Corporate leaders are pointing to AI-driven productivity metrics as proof the technology is working. Their employees, increasingly, are not buying it.
There is a word circulating across Reddit threads and X timelines this month that captures something the quarterly earnings calls do not: workslop. It refers to the growing volume of AI-generated content, code, documentation, and email that workers must now review, correct, and clean up before it becomes usable. The term is blunt, but the frustration behind it is real, and it is spreading fast enough that enterprise software vendors should probably be paying attention.
The immediate trigger was a pair of high-profile executive appearances in early April. Microsoft CEO Satya Nadella and Salesforce CEO Marc Benioff both used their respective earnings calls to tout the productivity gains their AI tools are delivering. Nadella leaned on a specific number: customers using GitHub Copilot are generating code 55% faster. The figure was presented as a landmark, a sign that the industry's bet on generative AI is paying off. Benioff made similar noises about Salesforce's Agentforce platform. Both men left their calls sounding bullish.
Then, on April 10th, a leaked internal survey from a Fortune 500 tech company landed on Reddit and reframed the entire conversation. The document showed that since the firm rolled out AI tools, the volume of internal documentation and code output had tripled. That sounds like exactly the kind of result executives celebrate. The problem is that 68% of employees surveyed reported a decrease in the perceived quality of their work, alongside a sharp increase in the hours they spent reviewing and fixing AI-generated errors. Triple the output. Lower confidence. More hours spent on correction. That is not a productivity gain by any conventional definition.
What this survey captures, and what the workslop discourse is gesturing at, is a fundamental shift in where the labor actually goes. Generative AI has dramatically lowered the cost of producing a first draft, whether that is a block of code, a project brief, or a client email. What it has not done is lower the cost of producing a good one. The burden has moved from generation to curation, and curation, it turns out, is not free. It requires judgment, domain expertise, and time. When AI tools flood an organization with plausible-looking output, someone still has to verify it. That someone is usually the same employee who was already doing the underlying work.
This is being called the Productivity Paradox 2.0, and the comparison to the original paradox of the 1980s and 1990s is instructive. Computers were supposed to make office workers dramatically more productive. For years, the data did not show it. The gains eventually materialized, but only after organizations fundamentally rethought how work was structured around the new technology. The AI cycle appears to be following a similar arc, with the added wrinkle that the tools are generative, meaning they actively produce content rather than simply processing or storing it. The potential for noise is orders of magnitude higher.
What the market is watching
The enterprise AI sector is projected to exceed $300 billion in value by 2027, a figure that assumes sustained adoption well beyond the current wave of pilot programs. That assumption is now under pressure. If workers continue to associate AI tools with additional cognitive load rather than relief from it, adoption could plateau at precisely the stage where vendors need it to deepen. Burnout and skepticism are not abstract concerns; they translate directly into churn, underutilization, and the kind of quiet rejection that does not show up in a CEO's earnings remarks but does show up in renewal rates.
The vendors who move first on outcome quality, rather than output volume, will have a real competitive advantage here. That means building tools that help users verify AI-generated work more efficiently, flag low-confidence outputs, and integrate seamlessly with existing quality control workflows. Measuring success by lines of code generated or documents produced is a metric that serves the vendor's narrative. Measuring it by whether the engineer shipped something that worked, or whether the analyst trusted the summary enough to act on it, is a metric that serves the user.
The executives celebrating 55% faster code generation are not wrong that something significant is happening. They may just be measuring the wrong thing. Workers will tell you the same, if anyone asks them directly.
Also read: Anti-AI saboteurs are attacking data centers with incendiary devices as Europe's energy crisis hands them an unexpected opening • OpenAI's $852 billion valuation is starting to look like a bet the market is no longer sure it wants to make • A Texas man has been charged with attempted murder after throwing a Molotov cocktail at OpenAI CEO Sam Altman's San Francisco home in an attack driven by fears about artificial intelligence
