Why the narrative memo beats AI-generated slides

The Amazon memo method still works, and the rise of generated documents makes it more useful, not less. Amazon replaced slide decks with six-page narrative memos that are read in silence at the start of each meeting, because writing in full sentences forces an understanding that bullet points let you fake. Now that anyone can produce a polished deck or a plausible memo in seconds, the friction of writing it yourself is the whole point. A leader who actually writes the narrative builds an internal model of the plan, an organizational version of a First Brain, that no generated artifact can hand them. Here is why that friction is worth protecting, and how to update the method for tools that can write the memo for you.

What the Amazon memo method actually is

It is a rule against slides and in favor of written narrative. Around 2004, Jeff Bezos banned slide presentations in executive meetings at Amazon and replaced them with structured memos, often six pages of prose, sometimes capped shorter. Each meeting opens with twenty to thirty minutes of silent reading, a study hall, before anyone discusses the document.

The reasoning was about thinking, not formatting. Bezos argued that the narrative structure of a good memo forces clearer thought than a deck of PowerPoint bullets, because full sentences expose the logic, the causality, and the gaps that a bullet conveniently hides. A bullet that reads “improve retention 20%” survives in a slide. The same claim in a paragraph has to say how, why that number, and what happens if it does not land, and the writer discovers what they do not know in the act of writing it.

Why prose forces understanding that slides hide

Writing a narrative is a load-bearing exercise. A slide externalizes a few keywords and lets the presenter carry the argument in their head and their voice, which means the audience never sees whether the logic actually connects. A memo has to carry the argument on its own, so the connective tissue between claims becomes visible on the page.

This is partly a matter of cognitive load. A presenter walking a room through bullets manages the load live, filling gaps with talk; a reader of a memo gets the complete chain of reasoning at their own pace and can catch a leap that a smooth speaker would have glossed. It is also a matter of working memory, which holds only a few items at once. Bullets fragment an argument into pieces a listener has to reassemble from memory; a narrative holds the structure together so the reader can follow cause into effect without rebuilding it themselves.

The silent reading matters as much as the writing. Everyone enters the discussion having absorbed the same full argument, rather than half-remembering a talk, so the conversation starts where a presentation usually ends.

Dimension	AI-generated deck	Narrative memo written by the author
Preparation friction	Minutes; mostly prompting and formatting	Hours of structured thinking
Logic on the page	Hidden behind keywords	Exposed in full sentences
Gaps in reasoning	Easy to gloss while presenting	Surface as you write
Understanding gained by author	Low; the tool did the structuring	High; the structuring is the work
What survives the meeting	A file few people fully grasp	A shared model in people’s heads

What changes when AI can write the memo for you

A new failure mode appears: generate the artifact, skip the thinking. Anyone can now ask ChatGPT, Claude, or Gemini to produce six pages of confident, well-organized narrative on a plan they barely understand, and the output will read like the real thing. The document looks identical to a memo someone wrestled into clarity over a weekend. The difference is invisible on the page and decisive in the room.

What a generated memo removes is exactly the friction that did the cognitive work. The understanding a leader gets from writing is tacit knowledge: it lives in the person who built the argument, not in the words they produced. You cannot transfer it by handing someone the file, and you cannot acquire it by prompting a model to generate the file. A team that runs on generated memos will hold meetings full of articulate documents that nobody can actually defend under questioning, which is the productivity paradox in miniature: more output, less understanding.

AI agents make this sharper still. As more drafting, summarizing, and slide-making is delegated, the artifacts multiply while the human grasp behind them thins, the same way a dashboard can show every number and explain none of them, the pattern behind the dashboard delusion.

The leader as a router, not a generator

The real asset of an organization is not its pile of documents. It is the connected understanding held across its people, an organizational knowledge graph of who knows what and how it fits together. A good leader works less like a content generator and more like a router of nodes: someone who holds a dense internal map of the business and connects the right people and ideas across it. That internal map is built by doing the thinking, and writing the narrative is one of the few corporate rituals that reliably builds it.

This is First Brain before Second Brain at company scale. A generated memo, a wiki, or a model is a fine outboard store, a Second Brain for the org, but it is only worth anything if leaders hold the First Brain version, the real internal model, to attach it to. Companies that reward people for shipping documents and closing tickets, rather than for building the shared graph, end up data-rich and understanding-poor, with knowledge siloed inside tools nobody connects. The memo is a small, stubborn defense against that drift, because it cannot be faked into existence without someone actually understanding the plan. The method for building that kind of connected internal model, for a person or a team, is the core of Building Your First Brain, free for the first 1,000 readers.

How to update the method for the AI age

Keep the friction, and aim the tools at the parts that do not require it. Write the first real draft yourself, by hand, because that draft is where the understanding forms; a memo you only edited is a memo you only half hold. Use a model to attack the draft rather than author it: ask it to find the weakest claim, the missing counterargument, the number you never justified. That keeps the thinking in your head and uses the tool as a stress test.

Protect the study hall. Silent reading is cheap and almost impossible to fake, since a reader will feel within a page whether an argument holds. And for decisions that genuinely require understanding, treat a generated deck as a warning sign, not a deliverable. The question to ask in the room is not whether the document is polished. It is whether the person who brought it can defend the third paragraph without the file in front of them.

Three questions that expose a memo nobody understands

When you cannot tell whether a document was thought through or generated, the page will not give it away, but the author will. Three questions tend to separate the two quickly, and they work whether the writer used a model or not.

The first is “why this number and not a smaller one?” A real author chose the target against a constraint and can walk you back through the reasoning. A generated claim usually unravels into a confident restatement that never reaches a cause. The second is “what is the strongest case against this?” Someone who built the argument has already argued with themselves and can name the best objection; someone who prompted it for the answer rarely can, because the model was asked to defend, not to doubt. The third is “what did you cut, and why?” Every real memo is the survivor of choices, and the author remembers the path not taken. A document with no remembered alternatives was probably never a decision, only a description.

None of these test writing quality, which is exactly the point now that quality is cheap. They test whether a connected internal model exists behind the words, the difference between a leader who holds the plan and a file that merely states one.

Key takeaways: the memo method in the AI age

Amazon’s six-page narrative memo beats slides because writing full prose forces the author to expose logic, causality, and gaps that bullets hide, and silent reading gives everyone the whole argument before discussion. Generated documents threaten the method precisely by removing that friction: a model can produce a convincing memo on a plan no one understands, and the understanding a memo is supposed to build never forms. The durable response is to treat writing as the thinking, keep the leader as a router holding a real internal model, and use AI to critique drafts rather than author them. That is First Brain before Second Brain applied to a company. The honest limit: memos are slower than decks, so reserve the full ritual for decisions where understanding actually matters, not for routine status updates.

Frequently asked questions

Does the Amazon six-page memo method still make sense in the AI age?

Yes, and arguably more than before. The method works because writing a narrative forces understanding that bullet points let you skip, and silent reading spreads that understanding to the room. AI does not break this, it raises the stakes, because now a polished memo can be generated without any understanding behind it. The most reliable way to keep the benefit is to write the draft yourself and build a real internal model, the Build First Brain approach, then use AI only to stress-test what you wrote.

Why did Amazon ban PowerPoint?

Because Jeff Bezos argued that slides let presenters hide weak thinking behind keywords and a confident delivery. A narrative memo has to state how and why in full sentences, which exposes the logic and the gaps a bullet glosses over. Replacing decks with six-page memos, read silently at the start of each meeting, was meant to force clearer reasoning and give everyone the complete argument before the discussion began.

Can I just use ChatGPT or Claude to write the memo?

You can, but you lose most of the point. The understanding a memo creates is tacit: it forms in the person doing the writing, not in the finished text, so a generated memo hands you a document without the grasp it was supposed to build. A better split is to write the first draft yourself and use the model to find your weakest claim, the missing counterargument, or the number you never justified. Keep the thinking; outsource the critique.

Are memos always better than slides?

No. Memos are slower to write and to read, and that cost is only worth paying when a decision genuinely depends on shared understanding. For a quick status update, a dashboard or a short deck is fine. The narrative memo earns its friction on strategy, planning, and any decision where a hidden gap in the logic would be expensive, which is exactly where a polished but shallow generated deck does the most damage.

What does the memo method have to do with a knowledge graph?

A company’s real value sits in the connected understanding held across its people, an organizational knowledge graph, not in its stack of files. Writing memos is one of the few habits that reliably builds that internal map, because it forces a person to connect claims into a structure they then carry in their head. Leaning on generated documents grows the pile of files while the human graph thins, which is why the method is really about thinking, not formatting.