AI makes documents cheap to produce. Reading, choosing, and maintaining them still costs human attention.
Creating a report takes minutes. Giving it someone's attention still takes time.
A colleague asks for a quick take on a proposal. Ten minutes later, you receive a beautifully structured analysis with an executive summary, a risk matrix, and recommendations for three possible futures. You asked for a take. You now have a reading assignment.
AI has made it wonderfully easy to produce a plausible document. That is a real improvement for anyone staring at a blank page. But the cost of a document does not end when someone finishes writing it. It moves to whoever has to read it, check it, respond to it, and decide whether it should exist at all.
The new bottleneck is not always creation. It is attention.
The Ease of Making More
Before AI, putting together a polished report took enough effort that people tended to ask whether a report was needed. That friction was not entirely bad. Now a few prompts can produce a background brief, a meeting recap, a slide outline, and a follow-up memo about the same subject.
None of these outputs is automatically useless. The trouble begins when producing each one feels free while its audience pays the price. A manager receives five thoughtful-looking summaries of the same issue. The team spends an hour working out which claims were checked, which recommendations differ, and which version actually reflects the current plan.
More material can make a decision slower when it introduces more things to reconcile.
Polish Is Not a Reason to Publish
AI can turn rough notes into persuasive prose before the underlying idea has been tested. A table of risks looks serious; it may still contain guesses. An elegant presentation can travel farther than the evidence behind it.
That is why a useful editing question comes before the prompt: who needs this, and what will they do after reading it?
If the answer is "choose between two options," give them the options, the trade-offs, and the evidence. If the answer is "find out what changed," show the change. If nobody can name the reader or the decision, generating twelve pages will not make the purpose clearer.
Sometimes a short message is the right format. Sometimes a report or presentation earns its length because people need to inspect the reasoning. The point is to choose the form from the reader's job, not from the number of formats a tool can generate.
The Uncomfortable Skill: Subtraction
When producing content is easy, editing becomes a more valuable contribution. Which detail changes the conclusion? Which paragraph repeats what everyone knows? Which caveat belongs beside the claim rather than buried at the end?
Subtraction does not mean hiding disagreement or stripping away necessary context. It means making the important parts easier to see. A one-page brief that names its uncertainty can be more useful than a twenty-page report that sounds certain about everything.
AI can help with this work too. Ask it to compare drafts, surface conflicting claims, or show where an argument relies on an assumption. Then verify the consequential points yourself. A system that can generate confident prose can also generate confident mistakes.
The best test for a new document is simple: will it help someone understand, decide, or act in a way the existing material cannot? If yes, make it and give it a clear audience. If no, the most productive output may be nothing at all.
We do not need to stop making things with AI. We need to stop treating every easy-to-make thing as another thing someone must read.