Workers Lose Nearly a Full Workday a Week Fixing AI’s Mistakes

Workers save 11 hours a week using AI, then give back 6.4 of those hours fixing what it gets wrong, according to a new report from Glean, an enterprise AI search and workplace assistant company. Its research arm, the Work AI Institute, calls that cleanup work “botsitting,” and it now claws back more than half of what AI is supposed to save.

The lost hours are simple arithmetic. Harder to track is what happens to a worker’s judgment once verification stops entirely, a slide that already shows up in a federal discrimination lawsuit against Workday and in a fresh stack of cognitive science published this year.

Workers Save 11 Hours a Week on AI and Give Back 6.4

Glean surveyed 6,000 full-time digital workers across the United States, United Kingdom and Australia in December and January for its Work AI Index 2026. Eighty-seven percent said they already use AI on the job. Those workers reported saving 11 hours a week, then spending 6.4 of those hours on botsitting: feeding AI context, catching hallucinations and rewriting confident sounding but wrong answers. That is close to a full workday lost every week, the exact gap the report is built around.

Net time saved comes out well under five hours. Only 13% of workers say their organization is performing significantly better because of AI, despite adoption running at 87%.

Tool sprawl makes the math worse. Workers juggling multiple AI systems are 35% more likely to report frequent botsitting, and six in ten told Glean they rerun the same prompt across different tools because the first answer was not good enough. Glean calls that pattern the AI toggle tax. For every 10% more time employees spend feeding context into AI tools, they are 25% more likely to say the work leaves them worn out.

The Slide from Botsitting to Botshitting

Botsitting has an uglier companion, in Glean’s own vocabulary: “botshitting,” shipping AI-generated work that employees have not verified, do not fully understand or could not defend if asked. Sixty-nine percent of AI users admit to doing it. Heavy users, Generation Z workers, men and managers were the most likely of all, Glean found. Separately, close to 28% of AI users say they have blamed the technology itself for error-ridden work. Among heavy AI users, that figure climbs to 41%.

The report describes exactly how workers get there. “As the AI toggle tax increases, workers begin to cognitively offload,” the report’s authors wrote. “They hand more of their thinking and judgment over to the machine. They start to cut corners. They stop checking outputs, verifying sources, and asking whether the AI’s recommendations make any sense.”

First, workers stop fully understanding the output. Then they stop interrogating it. Eventually, they stop feeling responsible for it at all.

Glean calls that a slow surrender of agency, not a single bad decision. Frank Meltke, chief executive of the digital transformation consulting firm Contraco, pointed to the root cause in comments to CIO: “Workers are spending nearly a full day verifying AI output because nobody at deployment defined what verification was required.”

The Science Behind the Slow Surrender

Glean is not the only one documenting this. A January 2026 Wharton School paper by researchers Steven Shaw and Gideon Nave gave the pattern a name: cognitive surrender, defined as adopting AI outputs with minimal scrutiny. The pair extended Nobel laureate Daniel Kahneman’s model of fast and slow thinking with what they call a third system of artificial cognition, Forbes reported in May.

The skill loss shows up in high-stakes work, too. The International AI Safety Report 2026 found that three months after doctors started using AI assistance, their unaided ability to detect tumors had dropped 6%. A separate survey of 666 people found heavier AI use tracked with lower critical-thinking scores, an effect researchers tied directly to cognitive offloading, the same dynamic behind the old “Google effect” on memory.

Three Studies, One Number

Three research efforts, run months apart with no coordination between them, keep landing on close to the same warning.

The Stanford and BetterUp researchers coined the term “workslop” for AI-generated work that looks polished but falls apart under scrutiny. They found it costs about $186 a month per employee once cleanup time is priced against self-reported salaries, close to two hours of rework per incident. A June 2026 follow-up in Harvard Business Review, written by Oxford professor Matthias Holweg and Babson College professor Thomas Davenport, argued the damage compounds over time: as errors move between coworkers, an organization’s collective knowledge base deteriorates.

HR Runs the Rollout and Faces the Fallout

Glean broke its findings down by role, and HR stands out. Ninety percent of surveyed HR professionals said they already use AI at work, above the average for other departments. Most of that use sits at the lower-stakes end of the job.

• Writing and content – job postings, policy language and internal communications
• Administration and coordination – scheduling, correspondence and routine paperwork
• Hiring and screening – roughly a third of HR professionals say AI already shapes people decisions

Glean’s report goes further on that last point: “HR workers are more likely than the average employee to report that AI is already shaping consequential people decisions.”

The pressure to adopt has its own momentum behind it. Companies racing not to fall behind on AI have made the tools close to mandatory, a dynamic that lines up with Gallup polling showing tech employees who skip AI face triple the layoff risk of workers who use it. Glean’s own data shows what compliance costs: workers who spend 40% or more of their AI time botsitting are more likely than their peers to start job hunting.

Could Workday’s Lawsuit Reach 10,000 Other Employers?

Possibly. Workday faces a class action from job seekers who say its AI screening tools rejected them by age, race and disability, and a March 2026 ruling let key claims move forward. Legal analysts tracking the case say the exposure could extend well beyond Workday to the thousands of employers that run its hiring software.

The case, Mobley v. Workday, was filed in 2023 in the U.S. District Court for the Northern District of California by Derek Mobley, the lead plaintiff, who says he applied for dozens of jobs at companies using Workday’s tools and was rejected every time. A 2024 ruling from U.S. District Judge Rita Lin let claims proceed under a theory that Workday acted as an agent of the employers that deployed its software, opening AI vendors to direct discrimination liability for the first time in a case like this, according to law firm Seyfarth Shaw.

The case kept moving through 2026. In March, the court rejected Workday’s argument that the Age Discrimination in Employment Act (ADEA) only protects current employees, not rejected applicants. Then Workday won a narrower point: a June 2026 discovery ruling from Magistrate Judge Laurel Beeler found the company’s internal bias-testing results are shielded by attorney-client privilege, and blocked plaintiffs from forcing Workday to hand over its customers’ applicant data. A separate claim now moving against Eightfold AI, another hiring-software vendor accused of discriminatory screening, suggests the theory is not staying contained to one company.

What is settled so far:

• Workday can be sued directly as an agent of the employers using its hiring software, not just as a neutral tool vendor.
• Age discrimination protections extend to rejected job applicants, not only current employees.
• Workday’s internal bias-testing results are shielded from disclosure by attorney-client privilege.

What is still unresolved:

• Whether Workday or its customers are ultimately liable. No trial date has been set.
• How many of the more than 10,000 employers using Workday’s hiring tools carry similar exposure.

Glean’s own fix skips volume entirely. The report tells companies to build judgment and teach workers when not to use AI, rather than pushing more tokens through the pipeline. It also points to a small group Glean calls high AI achievers, workers who protect their own judgment, learn from the tool’s mistakes and know when to close the tab.

Separate research on AI’s effect outside the office found that effortless AI use can dull independent thinking, the same pattern Glean is now measuring at work. Most of Glean’s 6,000 respondents are still doing the opposite of that careful minority, trusting the output first and checking it later, if they check it at all.