David Solomon, CEO of Goldman Sachs, was reviewing a report generated by an AI engine — a report his commodities business had produced. His description of it, delivered at The Economic Club of Washington, is the most precise account of a specific and growing problem that anyone has offered: the first few sentences, he said, made you think wow, that’s really insightful. And then you go down below that, and it is all garbage.
That gap — dazzling surface, hollow substance — has a name. Researchers at Stanford’s Social Media Lab and BetterUp Labs coined it in a Harvard Business Review article: workslop. And it is quietly becoming one of the most expensive and corrosive side effects of enterprise AI adoption, in ways that go well beyond the wasted time. This piece is about what workslop is, what it actually costs, and why the deepest damage is not financial at all.
What workslop actually is
The researchers define workslop as AI-generated work content that masquerades as good work but lacks the substance to meaningfully advance a given task. It has structure, polish, and fluency. It reads well. And it collapses under scrutiny — an essay that makes no original claim, a report that sounds convincing but offers no real point of view, prose that circles an idea rather than developing one. Stanford’s Jeff Hancock describes the style as a kind of purple prose: elaborate and expressive on the surface, ultimately hollow.
Solomon’s description is the phenomenon exactly. The first few sentences dazzle because AI is extraordinarily good at producing the surface features of insight — confident framing, sophisticated vocabulary, the cadence of expertise. The garbage below is the absence of the thing those surface features are supposed to signal: actual analysis, real judgment, substance that advances the work. Workslop is the surface of competence detached from the competence itself.
What it actually costs
The numbers are larger than most leaders realize, because the cost is hidden. The Stanford-BetterUp survey of 1,150 US full-time workers found that roughly 40% had received workslop in the previous month. Each incident took an average of one hour and 56 minutes to resolve — to interpret, correct, or redo. Translated into money, the researchers estimated a cost of about $186 per employee per month, which for a 10,000-person company works out to roughly $9 million a year.
The mechanism of the cost is what makes workslop so insidious. It does not save work; it transfers work. The person who generates workslop saves their own time by offloading the real cognitive labor onto the recipient, who must now do the thinking the sender skipped — interpret the polished emptiness, discover the substance is missing, and either redo it or send it back. As the researchers put it, workslop shifts the burden downstream. The apparent productivity of the sender is purchased with the hidden, larger unproductivity of everyone downstream of them.
This is why workslop can make an organization measurably busier and no more productive — which connects directly to the broader finding that 95% of organizations see no measurable return on generative AI. Some meaningful portion of that missing return is workslop: AI producing output that looks like productivity while actually generating downstream cleanup that consumes the time it appeared to save.
The deeper cost: trust
The financial cost is real, but the researchers found something more damaging, and it is the part every leader should weigh most heavily. Workslop corrodes trust between colleagues.
In the survey, among workers who received workslop, 42% viewed the sender as less trustworthy. Around half judged the colleague as less creative, less capable, and less reliable than they had before. A third said they were less likely to want to work with that person again. A single piece of polished AI emptiness, sent to a colleague, measurably degrades how that colleague perceives your competence and character.
This is the cost that compounds. An organization runs on trust between its people — the confidence that when a colleague sends you something, it represents real thought. Workslop erodes that confidence transaction by transaction. Once someone has been workslopped by you, they begin to scrutinize everything you send, which slows every future exchange. And as this spreads across a team, the collective knowledge base itself deteriorates: people trust each other’s output less, verify more, and the frictionless collaboration that made the organization fast grinds down. The HBR follow-up work describes exactly this — errors and hollow output compounding across teams until the organization’s shared knowledge decays.
Why AI produces workslop specifically
Workslop is not a random failure. It is the predictable product of a specific mismatch. AI is optimized to produce fluent, plausible, well-structured output — the surface features of good work. It is far less reliably able to produce the substance those features are meant to signal: genuine analysis, real judgment, a point of view earned through understanding. When a person uses AI to generate work they do not then do the hard part of verifying and deepening, the result is structurally guaranteed to be workslop: maximum surface, minimum substance.
This means workslop is not really an AI problem. It is a human-process problem that AI enables at scale. The AI produces the polished draft; the workslop is created when a human passes it on as finished work without supplying the substance. The tool makes the surface free. The substance was always the person’s job, and workslop is what happens when the person, seduced by the free surface, skips it.
Mapped to the Mutation Readiness framework
The workslop problem maps onto three dimensions of the Mutation Readiness diagnostic — the operational instrument of the Mutation transformation practice we run for enterprise leaders.
Signal Sensitivity — workslop is precisely an attack on signal-reading: it manufactures the signal of quality (polish, fluency, structure) while removing the underlying quality. Solomon’s ability to read past the dazzling first sentences to the garbage below is Signal Sensitivity in action. An organization with high Signal Sensitivity trains its people to detect the gap between surface and substance; one without it accepts the surface and pays the downstream cost. In the AI era, the ability to tell real substance from manufactured surface is becoming a core professional skill.
Ethical Guardrails — framed as containment-as-velocity. Workslop is what velocity without containment looks like at the level of daily work: AI lets people ship polished output fast, and without the guardrail of substance-verification, that speed produces reputational and knowledge-base damage. The guardrail here is a norm — AI-assisted output remains the sender’s responsibility, and passing on unverified workslop is a failure, not a shortcut. Containment-as-velocity means shipping fast without shipping slop.
Narrative Coherence — the deepest damage of workslop is to the trust that lets an organization act in concert. As workslop erodes colleagues’ confidence in each other’s output, it degrades the shared foundation Narrative Coherence depends on. An organization whose people no longer trust each other’s work cannot move together. Protecting against workslop is, at the deepest level, protecting the trust that makes coherent collective action possible.
The signals your organization is missing right now
The master signal is the gap between how much AI-generated output your organization is producing and how much of it actually advances the work versus generating downstream cleanup. Because workslop looks like productivity, most organizations are measuring the output and missing the cleanup — counting the busier as more productive.
Look for the specific tells. Are your people spending increasing time interpreting, correcting, or redoing colleagues’ AI-assisted work? Has trust between team members quietly declined as more output became AI-generated? Are your reports and documents getting longer and more polished while somehow advancing decisions less? Do your best people increasingly feel they are carrying invisible cleanup labor for others’ AI output? Each is a workslop signal, and each is invisible on a dashboard that counts output rather than substance.
Three practical questions
One: how much time are your people spending cleaning up workslop? The Stanford-BetterUp data says nearly two hours per incident, 40% of workers hit monthly, roughly $9 million a year for a 10,000-person company. Measure your own version. Ask your people directly how much AI-generated work they receive that they have to redo. The number will be larger than your dashboards show.
Two: is AI-assisted output treated as the sender’s responsibility in your organization? The single most effective guardrail the researchers identify is a clear norm: AI can assist, but the output remains each person’s responsibility, and passing on unverified slop is a failure. If your culture treats AI output as automatically acceptable because it looks polished, you are manufacturing workslop by policy.
Three: is workslop quietly eroding trust between your best people? The deepest cost is not the time — it is that 42% trust the sender less and a third don’t want to work with them again. Ask whether your collaboration is getting more guarded, whether people are verifying each other more. If trust is eroding, workslop is corroding the foundation your organization runs on, and no productivity metric will show it until the damage is done.
The closing thought
Solomon’s report dazzled for a few sentences and then dissolved into garbage. That is the signature experience of the AI era’s most underestimated problem. AI has made the surface of competence free — the confident opening, the polished structure, the cadence of insight — while leaving the substance as scarce and as effortful as it ever was. Workslop is what happens when people ship the free surface as if it were the scarce substance.
The cost is real and larger than it looks — roughly $9 million a year for a large company in cleanup alone. But the deeper cost is the trust between your people, which erodes every time a colleague opens a polished document and discovers there is nothing inside. That trust is the actual operating system of your organization, and workslop is quietly corrupting it, transaction by transaction, while the output metrics glow green.
The defense is not to abandon AI. It is to insist, as a matter of culture and guardrail, that the surface is never the deliverable — that AI-assisted work carries the sender’s full responsibility for the substance beneath it. The organizations that hold that line will get AI’s speed without AI’s slop. The ones that don’t will get busier, spend more, trust each other less, and wonder why the productivity never came.
The world has changed. The leaders who notice will be the ones the next decade is built around.
