Six Pages Was Never Proof of Anything

The article making the rounds argues that Amazon's writing culture was a filter for half-formed ideas, and AI has gone and broken the filter. The passage doing the emotional work is this one:

A document used to be proof of thought. If you handed me six pages, I knew that you had spent significant time with the problem. Therefore, I knew that it would be worthwhile spending my time to understand your thoughts.

I understand the feeling behind it. But look at the structure of the claim, because it is not one inference, it is two. Length proved time. Time proved thought. Neither link is a law of nature. Both are economic assumptions: at an hourly rate, six pages costs six hours of a life. When the price of pages collapses to zero, the signal collapses with them. What people are mourning is not broken trust. It is a dead proxy.

The book report test

Almost everyone has, at some point, handed in a six-page report on a book they did not read, written the night before.

Six pages. Proof of time. Zero thought.

The proof-of-time signal was always this noisy, and we tolerated it because faking was expensive enough that most people only faked the boring parts. But anyone who has sat through corporate document regimes knows the second link was failing in both directions. Six-pagers got ghostwritten, team-rotated, and recycled into what amounted to paperwork theater. If length were thought, the internals wiki of every large company would be the most valuable literature on earth.

Writing that forces thinking is real, but it was never the length doing the forcing, and the six-pager was never as reliable a detector of thought as the nostalgia claims.

Two claims wearing one coat

The article, like most in this genre, runs together two propositions that should be kept apart.

The first is about the author: writing is a tool for thought. Forcing vague intent into sentences makes the incoherence surface, which is why the practice produces better plans. This claim is roughly true, and it survives AI, because the benefit comes from working through the ideas. If a model does the thinking for you, the benefit disappears. If you use it to articulate, challenge, and refine your own thinking, the benefit survives, even if the model generates every sentence.

The second is about the reader: the document is evidence, a reason to spend your hour. This claim was always false, as the book reports prove, and AI merely made the falseness cheap to notice.

Once you separate them, the panic resolves into something smaller and more fixable. The real loss is that a reader can no longer assume the author spent time. Losing that matters. But you do not get it back by reverting to word counts, because you never actually had word counts doing the work. You had the ritual around the document.

The ritual was never the paper

The Amazon ceremony is the tell. Thirty silent minutes of reading, red pens in the margins, senior engineers interrogation-tested before a decision moved. That room is where trust lived. The six pages were the ticket you bought to get in.

Note what the article's own anecdote says about the room. A presenter reads his document aloud for the first time in the meeting, it becomes obvious to everyone, and nobody challenges him, because he is the most senior person there. The ritual was already dead in that meeting. Hierarchy killed it, not the model. The document culture survived the paper's death for years and did not notice, which suggests the culture was more about the paper than about the thinking all along.

The tell that the author does not see coming arrives in his own policy section. He endorses a team rule that includes "Longer is not better. If you have generated a long document from a short prompt, share the prompt instead." You do not write that rule if length was the evidence. He is legislating against the exact nostalgia he sells.

What replaces the word count

If length was a proxy for thought, the successor has to be something that costs a faker more than it costs a sincere author. Not volume. Accountability.

Stand behind every sentence. Generated or typed, if it is in your document, the claims are yours, and you answer for them in the room. This is the best line from the policies being written right now, and it works because it moves the cost from producing text to defending it, which AI cannot take off your table.

Make the thinking inspectable instead of inferred. Drafts, edits, revisions, and decision logs can help a reader follow how an argument developed and where to question it. They are supporting evidence, not proof: an agent can generate a convincing history, and a person can think carefully before writing a single draft. The useful part is being able to explain what changed, why it changed, and what evidence would make you change your mind again.

Match the ritual to the risk. One-way doors deserve the slow ceremony, humans reading humans with red pens. Two-way doors should ship, roll back, and stop pretending the paperwork was the safety. Most of what died with the six-pager was two-way-door writing that only looked serious.

Stay answerable. The test of a thought was never its length, it was the ten-minute questioning afterward. If you can survive it, the format does not matter. If you cannot, no format was going to save you.

Six lines instead of six pages

I have been building toward this format without noticing. ARF is a small spec for capturing agent reasoning: what, why, how, and backup, four fields, with the rollback plan required before the action ships.

Six lines, not six pages. And the backup field is where the thinking becomes testable. A model can produce a plausible rollback plan in seconds. The useful questions are whether the plan works, whether you can execute it with the access and state you will actually have, and whether you can explain what it will and will not recover. A short plan you can validate and defend is worth more than a document that takes a week and says nothing.

That is the same shift I wrote about in I Don't Type Every Word You Read, arriving on the reader's side of the page. The work moved. It lives in what you decide, what you keep, what you can defend, and what you can undo. It does not live in how long it took you to type.

The six-pager was never proof of anything. The people worth trusting did not need it, and the people faking it were buying a ticket, not doing the work. What AI actually ended was our excuse for counting pages instead of building rooms where counting is unnecessary.