AI workslop: it looked finished

Start here

Tuesday, half past four, and it arrives on time.

The competitor analysis you asked for is in your inbox a day early. Fourteen pages, clean headings, a summary table, three recommendations at the end. It reads well. You skim it twice and you cannot find the thing you actually needed, which was whether the two of them price by seat or by usage.

So you go back to it properly. Forty minutes in you realise the summary table is describing three companies in general terms that would fit any three companies. The recommendations are true and would have been true before anybody did any work. Somewhere around six o'clock you open the two websites yourself and find the answer in eleven minutes.

Nothing about that evening gets recorded anywhere. It does not appear in a project report, it does not come up in a one-to-one, and the person who sent it has no idea it happened. From where they sit, they delivered a day early.

Receiving this poor quality work created a huge time waste and inconvenience for me. Since it was provided by my supervisor, I felt uncomfortable confronting her about its poor quality and requesting she redo it. So instead, I had to take on effort to do something that should have been her responsibility. A project manager, quoted in the BetterUp Labs and Stanford research1

That quote is from the study, and it contains the whole mechanism. The work moved. The cost moved with it. And the person carrying the cost decided it was not worth the conversation.

Out in the wild

What you notice first, second and third.

The order matters, because by the time you reach the third one the first two have already been normalised.

Someone at a desk reading through a document with their agent alongside — **First: things arrive earlier and read better** Documents get longer and tidier. Decks get more slides. Everybody notices this and everybody reasonably reads it as the tools working, because for a while it partly is.

Two colleagues at a table going through work together — **Then: your own evenings get longer** You are doing more reading and more rewriting than you were, and you cannot point at what changed. It comes in four-minute and forty-minute pieces, none of which is worth mentioning on its own.

A meeting where a decision is being talked through — **Last: you start checking who sent it before you open it** This is the expensive one, and it is the one nobody says out loud. Two or three names have quietly moved into a different category, and neither they nor you have ever had that conversation.

Workslop is work that looks done and moves the thinking onto somebody else. Forty in a hundred American desk workers received some in a month, and each incident took about two hours to sort out.1

The gist

BetterUp Labs and Stanford's Social Media Lab surveyed 1,150 full-time American desk workers in September 2025 and named the thing workslop.1
40% had received some in the previous month. Each incident took an average of about two hours to resolve.1
They put the cost at about 186 dollars per employee per month, roughly 170 euros, and more than nine million dollars a year for a company of ten thousand.1
The damage is not only time. Receiving it changes how people see the colleague who sent it, and almost nobody raises it.1
The sum for your own company is further down, with every assumption on screen and adjustable.

This lands if any of these is true

Work arrives faster and you are spending longer on it
Something came in that was long, tidy and impossible to act on
You have started checking the sender before you open the file
Your productivity numbers look fine and the week feels heavier

What you take away

The research, with the sample, the method and what is missing from it
A sum you can run on your own company in two minutes
Why the person sending it has no idea, and what that means for the fix
One rule that costs nothing and stops most of it

In plain terms

The work moved downstream, and it got bigger on the way.

The researchers define workslop as AI-generated content that looks good and lacks substance: it creates the illusion of progress while leaving colleagues to do the real thinking and the clean-up.1 That definition is doing a lot of work, and the important half is the second one.

Consider what actually happened in the example at the top. The sender spent twenty minutes and produced fourteen pages. The receiver spent an hour and then eleven minutes doing it properly. Total time went up. Total quality went down. And on every dashboard in the company, a deliverable arrived a day early.

Nobody saved any time. One person's twenty minutes became somebody else's evening, and only the twenty minutes got counted. Paul Musters

There is a second cost that does not appear in hours at all. The research is clear that receiving this changes how people see the person who sent it, and the person sending it almost never finds out.1 Read the quote from the project manager again. She did not confront her supervisor, she absorbed the work, and now she has a private opinion about her supervisor's output that will affect every handover between them for a year.

That is the part that makes this different from ordinary bad work. Ordinary bad work gets sent back. This gets absorbed, silently, by somebody who has decided the conversation is more expensive than the fix.

What the research says

Three numbers, and then what they mean at your size.

BetterUp Labs ran this with Stanford's Social Media Lab, surveying 1,150 full-time American desk workers in September 2025. The findings were published in Harvard Business Review, and the headline numbers are these.1

40%

of American desk workers received workslop in the previous month

1

2 hrs

average time to resolve each incident

1

$186

monthly cost per employee, roughly 170 euros, and over $9M a year for a company of ten thousand

1

Nine million dollars in a company of ten thousand is a headline that makes it into a lot of slides and means very little to anybody running a company of a hundred and fifty. So here is the same published per-employee figure, scaled down to sizes that might be yours.

The published figure of 186 dollars per employee per month, applied to four company sizes

50people$112k

150people$335k

300people$670k

500people$1.1m

Each figure is 186 dollars times twelve times the headcount, rounded. That assumes everybody in the company is a desk worker of the kind surveyed, which nobody's company is, so treat these as the top end of the range rather than a bill.1

The published per-employee cost, scaled to company sizes between 50 and 500 people.1

Even taking a third of those numbers, this is a line item somebody would notice if it appeared on a budget with a name on it. It never does, because it arrives as forty minutes here and an evening there, spread across people who each decided their own piece was too small to mention.

What the study does not tell you

Two honest limits. The sample is American desk workers, so the share and the hourly value both travel badly to a Dutch company of a hundred people. And BetterUp does not publish the full method behind the 186 dollars, so when you do the arithmetic yourself further down you will almost certainly land lower. Both are estimates built on different assumptions, and the one you build with your own inputs is the one you can defend in a meeting.

BetterUp is a coaching and development company, which is worth knowing when reading a study that concludes with the value of coaching and development. The definition and the survey design are solid and the framing is theirs.

The mechanism

The sender saved an hour. That hour did not disappear.

The sender genuinely believes it is good

This is the one people get wrong. Somebody who produces fourteen tidy pages in twenty minutes is not cutting a corner they can see. The output is longer and better written than what they would have made alone, so by their own standard it improved. They are missing the specific judgement about whether it answers the question, and that judgement is exactly what nobody has ever taught them to apply to a machine's output.

Everything visible rewards volume

Delivered early. More slides. Longer report. Every signal a company routinely collects says this went well. The cost lands on somebody else's calendar, in a form nobody logs, so the ledger shows a gain on one side and nothing on the other. Any system that measures output and not usefulness will produce more of this every quarter.

It went out early and it came back as somebody else's evening. Only one of those two things is in a report. Paul Musters

Saying something is socially expensive and the fix is cheap

Redoing it yourself takes an hour. Telling a colleague, or worse your own manager, that what they sent was hollow takes courage and costs you something in the relationship. Almost everybody takes the hour. The project manager quoted in the study took the hour.1 Multiply that decision by everybody in the building and you have a problem that is enormous in total and invisible in every individual case.

Is this a level problem?

Six questions, three minutes, nobody sees the answers. Where your company sits, what that costs, and what one level up changes.

Do the Culture Level scan

One thing to try

Do the sum with your own numbers.

Every assumption below is on screen and every one of them is yours to change. Two of the defaults come from the study and two of them are about your company, and only you can fill those in.

Desk workers in your company yours Share who receive one in a month study Hours to sort out each one study Loaded cost of an hour, in euros yours

people × share × hours × hourly cost, once a month, then twelve times over

Incidents a month60

Hours a month120

A year, at your numbers€93,600

One incident per affected person per month is the conservative reading of the study, and it is what this uses. BetterUp's own model produces a higher figure per employee than this arithmetic does, and they do not publish the method behind it, so the two are separate estimates rather than a check on each other. Yours is the one you can defend.

Run it once with the defaults and once with the honest version for your company, and take the lower of the two into any conversation. The number matters less than what it does in a room: it turns a vague complaint about quality into a figure somebody has to answer.

One thing worth doing alongside it. Ask three people how many hours they spent last month fixing something that arrived looking finished. Do it separately, and expect the first answer to be none, followed thirty seconds later by an example.

Level by level

The people producing it are not the people you would guess.

Every leadership team that runs this exercise arrives with a name in mind. That name is usually wrong, because this turns on a habit rather than on effort or seniority: whether somebody reads a machine’s output as a draft or as a result.

An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.

So the measuring runs per person: how somebody thinks and works, and how far along they are with AI. Somebody who produces workslop is usually somebody in the middle of the range, confident enough with the tools to move fast and without the habit yet of asking whether the thing answers the question. That is a coachable position and it is a very specific one.

It also changes what you do about it. A general message about quality lands on everybody and changes nobody. Knowing which four people are in that position lets you have four short conversations instead, and those four conversations are the whole intervention.

About emaho

emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.

See your own profile, free Find your AI Culture Level

Fifteen minutes per person. No credit card, no strings.

What to do

Two sentences, on everything you hand over.

There is one rule here and it costs nothing to introduce. Anything you send to a colleague carries two sentences at the top, written by you rather than by anything else.

The first says what you are asking them to do with it: decide, review, act, or simply know. The second says what you are least sure about. That is the whole rule, and it works because a person who has to write those two sentences has to have read their own document, and reading your own document is precisely the step that workslop skips.

Say out loud that using AI to produce it is fine and expected. The rule is about somebody taking responsibility for a thing before it leaves their desk, which is the difference between a draft and a deliverable and always was.

What changes once it is in place

The receiver gets a way to send something back without insulting anybody. “Your second sentence says you were unsure about the pricing, can you check that before I read the rest” is a normal, cheap sentence. Without the rule the only available move is to say the work was poor, and almost nobody makes that move. This is the real point of it.

And you get an early warning that costs nothing to collect. When somebody's second sentence is consistently empty, or consistently the same, that is the signal. It arrives long before anybody would have raised it in a review, and it points at exactly the coaching conversation worth having.

One thing to stop while you introduce it. Stop treating a long document as evidence of effort. Say in a leadership meeting that fourteen pages is now a neutral fact rather than a good sign, and mean it the next time somebody sends you fourteen pages.

What did your sum come out at?

If you run the calculation, send me the number and your headcount. I have a rough sense of the range for companies your size and I will tell you where you sit in it. I read these myself. One message back, no follow-up.

Message me on WhatsApp

The cost of waiting

Somebody downstream does the thinking that got skipped.

The measurable cost is the one you just calculated, and it is the least interesting of the three. It runs quietly, it is spread thinly enough that nobody raises it, and it is the version you can put on a slide.

Underneath sits trust between colleagues, and that one does not come back easily. Somebody who has absorbed three of these has adjusted their expectation of a person permanently, without ever telling them. The next time that person sends something genuinely good it gets read with a bit less attention, which is a strange and expensive way to lose the value of somebody's best work.

And there is the story it writes about AI inside your company. Every one of these incidents is somebody experiencing the tools as a thing that makes their week worse. Enough of them and the honest internal verdict becomes that this stuff creates more work than it saves, which is both wrong and, from where they are sitting, entirely reasonable.

The common ones

Workslop, asked and answered

What is workslop?

Workslop is AI-generated work that looks polished and lacks substance, so it creates the illusion of progress while leaving colleagues to do the real thinking and the clean-up. The term comes from research by BetterUp Labs with the Stanford Social Media Lab, published in Harvard Business Review in September 2025. Typical forms are slick slides, long reports, over-tightened summaries and code without context.

How common is AI workslop at work?

In a survey of 1,150 full-time U.S. desk workers conducted in September 2025, 40% said they had received workslop in the previous month. That is two in five people, in a single month, which is why most people recognise the description immediately even though almost nobody had a word for it.

How much does workslop cost a company?

The researchers put it at about 186 dollars per employee per month, roughly 170 euros, and more than nine million dollars a year for a company of 10,000 people. Each incident took an average of about two hours to resolve. Those figures are based on U.S. desk workers, so the share and the hourly value both travel imperfectly to a smaller European company.

How do you find out whether your company is producing workslop?

Look at whether there is a written standard for what finished means. emaho places companies on five AI Culture Levels, and a shared, written definition of good work is Level 3, Blueprint, where about one in ten organisations sits. Below that, quality depends on who happens to care, which is exactly the condition in which polished, hollow work travels unchallenged. The scan is six questions and three minutes.

Why is workslop worse than ordinary bad work?

Because it gets absorbed instead of sent back. Ordinary poor work is visibly poor, so somebody returns it. Workslop arrives looking finished, so the receiver quietly does the missing thinking themselves and says nothing. The research quotes a project manager who felt unable to challenge her own supervisor's output and took on the work instead.

Does the person sending workslop know they are doing it?

Usually not, and that matters for how you fix it. Somebody producing fourteen tidy pages in twenty minutes has made something longer and better written than they would have alone, so by their own standard the work improved. What is missing is the judgement about whether it answers the question, which is a habit almost nobody has been taught to apply to a machine's output.

How do you stop AI workslop without banning AI?

Require two sentences at the top of anything handed to a colleague, written by the sender rather than by a tool. What you are asking the reader to do with it, and what you are least sure about. Writing those two sentences forces the sender to read their own document, which is exactly the step workslop skips, and it gives the receiver a cheap way to send something back without insulting anybody.

How do you know which people are most likely to send hollow AI work?

It is rarely the name a leadership team guesses first. emaho measures one Operating Profile per person, personality type and AI level in a single profile, which locates the people who read a machine's output as a result rather than as a draft. That is usually somebody in the middle of the range, moving fast and confident, and it is a coachable position rather than a character flaw.

Does workslop damage relationships between colleagues?

The research finds that receiving it changes how people see the colleague who sent it, and that the sender almost never finds out. The practical effect is that somebody's future work gets read with less attention, including the work that is genuinely good, and the conversation that would fix it never happens because it feels socially expensive.

Why do productivity dashboards miss workslop?

Because they measure output and not usefulness. A document delivered a day early with more slides registers as a gain. The hour the receiver spent decoding it, and the eleven minutes they then spent doing the job properly, appear nowhere. Any system that counts one side of that exchange will produce more of it every quarter.

How do you calculate what workslop costs your own company?

Multiply your number of desk workers by the share receiving one in a month, by the hours each takes to resolve, by your loaded hourly cost, and then by twelve. Using the study's 40% and two hours with your own headcount and rate gives a figure you can defend in a meeting, and it will usually come out lower than the published per-employee number because the underlying method for that figure is not published.

What does emaho do about AI workslop?

emaho puts the checking step inside the way people already work. Everyone gets an Operating Profile, personality type and AI level in one profile, and on that we build a personal set of AI agents that fit how they work, with the review built in rather than hoped for. That turns a general message about quality into four specific conversations. The first profile is free, fifteen minutes per person, built for companies between 20 and 500 people.

Is workslop the reason AI shows no productivity gain?

It is one plausible mechanism among several. The same Harvard Business Review article notes MIT Media Lab research finding that 95% of organisations see no measurable return on generative AI investment, and offers workslop as part of the explanation: activity rises, output looks better, and the work simply relocates from one person's twenty minutes to another person's evening.

Who in a company tends to produce workslop?

Rarely the people a leadership team names first. It is usually somebody in the middle of the range: confident enough with the tools to move fast, and without the habit yet of reading a machine's output as a draft rather than a result. That is a coachable position and a specific one, which is why four short conversations beat a general message about quality.

What should a leader stop doing about this?

Stop treating length as evidence of effort. Say in a leadership meeting that a fourteen-page document is now a neutral fact rather than a good sign, and act on it the next time one arrives. As long as volume reads as diligence, the incentives keep pointing the wrong way regardless of what any policy says.

Where this goes next

What travels with it

Each one is a different way that work moves without the cost of it moving too.

12

No real impact

Everybody is faster and nothing moved. Where the individual gains go on their way to the company.

Read this one 11

Coordination cost

What happens to a review queue when the work arriving in it doubles and the reviewing does not.

Read this one 05

AI agrees with everyone

Why the thing that produced those fourteen pages told the sender they were on the right track.

Read this one

All 25 AI challenges, and what to do about each

Twenty-five things that break inside a company once people start using AI, each with the research behind it and the level where it starts to bite. This one starts to bite at Wild West, and 8 of the others start there too.

Open the full set

What now

The four people those two sentences are really for.

A message about quality reaches everybody and changes nobody. Fifteen minutes per person tells you which handful are in the position where this happens, and those are four short conversations rather than a policy.

Name the handful it is really forFifteen minutes each tells you which people send work that nobody can check quickly.
Ask for the reasoning, not the documentA page that cannot explain how it got there is the tell, and it takes one question to find.
Move the checking to where the work is madeChecking at the receiving end is already too late. Four people have read it by then.

Start with your own profile, free Ask me anything first Plan a 30-minute call

First profile free · no credit card · built for companies of 20 to 500 · you decide what your team gets to see

Not ready to put your team in anything yet? Start with the level of the company instead. The Culture Level scan is six questions, three minutes, and asks nothing of you.

Paul MustersFounder of emaho

Paul Musters

Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.

LinkedIn · paul@emaho.world · WhatsApp

Sources and numbers used on this page

Kate Niederhoffer, Gabriella Rosen Kellerman, Angela Lee, Alex Liebscher, Kristina Rapuano and Jeffrey T. Hancock, AI-Generated “Workslop” Is Destroying Productivity, Harvard Business Review, 22 September 2025, updated 25 September 2025. Research by BetterUp Labs with the Stanford Social Media Lab. Figures and the definition are taken from BetterUp Labs' own research page for the study, since the Harvard Business Review article is paywalled beyond its opening. Insights are based on an online survey of 1,150 full-time U.S. desk workers conducted in September 2025. Source of: the term workslop and its definition as AI-generated content that looks good but lacks substance, creating the illusion of progress while leaving colleagues to do the real thinking and clean-up; 40% of U.S. desk workers having received workslop in the previous month; an average of about two hours to resolve each incident, reported in the article as one hour and fifty-six minutes; a cost of 186 dollars per employee per month; more than nine million dollars a year for a company of 10,000 people; the finding that receiving workslop changes perceptions of the colleague who sent it; and the quotation from a project manager. A follow-up article by the same team, Why People Create AI “Workslop” and How to Stop It, appeared in Harvard Business Review in January 2026 and has not been read for this page. BetterUp sells coaching and development software. Percentages circulating in press coverage for how receiving workslop changed perceptions of colleagues could not be verified against a primary source and are therefore not used here.

Numbers are quoted as published, and the euro figures are approximate conversions. The Tuesday afternoon at the top and the three scenes are composites drawn from client situations rather than transcripts. The four company sizes are the published per-employee figure multiplied by headcount and twelve, which assumes a workforce made up entirely of desk workers.