How Curely AI Is Putting Clinician Pajama Time to Rest

Pajama time, the documentation that follows clinicians home after the last patient leaves, is one of the most measurable contributors to burnout in modern medicine. The strongest recent evidence ties heavy after-hours electronic health record work to higher burnout, lower professional satisfaction, and even lower knowledge scores. Ambient AI documentation is now the most credible tool for reducing that burden, though the evidence shows its real value lies in restoring clinician wellbeing and attention rather than in dramatic time savings. At Curely we build documentation for the clinicians we serve directly, with particular attention to the low-resource settings, including much of Africa, where every reclaimed minute is also a minute returned to a waiting patient.

What pajama time actually costs

The term describes a specific failure. Care does not end when the patient leaves the room. It continues at home, at night, in the electronic health record (EHR).

Time studies have shown for years that physicians spend a large share of the workday inside the EHR, with substantial after-hours work that spills into evenings and weekends. One widely cited analysis found primary care physicians spending close to six hours of the workday in the EHR, including roughly ninety minutes after clinic hours.

The most rigorous recent signal comes from a 2024 survey of 9,731 United States family medicine residents conducted by the American Board of Family Medicine. Nearly a third reported three or more hours of after-hours EHR work per night. After adjustment, that group had higher odds of burnout, lower odds of professional satisfaction, and lower in-training examination scores. The reported odds ratios were 1.61 for burnout, 0.61 for professional satisfaction, and 1.28 for lower exam scores. This is strong evidence of association. As a cross-sectional survey it cannot prove that documentation causes burnout rather than the reverse, but the direction and consistency across studies are clear.

The burden is also unevenly distributed. The same work found that older residents, women, international medical graduates, and those underrepresented in medicine were more likely to carry heavy pajama time, consistent with earlier findings that women physicians receive higher volumes of patient messages and clinical tasks. The cost of documentation is not paid equally.

The evidence on ambient AI documentation, and its limits

Ambient AI scribes listen to the clinical conversation and draft the note, which the clinician then reviews and signs. The category has moved quickly from pilots to controlled trials. The honest reading of the evidence is encouraging on wellbeing and mixed on raw efficiency, and the distinction matters.

On burnout and clinician experience, the evidence is moderate to strong and consistent. A quality improvement study across six health systems found burnout among ambulatory clinicians fell from 51.9 percent to 38.8 percent after thirty days with an ambient scribe, alongside lower cognitive task load and less after-hours documentation. A pragmatic randomized trial at UW Health reported a clinically meaningful reduction in burnout and roughly thirty minutes of documentation time saved per provider per day. Across these studies, clinicians consistently report lower cognitive load and more attention for patients.

On objective time savings, the evidence is emerging and inconsistent, and we think it is important to say so plainly. The first randomized trial of two ambient scribes, conducted at UCLA, found one tool reduced time per note by about 41 seconds against 18 seconds in the control arm. A matched analysis at UChicago Medicine found an 8.5 percent reduction in total EHR time and a larger drop in note composition specifically. A separate time-motion study found a reduction of less than one minute and no change in visit length.

The takeaway is that ambient documentation reliably changes how documentation feels, and less reliably changes how long it takes. For a clinician those are not the same thing. The wellbeing gains appear real even in studies where the clock barely moves, which suggests the relief comes as much from offloading cognitive work as from saving minutes.

Why the stakes rise in under-resourced systems

In well-resourced systems, pajama time mostly steals evenings. In understaffed systems, including much of Africa but also rural clinics and safety-net hospitals in wealthier countries, the same minutes are subtracted from patients who are already waiting.

The most acute version of this is in the WHO African Region, which has about 1.55 physicians, nurses, and midwives per 1,000 people, well below the density of 4.45 considered necessary to deliver essential services and reach universal health coverage. The region carries close to a quarter of the global disease burden while accounting for a small fraction of the global health workforce, and the projected shortfall reaches 6.1 million health workers by 2030. A clinician seeing thirty or more patients in a day has no evening buffer to absorb administrative overflow. The same pressure, in a milder form, applies wherever clinicians are stretched thin. Documentation competes directly with the next person in the queue.

This changes the design problem. In a well staffed clinic, a scribe that saves a few minutes per note is a wellbeing tool. In an understaffed one, the same minutes can mean another patient seen, a shorter queue, or a clinician who finishes the day with enough left to return tomorrow. The catch is that the evidence base for ambient documentation comes almost entirely from high income, well-resourced settings. Its performance under constraint, across local languages, code-switching, accented speech, intermittent connectivity, and paper-based or hybrid records, is not yet established. That gap is a research priority, not a footnote, and we treat it as one.

How Curely approaches the documentation burden

At Curely AI, we believe the future of clinical documentation is not better typing. It is documentation that happens quietly in the background while clinicians focus entirely on patient care.

Our ambient documentation agent is one component of a broader clinical intelligence platform designed to reduce administrative burden across the entire patient journey. Rather than building a standalone AI scribe, we are building an integrated intelligence layer that supports clinicians before, during, and after every clinical encounter. Documentation is simply one part of that larger vision.

What makes our approach different is where we choose to build.

Most ambient documentation systems have been designed and validated in well-resourced healthcare environments with reliable connectivity, mature electronic health record infrastructure, and predominantly English-speaking clinical workflows. We are building for the realities of healthcare delivery that exist across much of Africa and other resource-constrained settings, where infrastructure is inconsistent, clinicians manage overwhelming patient volumes, consultations naturally shift between multiple languages, and digital records often coexist with paper-based workflows.

These realities fundamentally change how clinical AI should be designed. Documentation systems must remain reliable despite intermittent internet connectivity, understand multilingual conversations and code-switching, integrate with both digital and hybrid record systems, and reduce workload without disrupting existing clinical practice. Technology should adapt to healthcare environments, not require healthcare environments to adapt to technology.

Equally important, we believe clinicians must remain firmly in control of the medical record. Ambient AI should accelerate documentation, not replace clinical judgment. Every AI-generated note is intended to serve as a draft that the responsible clinician reviews, edits where necessary, and approves before it becomes part of the patient's permanent record. This human-in-the-loop approach reflects both good clinical governance and the strongest evidence available today. The published studies demonstrating improvements in clinician wellbeing evaluate AI-assisted documentation, not autonomous documentation.

Our commitment extends beyond building the technology. We believe healthcare AI should be evaluated with the same scientific rigor expected of any clinical innovation. While current evidence shows that ambient documentation can meaningfully reduce documentation burden and improve clinician experience in well-resourced settings, its effectiveness in constrained healthcare environments remains underexplored. Rather than assuming those findings generalize globally, we believe they should be validated where they matter most.

For that reason, we measure success using outcomes that directly affect clinicians and patients: reductions in after-hours documentation, lower cognitive workload, improved workflow efficiency, clinician satisfaction, and ultimately more time available for patient care. We are equally committed to reporting where our systems underperform as where they succeed. Trust in healthcare AI is earned through transparency, reproducible evidence, and continuous improvement, not marketing claims.

Our ambition is simple but significant. We want to build clinical intelligence that returns time to healthcare professionals, restores attention to the patient encounter, and enables every clinician, regardless of where they practice, to spend less of their life documenting care and more of it delivering care.

What we are watching

Three questions will decide whether ambient documentation earns its place in care, and they matter most where resources are tightest.

First, whether it holds up outside English and outside well-resourced clinics. Local-language and accented speech recognition remains the hardest technical problem in this category and the one most likely to determine real-world accuracy. Performance demonstrated on North American speech does not transfer automatically.

Second, whether time saved converts into care delivered. In wellbeing terms the case is increasingly made. In throughput and access terms, which matter most where clinicians are scarcest, it remains open and needs to be measured directly.

Third, whether the note stays safe. Faster documentation is only progress if accuracy, completeness, and clinician oversight hold under real clinical load.

Pajama time is a solvable problem. The evidence now points to a clear direction even as it cautions against overstatement. Our work is to make that direction real for the clinicians and patients who have the least slack in the system, and to keep our claims tied to what we can actually show.