Generative AI promised to supercharge productivity. Billions poured in. Yet many organizations see scant gains. Instead they confront a subtler threat. Low-quality AI output floods inboxes, shared drives and decision documents. Colleagues spend hours fixing what looks finished but falls short.
Researchers call this workslop. The term, coined by BetterUp Labs and Stanford’s Social Media Lab, describes AI-generated content that masquerades as good work but lacks the substance to advance a task. A September 2025 Harvard Business Review article laid it out in detail.
Survey data from 1,150 U.S. full-time workers backs the claim. Forty percent received workslop in the prior month. Recipients spent an average of one hour and 56 minutes sorting each case. The monthly cost per worker reached $186. Scale that to a 10,000-person company and the annual hit exceeds $9 million. These figures come straight from self-reported salaries and time estimates in the BetterUp-Stanford research.
But numbers tell only part of the story. Fifty-three percent of recipients felt annoyed. Forty-two percent judged the sender less trustworthy. Roughly half saw that colleague as less creative, capable or reliable. One in three grew less willing to collaborate with them again. Trust inside teams frays. So does the quality of collective knowledge.
From isolated errors to systemic atrophy
Matthias Holweg, Oxford professor of operations management, and Thomas Davenport of Babson College took the idea further in a June 2026 HBR piece. They described a feedback loop. Polished-yet-flawed AI documents enter workflows. Downstream teams verify, correct or redo them. Errors compound. Organizational memory thins. The authors labeled the broader outcome knowledge decay.
This differs from simple hallucinations. Hallucinations produce outright false facts. Knowledge decay describes what happens when mediocre output accumulates over months. Workers stop relying on internal files. Processes built on shaky information yield shaky results. Institutional expertise atrophies as employees lean on AI instead of building their own judgment.
Recent evidence sharpens the picture. A Nature article from June 18, 2026 reports that AI-driven deskilling has begun in medicine and computer science. Seventy percent of nurses and 77 percent of physicians worry about losing skills through over-reliance. Studies show performance drops when AI tools are taken away. One Polish endoscopy analysis found adenoma detection rates fell from 28.4 percent before AI assistance to 22.4 percent on days without the tool after its introduction.
Similar patterns appear in software engineering. Developers who lean heavily on code assistants show reduced ability to solve problems unaided. The phenomenon echoes older concerns about automation. Pilots lose manual flying skills when autopilots dominate. The same risk now faces knowledge workers.
A July 2025 MIT Media Lab report found 95 percent of organizations saw no measurable return on generative AI spending despite billions invested. Goldman Sachs reached a comparable verdict in March 2026. It detected no meaningful link between AI adoption and productivity gains at the economy-wide level. Seventy percent of S&P 500 executives still talked up AI on earnings calls. The gap between rhetoric and results widened.
Hiring offers a stark example. AI-generated resumes flood recruiters. AI-written job postings mislead applicants. Automated screening tools discard strong candidates. Trust in the entire process has sunk to all-time lows for both sides, according to Holweg and Davenport. The very systems meant to streamline talent acquisition instead degrade it.
Worker reactions have turned measurable. A 2026 survey of 2,400 employees across the U.S., U.K. and Europe found 29 percent actively sabotaging corporate AI strategies. They ignored guidelines, skipped training or skewed performance data. Among Gen Z respondents the share climbed to 44 percent, largely from fears of displacement. Tech companies announced more than 95,000 job cuts in 2026, nearly half attributed to AI. Analysts question how many of those roles were truly replaced by mature systems.
And the fix demands the very human effort AI was supposed to replace. Companies must now add layers of verification, quality standards and oversight. That work consumes employee time. It undercuts the original efficiency case. Indiscriminate mandates produce indiscriminate results. Public large language models applied to unsuitable tasks spit out generic prose laced with mistakes. Proprietary models trained on company data fare better when paired with clear guardrails.
Authors of the 2025 HBR piece, including Kate Niederhoffer of BetterUp, Gabriella Rosen Kellerman, Angela Lee, Alex Liebscher, Kristina Rapuano and Jeffrey T. Hancock of Stanford, offered practical advice. Avoid blanket orders to use AI everywhere. Model discernment. Develop explicit policies. Cultivate a “pilot” mindset of high agency and optimism rather than passive reliance. Frame AI as a collaborative tool with firm standards for output quality.
Examples from the survey illustrate the frustration. One finance professional received an AI-drafted analysis that forced a choice: rewrite it, demand revisions or accept mediocrity. “It is furthering the agenda of creating a mentally lazy, slow-thinking society that will become wholly dependent upon outside forces,” the respondent said.
A tech manager struggled to decipher an unclear AI-generated email. It took an hour or two to gather the team and restate the points clearly. A retail director spent extra hours verifying claims, scheduling meetings and redoing the work. Each case shifted burden downstream. Time supposedly saved evaporated. Friction grew.
These stories align with broader patterns. The Guardian reported in April 2026 that bosses tout productivity while workers drown in workslop. Charter Works examined how smarter, bounded AI use can curb the problem. A Springer Nature paper from late 2025 framed AI deskilling as a structural issue that extends beyond medicine into many fields.
Executives face a hard question. Individual tasks may finish faster. Yet the cumulative effect on organizational decision-making can turn negative without controls. Companies that froze hiring on expectations of AI gains now discover quality can slip faster than headcount shrinks. Klarna’s well-publicized pause offers one case among many.
Knowledge decay reframes the entire debate. The issue is no longer whether AI speeds up one worker’s output. It is whether widespread adoption improves or degrades the shared base of expertise that organizations depend on. Early data and expert analysis point toward degradation for those who adopted without discipline.
Leaders who treat AI as a simple labor substitute risk institutional atrophy. Those who treat it as a tool requiring human judgment, standards and occasional pushback stand a better chance of preserving what matters most: the ability to think, decide and innovate without constant digital crutches. The coming years will test which path more companies choose.
WebProNews is an iEntry Publication