For those who’ve ever used a web-based affected person portal to message your physician in the midst of the night time, you gained’t be shocked to study that responding to these messages takes an more and more huge chew out of clinicians’ workdays.
So lately, hospitals have begun adopting an AI device that may draft responses for them. The device was purported to make a time-consuming job go extra shortly and easily, mentioned Philip Barrison, an MD-PhD scholar on the College of Michigan Medical Faculty who research AI in healthcare.
As a substitute, the device has given docs and nurses a brand new to-do listing. First they should learn the AI-generated response and resolve if it “is definitely one thing that they suppose they might say,” Barrison mentioned. People are suggestible, and one thing and deciding whether or not you’ll have considered it by yourself is a cognitively complicated job.
Even when the message seems to be right, the clinician nonetheless must “edit it to the purpose the place they suppose it’s acceptable” to ship to a affected person, Barrison mentioned. The AI device introduces a completely new set of difficult judgment calls into what was a comparatively simple course of. Consequently, many clinicians have chosen to not use it in any respect.
They’re lucky to have the selection. Buoyed by expectations of price financial savings and skyrocketing productiveness, firms are more and more asking (and typically requiring) workers to make use of AI to make their work extra environment friendly. Meta, for instance, final yr instructed some employees to make use of AI to “go 5X quicker by eliminating the frictions that sluggish us down.” The CEO of Shopify instructed workers they’d have to show they “can not get what they need finished utilizing AI” earlier than the corporate would approve new hires. Some firms are even evaluating or rating workers based mostly on how a lot they use AI instruments.
Employees in some sectors have discovered main time financial savings from AI. However for others, the instruments simply change the work moderately than making it quicker. Employees is perhaps spending much less time writing affected person portal messages, for instance, however extra time enhancing the releases the AI device writes.
At finest, this mismatch between employer expectations and worker actuality could be an annoyance. In different circumstances, nonetheless, it may end up in employees being laid off for failing to fulfill unrealistic effectivity calls for. Some critics say the overzealous adoption of AI in high-stakes settings like healthcare even places individuals’s lives in danger. Now employees, unions, and specialists are more and more calling for guardrails to guard workers from inflated expectations round AI — and clients, college students, sufferers, and most of the people from errors that may occur when managers put AI adoption above all else.
The hidden prices of AI use
Firms are more and more presenting workers with a selection: Use AI to be extra productive or “you’re going to be automated out of a job,” mentioned Aiha Nguyen, director of the labor futures program on the analysis group Knowledge & Society.
However the results of AI on productiveness aren’t as simple as some CEOs have claimed. In a single 2025 examine, software program builders believed AI made them quicker, however actually they took 19 p.c longer to finish duties. (The researchers tried to repeat the experiment this yr however had bother recruiting builders who would comply with work with out AI.) And in a latest survey of 5,000 white-collar employees, 40 p.c of rank-and-file workers mentioned AI saved them no time in any respect.
Employees throughout closely AI-exposed fields level to hidden timesucks that include utilizing the expertise. Julie, an artwork trainer, wrote in a response to a Vox reader survey that her college’s directors routinely counsel utilizing AI for lesson-planning, emails, and progress report feedback. She’s tried AI-generated lesson plans, however they don’t account for the truth that children may match by an exercise at totally different speeds.
“First, I’m checking what AI suggests, then I’m enhancing them. Why add a step I can accomplish by myself?”
— Julie, an artwork trainer who wrote in response to a Vox reader survey
“First, I’m checking what AI suggests, then I’m enhancing them,” she mentioned. “Why add a step I can accomplish by myself?”
For an worker at an East Coast communications company, an inner AI device was supposed to hurry up the method of drafting press releases and different paperwork in regards to the pharmaceutical business.
“The purpose is, I feel, to have the ability to plug and chug into this machine and be capable of flip a whole lot of supplies round quite a bit faster than we already do,” mentioned the worker, who requested to stay nameless for concern of profession repercussions.
However when the worker tried to make use of it for primary analysis, it made too many errors. Double-checking its work erased any time financial savings. When the worker tried utilizing it for communications with shoppers, its people-pleasing tendencies turned an issue, because the device put a “bizarre glad spin” even on messages warning of dangerous information.
“A part of the explanation we take a human velocity to show issues round is as a result of there’s a lot nuance behind every part that we do,” the worker instructed me. “AI is simply not going to have the ability to catch it.”
It’s not simply that AI makes errors. With the arrival of agentic AI, employees are more and more being requested to edit and oversee the output of a number of AI instruments, a brand new form of work that may have sudden prices.
One latest examine of 1,488 employees throughout industries, for instance, discovered that extreme oversight of AI brokers may result in what the researchers known as “AI mind fry,” a form of cognitive fatigue. “Individuals described a ‘buzzing’ feeling or a psychological fog with problem focusing, slower decision-making, and complications,” the researchers wrote in Harvard Enterprise Evaluate. Mind fry was additionally related to an elevated variety of errors and an elevated need to give up one’s job.
The researchers additionally discovered that whereas utilizing one or two AI instruments elevated productiveness, including extra instruments produced diminishing returns, and after 4 instruments, productiveness truly declined.
What employees really need from AI
Regardless of such findings, firms proceed to strain workers to make use of AI, and to quote AI funding as a rationale for layoffs, whilst firms that attempt to hyperlink employees reductions to AI adoption are inclined to wrestle on the inventory market.
Some employees and organizations, nonetheless, are starting to push again. Nationwide Nurses United, the nation’s largest nurses’ union, has criticized using AI instruments in hospitals to estimate staffing wants or to suggest remedy protocols for sufferers.
There’s no assure that these instruments will consider a affected person’s particular person profile, together with underlying medical situations, the way in which human clinicians can, Cathy Kennedy, the union’s president, instructed me. AI is meant to “assist us do our work extra effectively, however on the finish of the day, it makes it much more burdensome,” she mentioned.
Hospitals want to guage, with nurses on the desk, whether or not AI instruments actually work as marketed, Kennedy mentioned. “We now have to cease — we’ve to return and actually see if that is actually doing what it must do,” she mentioned.
The identical is true throughout industries, Barrison, the healthcare researcher, instructed me. “Organizations have to be ready to say when, in the event that they have been looking for a return on funding, in the event that they have been looking for worth in a expertise — how do you outline what that worth is? And if there’s not worth there anymore, how do you flip it off?”
Some employees have discovered ways in which AI truly helps them do their work — simply not those administration anticipated. Julie, the artwork trainer, likes to make use of Claude to study extra about subjects she’s much less acquainted with, like kiln-firing ceramics.
In the meantime, researchers have discovered that AI can truly cut back worker burnout, if it’s used to finish duties workers discover burdensome. “All people in each job has an inventory of issues that they procrastinate on,” mentioned Julie Bedard, a managing director and associate at Boston Consulting Group who led the AI mind fry examine. “These are the locations I get, unsurprisingly, a whole lot of enthusiasm to strive AI with.”
However employers gained’t discover out what these burdensome duties are except they hearken to rank-and-file workers. “Employee requirements and employee rights ought to proceed to be on the coronary heart of all of this,” Nguyen mentioned, “moderately than simply focusing an excessive amount of on the AI.”
