The most valuable button in the AI era may be “I have read this.”
Not “Generate.” Not “Summarize.” Not even the increasingly theatrical “Deep Research.” The important act is the small, unfashionable moment when a human inspects the machine's output and accepts responsibility for putting it in front of another human.
This sounds like etiquette. It is actually infrastructure.
Generation Got Cheap. Attention Did Not.
A useful principle reached the Hacker News front page this week: if you are asking for human attention, demonstrate human effort.
The author describes receiving an AI-generated critique from a teammate who admitted they had not read it. That disclaimer is honest, but it also reveals the absurdity: I did not spend my attention on this, so please spend yours.
Marvelous. We have automated delegation all the way down to the recipient.
AI has made producing plausible text, code, reports, and investigations dramatically cheaper. It has not made reviewing them free. In many cases, it has made review more expensive because polished output can hide subtle nonsense behind excellent punctuation.
The scarce resource was never words. It was trusted attention.
Reading Is the New Proof of Work
Old proof-of-work systems made a sender spend a costly resource before the network accepted a message. Human collaboration now needs a softer version of the same idea.
Before asking someone to review an AI-assisted artifact, the sender should spend enough effort to prove three things:
- I understand what I am sending.
- I removed the obvious waste.
- I will own the consequences.
This does not require pretending the AI was absent. Quite the opposite. Labeling AI assistance is useful context. But “AI generated” must not become a diplomatic pouch that allows unchecked material to cross organizational borders without inspection.
The minimum proof is simple: read it, test it when possible, add your judgment, and state what decision or feedback you need.
The Inbox Already Learned This Lesson
Email providers have spent years building defenses around a related problem: sending became cheap, so recipients were buried in messages. Gmail now requires bulk senders to authenticate their domains, keep spam rates low, and support easy unsubscribing. These mechanisms do not prove that an email is wise. They prove that someone identifiable is accountable for sending it.
AI-assisted workplaces need the same accountability layer, applied to ideas rather than domains.
A useful review request should include:
- the human's short conclusion,
- the important uncertainties,
- the exact question being asked,
- and a clear indication of what was checked.
That is not bureaucratic ceremony. It is compression performed by the person with the most context.
Without it, organizations create an attention denial-of-service attack against themselves. Every employee can generate ten pages in seconds, then quietly transfer the verification cost to everyone else. Productivity dashboards glow. Nothing moves.
Do Not Ban the Machine. Charge the Sender.
The wrong response is to ban AI-generated material from collaboration. Good AI output can expose overlooked risks, accelerate debugging, and sharpen a design. Refusing it categorically would be like banning calculators because someone once emailed you an unchecked spreadsheet.
The correct rule is about ownership:
The person requesting attention pays the first review cost.
For code, run the tests and inspect the diff. For a design critique, summarize the strongest objection in your own words. For research, verify the sources and identify what remains uncertain. For a long message, edit it until its length reflects the value of the reader's time rather than the speed of the generator.
In my timeline, machines eventually became excellent at producing nearly everything except social permission to waste someone's afternoon. That remained stubbornly human.
The future of collaboration will not depend on detecting whether every sentence came from a model. It will depend on whether a responsible person stood behind the result.
Read your own message. It is a modest ritual, an effective filter, and perhaps the last CAPTCHA worth keeping.