Workslop. When AI Creates More Work Instead of Better Work
The most tiring AI document is not obviously bad. It is complete, polished, and empty.
Every heading is in the right place. Every sentence sounds reasonable. Nothing is wrong enough to reject immediately, so someone has to keep reading. By the end, the reader has spent ten minutes discovering that the document did not contain a decision.
Call it workslop. AI makes an output cheap for the person who sends it and expensive for everyone who has to inspect, interpret, correct, or quietly ignore it.
The sender saves ten minutes. The cost moves downstream.
Production got cheaper. Review did not
When the price of something falls, people use more of it. AI has sharply lowered the cost of producing emails, reports, plans, meeting summaries, and internal proposals. It has not lowered the cost of deciding whether those things are true, necessary, or useful.
This creates an asymmetry. One person can generate thirty messages in the time it once took to write three. Every recipient still has to open each one, work out what matters, and decide whether a response is required.
The organization looks more productive because more artifacts exist. The people inside it spend more time processing artifacts that should not have existed.
Workslop is therefore not simply bad writing. It is a coordination failure.
Three kinds of workslop
The first kind is informational. A report has the shape of an analysis but no fact or distinction that changes what anyone should do. It expands a small amount of material until the page looks finished.
The second kind is decisional. Someone asks AI which option is best, copies the recommendation, and forwards it as if the answer carried responsibility. The missing work is not generation. It is choosing which tradeoff the organization is willing to accept.
The third kind is communicative. One model writes an email. Another model summarizes it for the recipient. A third model drafts the reply. Two people remain formally in the conversation while most of their attention is spent supervising text generated for the other person's model.
Each artifact can pass a basic quality check. The damage appears in the total volume and in the gradual separation between sending words and owning what the words mean.
Plausibility is a weak quality standard
Obvious nonsense is easy to remove. Workslop survives because it is plausible.
The document uses the expected vocabulary. The recommendations are balanced. The conclusion asks for more alignment, clearer priorities, and better execution. Nobody can point to the one sentence that makes it unacceptable.
This is the same reason workslop is exhausting. The recipient has to supply the missing judgment. What is the actual claim? Which fact supports it? What decision is being requested? What happens if we do nothing?
A clean format can hide the absence of those answers for a surprisingly long time.
Restraint becomes part of the work
The first defense is to ask whether an artifact needs to be sent at all. Silence can be a contribution when the alternative is another document that transfers sorting work to other people.
The second is to use AI for compression. If a three-thousand-word draft contains three hundred words of value, the useful application of AI may be finding and preserving those words. Generation is already abundant. Expansion needs a reason.
The third is ownership. A person who sends a document should be able to state its central claim in one sentence, identify the evidence that matters, and name the decision the reader is expected to make. AI participation does not remove that obligation.
Over time, names become quality signals. Some senders earn the assumption that a document is worth opening because they have repeatedly respected the reader's time. That trust may become more valuable as polished output becomes easier to manufacture.
AI can help us make more things. The mature use of it will often look like making fewer things, choosing them more carefully, and standing behind what remains.