A new term from Harvard Business Review, "workslop," names the feedback loop that enterprise AI adoption created: polished-looking output laced with errors, compounding across teams until the damage is impossible to ignore.
The pitch was simple. Flood your workflows with generative AI, watch productivity climb, and leave slower competitors behind. Hundreds of billions in enterprise spending followed that logic. Now the receipts are coming in, and they don't match the forecast.
Last September, Harvard Business Review published a piece by Oxford's Matthias Holweg and Babson's Thomas Davenport that named what a growing number of organizations were quietly experiencing: workslop. Not bad AI. Not hallucinated output that anyone would catch. Workslop is the subtler problem, content that looks polished, sounds confident, and contains just enough error or emptiness to corrupt whatever it touches. A summary that misses the point but reads like it didn't. A draft that sounds authoritative while quietly shifting the facts. Each piece passes inspection; the accumulation doesn't. BetterUp Labs and Stanford research put a number on the rework: 41% of workers say they've encountered it, and it costs close to two hours of correction per instance, adding up to more than $9 million a year in lost productivity for an organization of 10,000.
The deeper problem is structural. When AI-generated output flows into a company's shared documents, wikis, reports, and institutional records, it doesn't stay isolated. It becomes the source material for the next round of AI-assisted work. Errors don't get caught; they get cited. The knowledge base that took years to build starts degrading from the inside, and the people most likely to notice, the experienced employees who'd recognize something was off, are often the ones already being replaced or sidelined in the name of efficiency.
What makes this particularly uncomfortable for the companies running the AI-spend thesis is how thoroughly the macroeconomic data has refused to cooperate. A July 2025 MIT Media Lab study of more than 300 corporate AI initiatives found that 95% of organizations saw no measurable return on their generative AI investment. The researchers didn't find a technology problem. They found an organizational one: companies that couldn't integrate AI into their actual workflows, structures, and cultures. Buying the tools wasn't the same as using them well.
Goldman Sachs made a similar point from a different angle. Chief economist Jan Hatzius said in early 2026 that AI contributed "basically zero" to US GDP growth in 2025, and added that "there's been a lot of misreporting" of that figure. Of the 2.2% GDP growth recorded last year, only about 0.2 percentage points could be traced to AI investment, and even that number was depressed by the fact that the most critical hardware, chips from Taiwan and South Korea, boosted those countries' output, not America's. Goldman separately found no meaningful relationship between AI adoption and productivity at the economy-wide level, though it did identify a 30% productivity boost in two narrow, specific use cases. The gap between that finding and the general triumphalism around AI in the enterprise is considerable.
None of this means the technology doesn't work. It means the way most companies deployed it didn't.
What the resistance inside companies is actually telling you #
A 2026 survey by Writer and Workplace Intelligence of 2,400 knowledge workers across the US, UK, and Europe found that 29% of employees admit to actively sabotaging their employer's AI strategy. Among Gen Z workers, that figure climbs to 44%. The sabotage ranges from refusing to use mandated tools to deliberately generating low-quality output to make AI look ineffective. Thirty percent of those who admitted to it said the main driver was fear of losing their job.
You could dismiss this as insubordination, but that would be the wrong read. When nearly half of your youngest employees are working against your AI rollout, that's not a culture problem, it's a signal. Those workers can see what the MIT researchers confirmed: the tools as deployed aren't delivering what leadership promised, and the people being asked to trust the process have good reasons not to. The irony is that the workslop problem and the sabotage problem have the same root. Both are symptoms of AI adoption that was driven by pressure and narrative rather than by careful thinking about what these tools actually change and what they leave untouched.
The companies that will come out of this with their knowledge bases intact are not the ones that deployed fastest. They're the ones that set clear standards for what AI output requires before it enters shared systems, that treat their employees' skepticism as data rather than resistance to manage, and that understand the difference between automating a task and automating away the judgment that made the task worth doing. Frankly, most organizations haven't had that conversation yet. The ones that start it now will have a meaningful head start on the ones still waiting for the productivity gains to appear.
Also read: Two ex-OpenAI founders built a tool to measure how well AI models actually know who you are • Microsoft is rewriting the economics of enterprise AI and the bill shock is just getting started • Stanford's 2026 AI Index confirms the enterprise window is closing faster than most founders think