Stop Drowning Your Team in S***

We were taught that comprehensive means good. AI just turned that belief into a liability.

For most of my working history, a long, comprehensive document meant someone who was fastidious and doing a good job. Length was a reasonable proxy for effort, and effort stood in for quality. The shorthand worked because writing was expensive. Producing more meant you spent longer thinking about it, so more was a fair sign you cared.

AI took the bottom out of that assumption. Producing text is now free and instant. The shorthand is broken, and most leaders haven’t noticed they’re still rewarding it.

Length is output, not outcome

Here is the distinction that matters. Length is output. The outcome is whether the reader understood and acted correctly. When writing was costly, output tracked outcome closely enough to use as a proxy. Now output is free and infinite, so it tells you nothing about the outcome, and often signals the opposite: that the writer skipped the harder work of deciding what to leave out. Product people have preached “outcomes over outputs” for a decade, then let AI bury them in output and called it progress.

The problem is old. AI removed the brake.

None of this is new. Pascal apologised for it in 1657: “I have made this longer than usual because I have not had the time to make it shorter.” “Brevity is the soul of wit” is delivered by Polonius, the biggest windbag in Hamlet, which is the joke. The skill has always been the same: knowing what matters and including only that. Being able to tell the difference is what good sense looks like, and it always has been. What changed is the brake. Writing used to cost enough to force the cut, and AI removed the cost, so now nothing makes you choose. If the natural brake is gone, install a deliberate one: make “what is the shortest version that still lands?” a required step before anything ships.

This is hard to stop because the bad instinct feels like good work. Producing a thorough document gives you the warm sense of a job well done, and that feeling is the trap. It is fast, effortless thinking optimising for the sensation of output, not the result. Most instincts are poor long-term outcome drivers, and this is one of them. The fix is deliberate: name the feeling so people can catch it. “It feels thorough” is the tell, not the proof. Add a cut pass, or a reviewer who asks one question. What decision does this document need to drive?

What it costs you

The cost of getting this wrong is not lost minutes, it is lost understanding. When documents bloat, people stop finishing them, and decisions get approved on the strength of documents nobody read, including the person approving them. A Stanford and BetterUp study published in HBR last year named the symptom “workslop”: AI output that looks finished but lacks substance, shifting the work of completing it onto whoever receives it. Forty percent of workers had received it in the past month, losing nearly two hours to each instance. Half thought less of the colleague who sent it. Length does not just waste the reader’s time. It costs you their trust.

Make it concrete. A lengthy PRD probably means your key engineers did not read all of it, which is a real failure. Or they did, and wasted time they could have spent understanding the problem, which is a productivity loss at best. Either way you lose. Get to the point with the right amount of context and you hand that time back: time to build the right thing, and to drive the change you actually care about.

The obvious objection: can I not just have AI summarise the bloated document? No. Summarising lossy output does not recover what was missing, it hides the gaps under fluent prose. You cannot compress your way out of a document that never had the right information in it. Concision is a discipline you apply when writing, not a patch you apply when reading.

Fix the standard, starting with yourself

I learned this years before AI, as a project manager. My emails read like letters, and they got ignored, even internally. So I cut them down: a one-line point at the top, the action in plain bullets, nothing else. Follow-through went up immediately. The technique has formal names, BLUF and Barbara Minto’s Pyramid Principle, but the test is simpler. Put yourself in the reader’s chair, especially a busy one, and ask honestly: would I sift through all of this to find the gold, or would I just wait for a conversation to get the ask? If the answer is wait for the conversation, the document has already failed.

So the fix is not a tool, it is a standard, and it starts with you:

• Redefine good as outcome, not output. A short document that gets read and acted on beats a thorough one that does not, every time. Reward that, not page count.
• Model it. Your team copies what you produce, not what you say. Ship bloat because it is easy and scores well by the old measure, and you have licensed everyone below you to do the same. You are training the behaviour you are complaining about.
• Teach the discipline, and give it a house format so it sticks and survives you.

And none of this has to be a burden. The same AI that inflates your documents will keep them lean if you point it that way. Two habits cost almost nothing. First, read your own draft and watch yourself: if you start skimming, or scrolling faster than you can read, that is the bloat talking. Second, build a skill or prompt that challenges the draft, strips what does not earn its place, and holds you to the right amount of context, so every word pays for itself. A few extra steps, not a new job. A small change in process, a large change in outcome.

AI did not cause this problem. It removed the friction that used to hide it, and it will multiply whatever standard you hold. Hold the old one and it floods you with confident, unread bloat. Hold a better one and it makes your team clearer than they could have been before. The bottleneck has moved from producing the work to deciding what is worth producing. The skill now, and always, is knowing what to leave out. Only school and academia ever rewarded word count. Outcomes never did. Measure the outcome, not the output.

Oh, and the word in the title? I meant slop. Workslop, to be precise, the AI kind named in that Stanford study. What did you think it was?