Artificial intelligence was sold as a turbo‑charger for productivity, but new evidence shows a sizable chunk of those gains is being clawed back by a hidden tax: the hours workers spend correcting, validating, and re‑doing AI output. Together, recent studies from Workday, Stanford University’s Social Media Lab, BetterUp Labs, MIT, and Harvard Business School sketch a consistent picture of early disruption, mounting “rework,” and a growing need to redesign work and skills—not just add more tools.[1][2][3][4][5]
AI’s productivity promise
Workday’s new report, “Beyond Productivity: Measuring the Real Value of AI,” surveys 3,200 full‑time employees at companies with at least $100 million in revenue across North America, Europe, and Asia. As Workday summarizes, “Nearly 40% of AI time savings are lost to rework, including correcting errors, rewriting content, and verifying outputs from one-size-fits-all AI tools.” Only “14% of employees consistently get clear, positive net outcomes from AI,” a finding that directly challenges the narrative of effortless efficiency.[3][6]
Yet enthusiasm for AI remains high among heavy users. Workday notes that “employees who use AI every day are overwhelmingly optimistic – more than 90% believe it will help them succeed.” The catch is that those same power users “carry the biggest burden: 77% review AI-generated work just as carefully as work done by humans, if not more,” effectively paying back part of their time savings in vigilance.[2][6]
The growing rework burden
The centerpiece of Workday’s findings is what it explicitly calls a rework drain on productivity. In one summary, the company states that “nearly 40% of AI time savings are lost to rework,” which includes “correcting errors, rewriting content, and verifying outputs.” In practice, that means time saved on drafting or analysis is immediately spent validating references, aligning style, and fixing hallucinations.[2][3]
Younger workers are on the front line of this phenomenon. Workday highlights that “employees aged 25–34 make up nearly half (46%) of those dealing with the most AI rework,” despite often being considered the most tech‑savvy cohort. These employees “spend the most time checking and fixing AI output,” turning them into de facto quality‑assurance layers for generative tools rather than beneficiaries of pure automation.[3]
Workslop: polished but empty output
The Workday data echoes a second stream of research from Stanford’s Social Media Lab and BetterUp that introduces a new term into the corporate lexicon: “workslop.” BetterUp defines AI workslop as “unhelpful, low-quality AI-generated content” that “is quietly draining productivity and trust at work.” Stanford professor Jeff Hancock explains that the difference between sloppy work and workslop is that the latter “doesn’t require any effort to create, while sloppy work still requires a little bit of effort,” making it easy “to generate a lot of useless or unproductive content very easily.”[4][7]
The costs are far from theoretical. An analysis reported by Entrepreneur notes that “the average annual cost of workslop for a 10,000-person organization is about $9 million per year.” The underlying survey, covering 1,150 U.S. desk workers, found that employees spent “around two hours” resolving each incident of workslop, producing “an invisible tax… about $186 per month” per worker. As the report puts it bluntly, “rather than saving time, it leaves colleagues to do the real thinking and clean-up.”[8][4]
Training and transformation gaps
Workday’s research argues that the rework problem is not simply a technology flaw but a transformation failure. Its analysis finds that “89% of organizations have updated fewer than half of roles for AI,” even though leaders “prioritize training (66%)” on paper. Among those carrying the heaviest rework burden, “only 37%… report receiving skills training,” suggesting a large mismatch between AI expectations and the support employees actually receive.[9][3]
The report also links better outcomes to deliberate capability building. In organizations where AI is viewed positively, Workday notes that “reinvesting saved time into skills and redesigned work reduces rework and improves outcomes.” Workers who report clear benefits from AI are significantly more likely to say they used freed‑up hours for “higher‑value work” and to have received “increased access to training,” pointing to learning investment as a key differentiator.[9][3]
The J‑curve of AI adoption
The productivity paradox identified in offices mirrors patterns observed in factories and research labs. A recent MIT‑backed study using U.S. Census Bureau data describes a “J-curve” trajectory in manufacturing: “AI adoption in manufacturing leads to early productivity setbacks before delivering long-term growth gains.” The researchers found that “AI adoption tends to hinder productivity in the short term,” with firms seeing an average decline of 1.33 percentage points, and in some adjusted scenarios “short-run negative impact” as high as “around 60 percentage points.”[5][10]
Crucially, the MIT team emphasizes that organizational design mediates this effect. Older firms with legacy systems “actually saw declines in the use of structured management practices after adopting AI,” a change that “accounted for nearly one-third of their productivity losses.” By contrast, “younger firms that had already integrated digital tools or data infrastructure showed fewer short-term losses and rebounded faster,” suggesting that AI becomes an accelerant only when layered onto robust processes rather than chaotic workflows.[10]
Why workflows—not tools—must change
Harvard Business School research dovetails with these findings, arguing that the real value of AI depends on redesigning how work gets done. While specific studies vary, a recurring theme in this body of work is that successful AI implementation “requires redesigning workflows, not just deploying tools,” with leaders needing to rethink decision rights, oversight, and the division of labor between humans and algorithms.[11][5]
Taken together, these strands of evidence point to a clear conclusion: the rework burden is not an inevitable side‑effect of AI but a symptom of partial transformation. Workday’s data on role design and training, Stanford and BetterUp’s concept of workslop, and MIT’s J‑curve all converge on the same message: without intentional changes to roles, skills, and workflows, AI will continue to generate impressive‑looking output that others must quietly fix.[7][4][5][2]
Sources [1] New Workday Research: Companies Are Leaving AI Gains ... https://newsroom.workday.com/2026-01-14-New-Workday-Research-Companies-Are-Leaving-AI-Gains-on-the-Table [2] New Workday Research: Companies Are Leaving AI Gains ... https://www.prnewswire.com/news-releases/new-workday-research-companies-are-leaving-ai-gains-on-the-table-302660517.html [3] Workday study: 40% of AI time gains lost to rework https://www.stocktitan.net/news/WDAY/new-workday-research-companies-are-leaving-ai-gains-on-the-slg4teud05rg.html [4] AI Workslop Is a $9 Million Issue: Stanford, BetterUp Study https://www.entrepreneur.com/business-news/ai-workslop-is-a-9-million-issue-stanford-betterup-study/497483 [5] The 'productivity paradox' of AI adoption in manufacturing firms https://mitsloan.mit.edu/ideas-made-to-matter/productivity-paradox-ai-adoption-manufacturing-firms [6] New Workday Research: Companies Are Leaving AI Gains ... https://www.barchart.com/story/news/37034540/new-workday-research-companies-are-leaving-ai-gains-on-the-table [7] What is AI workslop? Research on costs and solutions https://www.betterup.com/blog/hidden-costs-workslop [8] How AI-Generated 'Workslop' Is Costing Companies https://finance.yahoo.com/news/9-million-problem-ai-generated-223107291.html [9] Learning investment emerges as AI's key differentiator https://learningnews.com/news/learning-news/2026/ai-speed-outpaces-skills-as-rework-erodes-gains [10] MIT Study Shows Drop in Productivity for U.S. Manufacturers After AI ... https://theaiinsider.tech/2025/07/23/mit-study-shows-drop-in-productivity-for-u-s-manufacturers-after-ai-adoption-followed-by-long-term-gains/ [11] MIT report: 95% of generative AI pilots at companies are failing https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/ [12] Employees waste 1.5 weeks a year 'correcting' AI output https://www.personneltoday.com/hr/employees-correcting-ai-workday/ [13] Company Announcements https://markets.ft.com/data/announce/detail?dockey=600-202601140530PR_NEWS_USPRX____LA62950-1 [14] 40% of AI-generated content is 'workslop', and it drives over ... https://www.unleash.ai/artificial-intelligence/40-of-ai-generated-content-is-workslop-and-it-drives-over-9-million-in-lost-productivity-annually-finds-betterup/ [15] Beyond Productivity: How Leaders Can Drive Real ROI ... https://blog.workday.com/en-gb/beyond-productivity-drive-real-roi-ai.html