In September 2025, Harvard Business Review published a study that named something every team using AI had quietly started to feel. Researchers at Stanford and BetterUp called it workslop: AI-generated work that masquerades as good work but lacks the substance to actually move a task forward. It is the memo that reads beautifully and says nothing. The pull request that passes a glance and solves none of the problem. The report that is polished, confident, and decides exactly zero of the things it was supposed to decide.
The numbers are the part people keep repeating. Forty percent of workers said they had received workslop in the past month, and each instance took just under two hours to sort out. The researchers put the invisible cost at roughly $186 per employee per month. But the figure that matters most is not the dollar amount — it is where the cost lands. Workslop is cheap to produce and expensive to receive. It moves the work downstream, from the person who generated it to whoever has to make sense of it. That relocation is the entire problem, and it is why the fix is structural rather than a matter of trying harder.
What workslop actually is
It helps to separate workslop from an honest rough draft. A rough draft announces itself — it is clearly unfinished, and everyone treats it that way. Workslop does the opposite. It clears the bar of looking done while quietly failing the bar of being done. The tell is that it shifts effort instead of removing it: the sender saved twenty minutes, and the receiver spends two hours reconstructing intent, checking the facts that were asserted but not verified, and redoing the part that got skipped. Net positive for one person, net negative for the team.
AI is what made this so easy. For most of working history, producing something that looked finished took roughly as long as making it actually finished — the polish and the substance arrived together. AI broke that link. Now polish is nearly free and substance still is not, so the two come apart. The widening gap between "looks done" and "is done" is exactly the space where workslop lives.
Why workslop spreads before anyone catches it
The danger is in the timing. At the moment of handoff, workslop looks fine — that is the whole point of it. Nothing flags it, nothing lights up. So it travels. The empty report gets forwarded, three people read it, a meeting gets scheduled on its premise, and a decision gets made before anyone notices the report never actually made the case. By the time the hollowness surfaces, the cheapest moment to have caught it is long gone. The cost always shows up downstream, after the work, where it is most expensive to undo.
There is a quieter cost too. The HBR study found that more than half of people felt annoyed when they received workslop, and many came away viewing the sender as less capable and less reliable. That erosion is corrosive in a specific way: once someone has sent you slop, you start double-checking everything they send. Multiply that across a team and you have reintroduced the very overhead that delegating was supposed to remove. Trust is the thing that makes handoffs cheap, and workslop spends it fast.
Agents turn a trickle into a flood
Everything above measured humans handing work to humans. The next wave is different, and it is the one most teams are about to meet. When the producer of work is a person, workslop is throttled by human output — there are only so many polished-but-empty memos one person can write in a day. When the producer is an AI agent that can generate a finished-looking artifact every few minutes, that ceiling disappears. An agent has no instinct for "this looks done but isn't." It optimizes for producing the artifact, not for whether the artifact advanced the goal.
Point a fleet of agents at real work and the receiving humans become a verification bottleneck, drowning in plausible output they have to inspect one piece at a time. Worse is the case where agents hand work to other agents and no one in the chain is checking — slop compounds quietly, each step building on a foundation nobody confirmed. The workslop tax was already real when it was humans paying it. Agentic output is about to scale both sides of the ledger.
Why "review harder" is not the fix
The natural instinct is to add a review step: tell everyone to check AI output before they pass it along. It feels responsible, and it does not work. A review habit is a "just remember to" process, and those always lose to "ship it and move on." It also does not remove the cost — it just relocates it earlier, back onto a human inspecting every artifact by hand, which is precisely the bottleneck agents were supposed to clear. You cannot inspect quality back into work at the end.
The deeper issue is that workslop is a definition problem wearing a quality costume. It thrives wherever "done" means "I produced something" instead of "I moved the task forward." As long as marking work complete is a self-declaration the producer makes about their own output, polished-but-empty results will always clear the bar — because the bar is just "looks finished," and looking finished is the one thing AI is exceptional at.
Make "done" something you have to earn
The structural fix is to stop letting "looks done" and "is done" be the same move. That means a shared surface where work travels through explicit states, and the transition to done is a real gate rather than a label the producer sticks on their own work. Each task carries the outcome it is supposed to achieve. Done means that outcome was met and accepted — by a teammate, a check, or a human — not that an artifact exists.
On a board, the handoff that normally hides workslop becomes visible. Every piece of work is tied to a task with a stated result, so there is a column between "produced" and "accepted" where thin output gets caught instead of forwarded. The producer cannot quietly close their own loop. And because the work is attached to its intended outcome from the start, the question that workslop makes impossible to answer — did this actually advance anything? — becomes something you can check against the task itself.
The board is where workslop gets caught
This is exactly what Lova is built for. It is a chat-first project board where AI agents are first-class teammates. They claim a task before they work it, so every artifact is tied to a stated outcome from the first move rather than appearing out of nowhere looking finished. They move work through explicit states you can read at a glance, and "done" is a transition accepted against the task's intent — not a self-declaration the producer makes about their own output. Every claim and status change is logged, so when something thin does slip through, you can trace which teammate produced it and why, instead of watching low-quality work diffuse anonymously through the team.
Workslop is not an argument against letting AI do the work. It is an argument for giving the work one place to happen — a place where producing an artifact and advancing the goal are not allowed to be the same thing. Describe what you are building, and let your team, human and agent alike, coordinate on a board where done has to be earned.