I was looking at our Claude usage at Hi Marley last week. The number two heaviest user in the company wasn't an engineer. It was someone on our People team. Working on a PowerPoint.

Stop and think about that for a second.

We have an organization full of engineers. We have data scientists. We have a product team shipping AI features daily. And the second-most aggressive AI user in the company is a non-technologist trying to build slides.

That's not a bug. That's a signal. The highest-leverage AI users in most companies are the people still chained to the legacy file formats.

1983 Called

Microsoft Word was released on October 25, 1983. PowerPoint shipped April 20, 1987. They were genius. They were innovations of their era — paired with the GUI, the mouse, and the dawn of personal computing. They gave non-programmers the ability to produce professional-looking documents and presentations. That was a real democratization.

They served us well. For decades.

But it has been forty-two years. Forty-two years of technology change while the file format underneath stayed essentially the same.

The original .doc format was a binary "Compound File" — Microsoft's own documentation literally describes it as "extremely complex" because they had to "prioritize performance over anything else." When they modernized in 2007 to .docx and .pptx, they switched to OOXML — zipped XML files. Sounds modern. The catch: the OOXML specification is roughly 6,000 pages.

For comparison, the original Markdown spec by John Gruber fits on a single web page.

And it's not just the spec. I tested this before publishing: I created the same simple document — one heading, four paragraphs — in three formats. The markdown file was 362 bytes. The HTML was 508 bytes. The .docx? 36,840 bytes. Roughly 100× larger for the same content.

And the vast majority of those bytes aren't your document. They're machinery — theme files, style definitions you're not using, font tables, an embedded thumbnail. Your actual text was about 0.27% of the unzipped contents.

That difference matters now in a way it didn't before. Because the writer is changing.

The New Substrate

When AI generates output for a human — or for another AI — it produces one of two things by default: Markdown or HTML.

Not because it was told to. Because that's the natural language of structured information for a model. Markdown is plain text with semantic conventions. HTML is structure plus presentation plus interactivity, all in plain text. Both are lightweight. Both are infinitely expressive. Both can be diffed, version-controlled, indexed, and retrieved.

Now imagine asking a model to write a Word document instead. It has to mentally translate its native expression into a 6,000-page-spec foreign format. The model has to unzip its own thinking, parse it through OOXML, and pretend it had a stylesheet hierarchy in mind. It can do it. But you've shackled the most powerful tool we've built to a substrate that wasn't designed for it.

FILE FORMAT FACE-OFF
Property Word / PPT Markdown / HTML
Format Zipped XML (OOXML) Plain text
Spec size ~6,000 pages ~1 page (MD)
File size (same content) ~37 KB ~0.4 KB (MD)
AI-readable Awkward Native
Diffable No Yes
Version control No Yes
Interactive No Yes (HTML)
Reading mode Stand-and-deliver Self-serve

The Direction of Information

There's another problem with these formats that we don't talk about enough. They have a direction.

A Word document is vertical. It reads top to bottom, in one direction. Like a scroll. A PowerPoint is horizontal. It reads left to right, in one direction. Like a strip of film.

Information in 2026 is not one-directional. It's networked. It branches. It links. It updates. It responds to what the reader knows and what the reader wants to know.

HTML was literally designed for this. Markdown plus a renderer can do most of it. Word and PowerPoint can do none of it.

And PowerPoint has a second problem: it presupposes a presenter. A meeting. A room. A "stand and deliver." The whole format assumes someone is going to walk an audience through it. HTML assumes a reader. It lives where it's read. That's a worldview shift, not just a format shift.

Try This

Here's an experiment. Next time you're about to ask Claude (or any model) to build you something, swap one phrase:

try-this.txt
> "build me a Word document for my exec session"
> "build me a single-page HTML document for my exec session"
# Bonus: use voice mode (Superwhisper or Cmd+G).
# Talk through what you want for 5-10 minutes.
# If you're using Claude, enable the front-end design skills.

What you'll get is not a document. You'll get a thing. A self-contained, interactive, readable, beautiful artifact. It will have its own typography. It might have a table of contents that scrolls with you. It might have collapsible sections. Tabs. A search bar. An embedded chart that updates when you hover.

And it'll be one file. Lightweight. Linkable. Hostable anywhere.

What We've Seen at Hi Marley

We've been doing this for months. Internal briefs, planning docs, decision memos, executive readouts — all built as single-page HTML droplets instead of decks or memos.

The efficiency gain is significant. Information is denser, easier to digest, easier to share. Decisions happen faster because everyone's looking at the same artifact in their browser, on their own time, instead of waiting for a meeting where someone clicks through 47 slides.

We even built our own internal platform for creating, hosting, sharing, commenting on, and versioning these single-page HTML artifacts. Auth, permissions, comments, version history. The whole thing.

I might open-source it. We'll see.

"But What About Collaboration?"

This is the objection I get every time. And it's a fair one. Two things shackle people to Word and PowerPoint:

One: collaboration. Real-time editing, comments, suggestions, track changes.

Two: auth and SSO. The infrastructure for "this is shared with my company, not the world."

For thirty years, those two needs were a moat. You couldn't reasonably build them yourself.

You can now. We're in the era of Claude Code. Building a collab-and-comment layer on top of HTML artifacts is a single-prompt away from being a working prototype, and a weekend away from being a product.

The ground shifted. The moat drained. And most companies haven't noticed because they're too busy asking Copilot to make them a better PowerPoint.

The Trap

Here's what I worry about most.

Microsoft, Google, and the big AI vendors all know everyone uses Word and PowerPoint. So their integrations are designed to shoehorn AI into those tools. Make slides faster. Generate paragraphs faster. Summarize the deck faster.

Which means the average non-technologist is being given an extremely powerful tool, pointed back at the same legacy substrate, and told "now make more of those."

The end state of that path: more PowerPoints, faster. Bigger Word docs, faster. The same one-directional, schema-bound, file-format-as-a-stone-tablet artifacts — just in greater volume. That's a real cost. To attention. To clarity. To everyone who has to sit through the meetings.

It is time to unshackle yourselves from these file formats.

Word and PowerPoint were one of the great gifts of the personal computer era. They earned their place. But they were designed for a world where the writer was a human, the reader was a human, and the medium was paper-shaped.

None of that is true anymore.

Try the experiment this week. Ask for a single-page HTML document instead of a deck. Talk to your machine. See what comes out the other side.

It's lighter than a stone tablet. I promise.