The AI Productivity Story Has A Catch, Harvard Business Review Warns

Companies that bet most heavily on generative AI now face a feedback loop that quietly degrades their own work, a problem researchers call knowledge decay.

Harvard Flags Knowledge Decay

Harvard Business Review published two articles this month with a stark message. The tools meant to speed work along, the authors warn, are quietly dragging it down across teams and entire departments. Writing in June, Oxford's Matthias Holweg and Babson's Thomas Davenport describe a slow rot they call knowledge decay, where polished but empty output erodes the records a company trusts.

The trouble is not the familiar one of AI inventing facts. Researchers traced this deeper damage to workslop, a term coined in September 2025 by BetterUp Labs and Stanford's Social Media Lab for output that looks finished yet adds almost nothing.

Also Read: Bitcoin ETF Exodus Hits Record $6.35B, But Panic Selling May Be Cooling

Workslop Erodes Workplace Trust

A survey of 1,150 full-time workers found 41% had received such material in a single month, with each instance eating close to two hours of someone's time. Analysts pegged the hidden bill near $9 million a year for a company of 10,000 staff, before counting the damage to morale and trust. In the same study, 53% of recipients said the output annoyed them, while 42% judged the sender as less trustworthy than before.

About half walked away seeing that colleague as less capable, and roughly a third said they would avoid working with them again. Hiring has absorbed some of the sharpest blows. AI-written resumes swamp recruiters, automated job ads mislead applicants, and screening tools filter out strong candidates while trust on both sides slides to record lows.

AI Payoff Stays Elusive

The trust problem sits atop a strikingly thin payoff. A separate report from MIT's Media Lab showed 95% of organizations saw no measurable return on their AI spending, even after pouring in tens of billions of dollars.

Cleaning up the mess, the authors note, means bolting human checks onto AI output, the very labor the tools were sold as a way to remove. The warning is not a blanket case against the technology. Models trained on a company's own data can still earn their keep, the authors argue, while public chatbots aimed at the wrong jobs churn out generic prose laced with mistakes.

The reckoning lands after a year of mounting doubt. Workslop first surfaced in September 2025, and the newer work shifts the question from whether AI speeds a single task to whether its spread leaves a company sharper or duller at every decision that follows.

Read Next: JaredFromSubway Bot Loses $7.5M After Taking Its Own Bait