The AI Productivity Trap Is Spreading Through Corporate America

NEW YORK — Generative AI is being integrated into daily workflows across corporate America, but the technology is producing a digital byproduct that threatens its promised efficiency gains. Known as “workslop,” the influx of poorly vetted, AI-generated communications and code has become a source of productivity loss and interpersonal friction inside major enterprises.

The term comes from the broader concept of “slop,” a word coined by technology developers and commentators as a counterpart to “spam” for unwanted email. It describes low-quality, AI-generated material increasingly cluttering search engines, social media feeds, and public websites. Inside the corporate firewall, that phenomenon becomes “workslop.”

A joint study by BetterUp Labs and Stanford University’s Social Media Lab surveyed 962 full-time desk workers in the United States. More than half of respondents, or 52.7%, admitted to sending workslop to colleagues, with the behavior appearing more often in organizations that encourage or mandate AI tools.

Workslop can look polished, professional, and well structured while lacking the accuracy, depth, or utility needed to be useful. Survey participants cited neatly formatted automated emails with broken hyperlinks and needlessly convoluted software code that complicates engineering systems.

The people receiving that material bear much of the correction burden. Thirty-eight percent of desk workers said they had received workslop and estimated that revising flawed documents and code bases cost them an average of 3.4 hours per month. In the previous year’s survey, 40% reported encountering workslop, but the average time spent resolving it was two hours.

For an individual employee, the lost time represents an estimated $186 per month. In an enterprise with 10,000 employees, correcting AI-generated errors can cost as much as $9 million a year in lost productivity.

The study also found a social cost. Employees who received workslop viewed the colleagues who sent it as less competent and less friendly. More than a third of recipients, 36%, said they actively wanted to avoid collaborating with those coworkers in the future, as uncritical AI deployment eroded workplace trust.

The findings arrive amid a wider dispute over the productivity returns of generative AI. Global consulting firms initially projected that the technology could add trillions of dollars in value to the economy, but deployment has encountered practical bottlenecks. A separate Upwork Research Institute study earlier this year found that 77% of employees using AI said the tools had increased their workloads and hampered productivity.

That friction is contributing to a retreat among technology executives who once led aggressive “AI-first” mandates. Shopify cofounder and CEO Tobias Lütke made AI use a “baseline expectation” across the e-commerce company last year. Employees were told they had to prove they could not complete a task with AI before requesting additional human resources or funding.

The Shopify directive formed part of a broader efficiency drive that included laying off roughly 20% of the company’s staff in May 2023 and introducing proprietary AI assistants such as Shopify Sidekick. Yet on *The Knowledge Project* podcast on Tuesday, Lütke criticized the way employees had used the technology, describing a culture of massive, unread communications and raw code produced without accountability for the final product.

“We call those ‘slop grenades’ that people toss at each other,” Lütke said. “That’s definitely a bad thing.”

He said unchecked AI use shifts work onto recipients. Instead of using AI to synthesize the key points in an email, employees were using it to draft long, unverified documents that consumed colleagues’ time. “You don’t really read it, and now it has to be reviewed by your colleagues, and they are like, ‘this doesn’t look right’,” Lütke said. “You’re just letting AI do the work for you.”

Duolingo has also begun retreating from an earlier AI-first push. In late 2023, the language-learning platform eliminated roughly 10% of its contractor workforce while moving aggressively toward generative AI, including OpenAI’s GPT-4 technology for conversational features and content creation.

CEO Luis von Ahn announced last year that Duolingo would adopt an “AI-first” model, assess staff performance according to AI utilization, and halt headcount growth unless teams could show that automation was not viable. By May, however, von Ahn acknowledged that the strategy had reached a ceiling.

In an interview with *Fast Company*, he said he had been overly optimistic about the technology’s capabilities because of its initial performance in writing demonstrations. He also noted that AI could not replicate human creativity when scaled to high volumes.

“We may need to write 1,000 different stories for people to learn a language, then you’ll find that 20% of the things were just pure slop,” von Ahn said. “Whenever we scale a lot [of] things with AI, we have to really be careful that slop doesn’t get through.”