AI Can Write Amazon's Six-Page Memos in Seconds. That's the Problem

Amazon's internal AI assistant Cedric promises to deliver "six-page narratives in seconds," according to Kristi Coulter, a former Amazon executive. That speed breaks the mechanism Jeff Bezos built in 2004, when he banned PowerPoint from Amazon's senior leadership meetings and replaced decks with narrative memos read in silence.

Bezos was explicit about why writing mattered. "There is no way to write a six-page, narratively structured memo and not have clear thinking," he said. The memo itself was never the point. The value came from what writing it forced the human to do.

How AI breaks the memo

Writing at Amazon served as a forcing function for rigorous thinking about important decisions. One Amazon veteran told Coulter that writing the memo forced them to confront their own ideas and clarify what they wanted to say. Writers often changed their mind several times mid-narrative.

The silent-reading format disciplined readers too. Everyone worked through the same argument before anyone spoke, so discussion started from substance. But the value of reading a memo falls in proportion to how little of the writer's thought it carries.

The evidence of breakdown is already visible:

• Steve Huynh, a former Amazon principal engineer, described a review meeting where a product manager began discussing a six-page memo he was supposed to have written — and it immediately became clear he was reading it for the first time.
• One Amazon manager now asks for longer documents because he plans to have AI summarize them, Coulter reports.
• Some employees tag suspect memos "AI;DR" — AI, didn't read.
• In a Stanford and BetterUp Labs survey on "workslop," 42% of workers who received low-effort AI-generated work said they saw the sender as less trustworthy.

Once a written product stops being evidence that someone thought about the content, readers stop giving it attention.

What did the memo actually do?

The trap for leaders is to treat this as a problem about memos. Businesses do not need memos. They need good decisions. The memo mattered only because, for a long time, it reliably communicated the results of the thinking behind it.

Michael Lewis's Moneyball offers the useful comparison. When the Oakland A's lost Jason Giambi to free agency, they did not hunt for another Giambi. They identified his measurable statistical impact and recreated it "in the aggregate" with several cheaper players. The lesson: don't substitute AI work for human work step by step. Look at how work produces value in total, then redesign the workflow to reproduce the end result.

A well-written memo bundles three distinct jobs into one artifact:

1. It forces authors to think.
2. It provides evidence the author has done that thinking.
3. It gives readers shared understanding of the issue.

Leaders should restore those jobs in two separate moves, not one.

Force the thinking, then prove it

The first move targets the first two jobs. Make the author, not the document, the thing that gets examined. Amazon's meetings already contained the seed: after silent reading ended, the writer answered for every line. The AI-era change makes the defense the initial gate — a memo earns an agenda slot only if its author can walk the room through the argument and take questions without the document in front of them. Cedric cannot sit in that room, and knowing the defense is coming does what the blank page used to do.

Terence Tao, the Fields medalist, has proposed the same rule for mathematicians: if the authors of an AI-assisted result cannot give a clear, expert-level talk on it, the result should not be published. A proof no human can properly explain "should be viewed as incomplete, even if it has been formally verified," he writes. The same holds for a memo.

Not all writing needs this treatment. A weekly project status report can be an assemblage of information, and automating it costs nothing. Visions of the future and arguments about major decisions are different: polished prose no longer guarantees deep thinking.

Who does the understanding?

The third job raises a counterintuitive problem. AI can help humans produce more valuable output than other humans can absorb, even when the content is genuinely good.

Mathematics shows the pattern. Mathematicians face a flood of journal articles and preprints containing AI-generated proofs nobody can digest. In September, OpenAI said its models had solved one of math's Millennium Prize problems. A month later, it released hundreds of AI-generated results in one batch. Software has verified many of the proofs' logic, but knowing a result is correct is not the same as understanding why. The Association for Human Mathematics stated bluntly, in a post reposted by Tao: "We reject OpenAI's assertion that this release advances our subject."

Pavel Etingof, a mathematician at MIT, argues AI results become part of mathematics only when a human reads them, understands them, and gets something out of them. A thousand unread proofs on a shelf add nothing. A memo, however intelligent, enters a company's decision-making only if the relevant humans absorb it. AI can shoulder the generative burden; understanding stays a human task.

That makes human understanding a scarce resource leaders may have to ration. The preprint server arXiv now limits authors to two submissions a month. A company could cap the number of memos that get a silent-reading slot each month.

For two decades, a six-page memo was reasonable evidence someone had thought hard. AI broke that link, and no policy will fully restore it. The alternative is more six-pagers than ever — and fewer people who understand what's in them.