Hospitals across the country are deploying artificial intelligence systems designed to streamline one of medicine's most time-consuming tasks: responding to patient portal messages. The tools automatically generate draft replies that doctors can review and send, promising to reclaim hours from clinicians' schedules.
Yet emerging research suggests these AI assistants may be backfiring. Rather than reducing workload, the systems are making some physicians worse at their jobs. Studies show that doctors spending time reviewing and editing AI-generated responses often take longer to complete messages than if they had simply written replies from scratch. The cognitive burden of evaluating machine-generated text, catching errors, and revising awkward phrasing consumes more time than straightforward composition.
The problem reflects a broader challenge with workplace AI adoption. When systems perform tasks at near-competent but imperfect levels, they can paradoxically increase worker burden instead of decreasing it. Clinicians cannot simply accept AI drafts wholesale, given the stakes involved in medical communication. Patient safety demands careful review. Yet the review process itself becomes the bottleneck.
Hospital administrators, seeking efficiency gains, have embraced these tools without fully accounting for the actual workflow impact. The friction between automation and quality control creates a catch-22. Faster adoption risks patient communication errors. Rigorous review negates time savings.
This tension extends beyond healthcare. Similar patterns emerge in legal document review, customer service, and knowledge work generally. AI tools that operate at 85 percent competence often create more work than they eliminate, because human oversight becomes mandatory rather than optional.
The lesson cuts against the prevailing narrative of AI adoption in institutions. Technology vendors and executives have promoted these systems as straightforward productivity multipliers. Real-world deployment reveals a more complex picture. Without careful job redesign and honest assessment of actual time costs, hospitals and other organizations risk investing in tools that sound efficient in theory but generate organizational drag in practice