Write for People. Package for Agents.

I caught myself doing the thing I tell everyone not to do. I let AI write a 10-page document and sent it to an executive. I don’t think he read it. I don’t think anybody reads those things.

In the last few months, I’ve fully integrated AI into my daily workflow. Research, analysis, writing drafts, specs, data modeling, vacation planning, lifting workouts… you get the idea. It’s the best thinking partner I’ve had.

And that’s exactly where it went wrong.

I started letting AI write the artifacts I was sharing with other people. Ten-page documents that say a lot and communicate nothing. Perfectly formatted. Something nobody asked for. Pure AI slop.

Then I started receiving similar documents from other people. That’s when I realized how bad it really was.

The Value is in the Thinking

Using AI to explore a problem makes me sharper. It’s not doing the work. It’s accelerating my thinking.

But the minute I copy-paste that output into a doc and send it, the value stays in my head. The other person just gets a wall of text. Volume is not communication. Most AI output is organized around structure, not insight.

The sender illusion is powerful. The larger the document, the more value it feels like you’re creating. Not that I thought I was being thorough in high school, but it reminds me of hitting a word count on papers. Pure volume, no intellect.

The One-Pager Discipline

The best board meetings I’ve attended follow a simple pattern. A one-page executive summary sent as a pre-read to set the frame. A three-hour meeting is the discussion and conversation. Everything worth discussing fits on one page. It’s easily digestible.

If that sounds brutal, it is. Most of what we write doesn’t survive the one-page test. That’s the point.

So I’m instituting a new rule. Everything I share with executives, engineers, or stakeholders gets a human-written one-page summary. Written by me. Not AI. The summary comes from genuine understanding. If you can’t distill it to one page, you haven’t fully understood it yet. Writing the summary improves your thinking and understanding. Outsourcing this to AI defeats the entire purpose.

The README Model

But what about the supporting facts? The research? The heavy analysis behind the one-pager?

If you’ve worked with technical products you know that every GitHub repository has a README file and a bunch of more complex files underneath it. The README is the human entry point. The codebase is the depth.

Why shouldn’t we do the same thing with how we share information?

Instead, we do the equivalent of sending someone the entire codebase with no README and expecting them to figure it out. That’s what a 10-page AI-generated document is. A repo with no entry point.

The one-pager becomes a human README. A structured folder of markdown files, research, and supporting analysis becomes the agentic layer underneath it. I now include a project folder with anything I send out.

Layer 1: Human-written one-pager. No AI.

Layer 2: Structured folder of markdown files, data, and analysis that the recipient can upload to their favorite AI tool and explore on their own terms.

This respects human attention and enables the depth needed for a deeper dive. The recipient chooses their own depth. Some read the summary and move on. Some spend an hour querying the folder with Claude. Both are valid.

The Hiring Signal

One place this already works: recruiting. We spend an awful lot of time recruiting CEOs for our ventures. Now I share a one-page opportunity overview with the candidates. As they go deeper in the process, they get access to the full research folder.

Whether the candidate knows what to do with a folder of markdown files tells you more about their AI fluency than any interview question. We’ve accidentally built a literacy test.

Sharing project folders will become as normal as sharing Google Docs in the next six months. The people who figure out how to structure those folders well will have a massive communication advantage.

The Bigger Shift

There’s a larger shift happening here. We’re moving from “documents as communication” to “documents as context” for agents.

The executive summary is the human interface. The folder is the agentic one.

This isn’t about generating less. I probably use AI more than almost anyone I know. It’s about knowing which layer is for humans and which layer is for machines. The 10-page doc was me confusing the two.

Write for people. Package for agents. Stop confusing the two. The best communicators in the AI era aren’t the ones who generate the most content. They’re the ones who distill the hardest and package the best.