Writing and Communication

"Writing is thinking. But not all thinking needs to be done at the keyboard."

The Real Job of Business Writing#

Most business writing is not literature. It's transmission: getting the right information from your head into the right shape for the recipient to act on it.

A proposal needs to make the case clearly and in the right order. A board memo needs to surface the risks with the right framing. A client update needs to be specific enough to be useful and concise enough to be read.

These are structural problems as much as they are language problems. And structural problems — organizing, sequencing, tightening, covering the right points — are exactly what AI handles well.

This chapter is about using AI for that structural work while keeping the judgment, the voice, and the relationship work in your hands.

Where AI Writing Helps#

First drafts from notes#

The worst part of writing a proposal or a report is the blank page. You know what you want to say. You have the notes, the numbers, the arguments. You just don't have the draft.

AI collapses this step. Paste your notes, describe the document, tell it who it's for and what decision it needs to support, and you get a draft to react to. A draft is almost always easier to fix than a blank page is to fill.

This is the highest-leverage use of AI writing. Not editing — starting.

Workflow:

1. Write your notes freely — bullet points, fragments, whatever's in your head
2. Tell Claude: "This is for a proposal to client, the goal is outcome, the audience is who. Draft it."
3. Read the draft. Mark what's wrong. Mark what's too long. Mark what's missing.
4. Ask Claude to fix the specific things — not "make it better," but "the second paragraph overstates our capacity — soften it" or "the pricing section needs to come before the timeline"
5. The fourth or fifth version is yours

Tightening and restructuring#

A document you've written yourself is often too long, too repetitive, and in the wrong order — because you wrote it in the order you thought of things, not the order the reader needs them.

AI edits well at the structural level. "This memo is 800 words. Cut it to 400 without losing any of the key decisions." "The executive summary buries the recommendation — move it to the top." These instructions produce useful results.

What it edits less well: tone, register, and the specific word choices that define your voice. Expect to put those back.

Formatting and standardization#

If you send client updates, investor memos, or team communications regularly, you probably have a format in mind that you never wrote down. AI can codify it and apply it consistently.

"Here is an example of a client update I'm happy with. Here are my notes from this week's client call. Draft the update in the same format."

Every subsequent update takes ten minutes instead of forty.

Translating between audiences#

The same information often needs to go to different audiences: the board, the operations team, the client, the press. Each needs a different framing of the same facts.

"Here is the operations team briefing. Rewrite it for the board — they need the risk framing and the financial implications, not the technical details." This is a legitimate use. The facts don't change. The emphasis and language do.

Where AI Writing Undermines You#

Anything where your voice is the product#

A personal note to a key client after a difficult conversation. A eulogy. The email to your team after a layoff. A message to a co-founder about a hard decision.

These communications work because they carry your specific presence. The reader needs to feel that you wrote it — that you thought about them specifically, that the words are yours. AI-drafted versions of these feel exactly like what they are: competent but impersonal.

You know which emails these are. They're the ones that feel important when you're sitting down to write them. Write those yourself.

When the content is the thinking#

Paul Graham's observation: "writing is a way of thinking." When you write a strategy memo yourself, you often discover what you actually think in the process. The writing is not just transmission — it's development.

If you AI-draft the memo, you skip that part. You may end up with a well-structured document that hasn't forced you to confront the gaps in your own thinking.

Use AI for drafts of things you already understand. Use writing for things you're still figuring out.

When accuracy is the job#

AI writing tools hallucinate numbers, misremember details, and state things with confidence that aren't true. For a rough internal draft that you'll fully review, this is manageable. For a client-facing document with specific figures, quotes, or regulatory claims, every fact needs verification before the document leaves your hands.

The rule: if a number or a claim matters, check the source yourself, not the AI's recollection of the source.

When the recipient will know#

Some writing needs to be demonstrably human. A cover letter for a key hire. A CEO public statement about a difficult situation. A personal recommendation for someone you know well. If the recipient will read it and think "did they actually write this?", write it yourself. The cost of that perception is higher than the time saved.

How to Stay the Author#

The concern with AI writing is losing your voice. The way to avoid it:

You write the substance, AI writes the structure. Your notes, your arguments, your judgments — in your words, however rough. The AI's job is to organize and expand them, not to supply the ideas.

React, don't accept. Treat every AI draft as a first draft from an assistant who doesn't know your style yet. Mark what's wrong, what's missing, what's overstated. Those corrections are your authorship.

Read it aloud. Any sentence that doesn't sound like you needs to be rewritten. This is a fast filter.

Keep the last edit. Whatever final version goes out should have your hands on the last pass. If you read a sentence and wouldn't have written it, change it. This is what it means to remain the author.

Practical Workflows#

The proposal#

1. Write your notes on the client situation — their problem, your solution, why you, the investment
2. Describe the output: "Draft a 2-page proposal for client. The goal is to convince them to engage us for project. The tone should be direct, not salesy."
3. Get the draft. Read it fully. Mark three things that are wrong.
4. Fix those specifically. Repeat until the structure is right.
5. Rewrite every paragraph that doesn't sound like you.
6. Review the numbers and claims manually.

Total time: 45 minutes instead of 3 hours. But only if you give the AI enough substance to work with.

The client update#

1. After the client call or project milestone, write three bullets: what happened, what's next, what you need from them
2. "Draft a client update email from these bullets. One page maximum. Professional but not stiff."
3. Review, adjust tone, check accuracy.
4. Send.

Total time: 10 minutes.

The board memo#

1. Write your full thinking first — what happened, what it means, what you're recommending, what the risks are. This is for you, not for the board.
2. "Convert this into a board memo. Recommendation at the top, supporting analysis below, risks in the third section. Keep it under 600 words."
3. Review. Check that the recommendation is right. Check that the risks are complete.
4. The writing that was hard is done. The thinking is yours.

Total time: The thinking takes as long as it takes. The memo takes 30 minutes.

The Temptation to Avoid#

The most common mistake: using AI to write something you haven't thought through yet.

This produces documents that are well-structured but wrong. The arguments are in the right order. The framing is professional. But the recommendation is off, because the thinking wasn't finished before the writing started.

AI can help you say what you've already decided. It cannot help you decide what to think.

Think first. Then draft.

Next: Meetings and Memory