I learned the term “workslop” from a sharp Harvard Business Review article by Kate Niederhoffer, Gabriella Rosen Kellerman, Angela Lee, Alex Liebscher, Kristina Rapuano and Jeffrey T. Hancock (“AI-Generated ‘Workslop’ Is Destroying Productivity,” September 2025). They describe it as the growing flood of AI-generated drafts, emails and documents that look like productivity but quietly erode it. Each version needs editing, alignment, and verification. This means that the time supposedly saved by AI ends up being spent cleaning up after it.
The GenAI Divide: high adoption, low transformation
In the article, they also refer to a major MIT study, The GenAI Divide: State of AI in Business 2025 by Aditya Challapally, Chris Pease, Ramesh Raskar and Pradyumna Chari. The report uncovers a remarkably similar pattern at the organisational level. Their data show that while over 80 per cent of companies have explored generative AI and nearly 40 per cent have deployed it, there is an issue. 95 per cent of enterprise pilots deliver zero measurable impact. The tools may look impressive, but they don’t integrate or learn.
MIT’s team calls this the GenAI Divide: the gulf between high adoption and low transformation. On one side, a small minority of firms is embedding adaptive, learning systems that evolve with use. On the other side, most organizations are stuck with static tools that never cross from pilot to production. What’s striking is how this divide mirrors the workslop dynamic that Niederhoffer and colleagues describe: enormous visible activity, yet minimal genuine progress.
The messy middle: where real AI work happens
It’s not that people aren’t using AI. In fact, Project NANDA’s interviews found that workers in 90 per cent of surveyed companies use ChatGPT or Claude for everyday tasks, even when their employers haven’t approved official subscriptions. MIT calls this the “shadow AI economy.” The real transformation, in other words, is already happening — but off the books, in the messy improvisation of day-to-day work.
So, what connects workslop and the GenAI Divide? Both point to the same structural failure: systems that don’t learn. The dividing line isn’t intelligence, as the MIT report notes, it’s memory, adaptability and context. People are productive with AI when it feels responsive and remembers what came before. Without that, AI becomes another layer of busywork, generating plausible-sounding content that humans must continually re-check and repair.
The result is a strange paradox: AI adoption is skyrocketing, but so is digital fatigue. Teams drown in drafts, summaries, and “smart” dashboards that still require manual judgment. The HBR authors warn that this flood of synthetic output risks turning knowledge work into administrative triage. Similarly, MIT researchers reach a similar conclusion from a different angle: the failure to build learning-capable systems keeps most organisations on the wrong side of the divide.
From slop to learning
If there’s a lesson here, it’s that the next phase of AI isn’t about flashier models or more automation — it’s about feedback. Tools must be able to learn from use, integrate with real workflows, and improve over time. Otherwise, we’ll keep mistaking noise for progress, and workslop for work.