Workslop Is a Tax Organizations Levy on Themselves

AI-generated junk work is now common enough to have a name. When an employee pastes unverified AI output into a deliverable and sends it on, the recipient inherits the job of figuring out what is wrong with it. Researchers at Stanford’s Social Media Lab and BetterUp Labs call this workslop, and their survey of 1,150 U.S. employees found that 41 percent had received some in the past month¹.

The instinct in most executive teams is to read this as a discipline problem: people being careless with a new tool. The survey data points somewhere less comfortable. More than half of respondents admitted to sending work they knew was subpar, and the reason they gave was workload pressure combined with vague instructions to use AI more. They were complying with an incentive someone above them created.

A definition that explains the behavior

Workslop is cost-shifting. The sender saves 30 minutes by skipping verification; the receiver loses two hours discovering the output is wrong. Total productivity falls while the sender’s numbers rise.

Each individual act is rational. The aggregate is a tax on the organization, collected disproportionately from its most conscientious people—the ones who do the checking. Any system that rewards visible output and ignores verification cost will produce this pattern, with or without AI. The technology lowered the price of plausible-looking work to nearly zero, which made an old dynamic suddenly visible.

The hours are the smaller cost

Wasted time is recoverable. Two other costs compound.

The first is trust. Once someone catches a colleague sending unverified output, they begin re-checking everything from that person. In the survey, recipients rated workslop senders as less capable and less reliable, and roughly a third said they became less willing to work with the sender at all. Verification overhead spreads through the org chart until collaboration runs slower than it did before the tools arrived.

The second is signal. When every document looks polished, it gets harder to tell who is thinking and who is forwarding. A company where all prose is fluent is a company where writing quality no longer carries information about talent—at the moment promotion and staffing decisions need it most.

How the mandate loop runs

The sequence repeats across companies. The board asks about AI strategy. The CEO answers with an adoption mandate. Middle managers, given no definition of good AI work, enforce the only thing they can measure, which is usage.

Overloaded employees comply in the cheapest available way: paste, generate, send. Eighteen months later the board asks why the productivity gains have not shown up. The missing step was the first one—deciding where AI creates value in this specific business, and what quality looks like there. A mandate issued without that work is a request for theater. The same gap between adoption pressure and work design shows up in the research on why AI adoption stalls: pushing the tool while leaving the workflow unchanged moves the cost onto the people inside it.

Four interventions, and where each gets hard

The diagnosis suggests four places to intervene. None is free.

1. Replace usage metrics with outcome metrics.
   Cycle time on named workflows, error rates, customer-facing quality. The trade-off: outcomes are slower to move and harder to attribute than adoption numbers, which is precisely why boards prefer adoption numbers.
2. Make senders own their output.
   One norm, stated plainly: if you send it, you vouch for it. The friction shows up in enforcement—a manager has to be willing to bounce polished-looking work back, which is uncomfortable in exactly the overloaded teams where workslop thrives.
3. Define the lanes per function.
   A short memo listing where AI is expected (first drafts, summarization, research sweeps) and where a human signs (anything customer-facing, performance reviews, decisions about people). The hard part is keeping it current; a lanes memo that lags the tools by two quarters reads as out of touch and gets ignored.
4. Learn from the people already doing it well.
   Every company has a few employees whose AI use is verified and woven into working processes. Building the playbook from their practice transfers judgment, not just mechanics—though it requires admitting the org chart does not know where its own expertise sits.

• • •

There is a version of this story in which leadership itself is the largest sender. A manager who runs an employee’s self-evaluation through a chatbot, or a CEO whose all-hands memo the team can smell, sets the effective standard regardless of what the policy says. In the survey, the angriest accounts came from people whose own work had been fed to AI without their knowledge; several described it as violating².

The pattern worth holding onto is that workslop is information. Each instance marks a place where someone was asked to demonstrate AI use without being told what good looks like, or was too stretched to verify their own output. Organizations that treat it as a behavior to punish will play whack-a-mole. The ones that treat it as a reading on their own incentive design will find it tells them, with some precision, where the design is broken.