The output standard

AI is very good at making work that looks finished. A clean email, a confident summary, a tidy table. Sometimes it is finished. Sometimes the real work has only been moved, from the person who made it to the person who receives it. The output standard is four rules that stop that happening.

Workslop

In September 2025, researchers at BetterUp Labs and the Stanford Social Media Lab gave the problem a name in Harvard Business Review. They defined workslop as “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.” Then they asked 1,150 full-time US employees how often it reached them.

The cost doesn't disappear. It moves.

The sender saves a few minutes. The receiver has to work out what's wrong, check what's true, cut what's padding, and often go back to the sender to ask what they meant. The time AI saved is spent again, by someone else, usually with interest. And because the sender's numbers look better, nobody sees it happen.

This is the same mechanism as module 1, at the scale of one email. The first principle of the framework is fix the bottleneck, and don't flood it. In most organisations the bottleneck is a person or a small team who checks, approves or signs: legal review, the pricing lead, the clinician who signs off. Unchecked AI output lands on them first. When engineering teams adopted AI heavily, Faros AI found they merged 98% more pull requests and review time rose 91%. Faster work upstream, slower work at the step that limits everything.

That's why the output standard is a guardrail and not a style guide. It protects the bottleneck.

Four rules

Anything AI helped make meets four rules before it goes to someone else.

• Labelled: say where AI helped. Not as a confession, as information. The reader knows what to look at more closely, and nobody is surprised later. One line, the same line everyone uses.
• Owned: your name is on it, and you can explain it. If you can't say why a number is what it is, or why one option ranks above another, it isn't ready to send. “The AI did it” is never an answer.
• Short: no longer than the reader needs. AI makes length free for the writer and expensive for the reader. A summary longer than the thing it summarises is a cost, not a help.
• Checked: you read it and checked the facts that matter before sending. Not every word. The facts that would cause harm if they were wrong: numbers, names, dates, commitments, citations.

Here are six outputs from an ordinary week. Decide which rules each one breaks, then see who ends up paying for it.

Why people send it

Workslop isn't a junior-staff problem. In the HBR survey, 40% of it moved between peers, 18% went up to managers, and 16% came down from managers to their teams. It travels in every direction, and it comes from the same few pressures.

• Volume gets rewarded. If people are pushed to use AI, or praised for volume, checking is the step that gets squeezed. Lesson 4 deals with usage targets directly.
• Nobody said what done means. Without a standard, “it looks finished” becomes the test. The four rules replace it with one that the receiver would recognise.
• The cost is invisible to the sender. They never see the hour and 56 minutes. A short note on the friction list, “rework from unchecked drafts”, makes it visible without blaming anyone.

Make it a habit, not a rule

Standards stick when they're easy and visible. Give people one line to label their work, so nobody has to invent the wording. Have managers use it first, on their own documents. And use the weekly team meeting to raise problems with AI output, as work to fix, never as a person to blame.

The pilot measures this directly. Rework at the bottleneck, and at the step after it, is one of the ten measures, recorded from day 0. If rework rises, AI output is flooding the bottleneck, and the output standard isn't holding.