near zero than to climb.

The clearest thing to do is double down on the high-complexity end of the work. Critical care respiratory therapy, neonatal and paediatric respiratory care, and sleep disorder management are areas where the clinical judgment required is highest and the irreplaceability is strongest. If you're currently working primarily in general medical-surgical settings, getting your Adult Critical Care Specialist (ACCS) or Neonatal/Pediatric Specialist (NPS) credential through the NBRC puts you in the part of the field that's hardest to displace and pays better.

Patient education is worth investing in deliberately. Teaching patients with COPD or asthma how to manage their condition at home, how to use inhalers correctly, and how to recognise early deterioration is a task with essentially no AI competition. It requires trust, communication, and the ability to read whether a patient actually understood what you said. Building that skill set makes you more useful in outpatient and pulmonary rehab settings, which are growing alongside the inpatient side.

On the documentation tools covered earlier: learn them, use them, don't fear them. Anything that gets the charting done faster gives you more time at the bedside. That's where your value is. The therapists who'll do best in this field over the next decade are the ones who let the administrative layer get lighter while they focus harder on the clinical work that only they can do.