Use .ai for AI files. CTA

File extensions matter more than we like to admit.

They’re not just a technical footnote — they’re a shared language. When you see .jpg, you expect an image. When you see .py, you expect Python. When you see .md, you expect Markdown.

And that’s exactly the problem.

Right now, the vogue in AI tooling is to use .md for everything: prompts, agents, skills, workflows, configs, instructions, personas, “AI apps,” and half a dozen other concepts that are not Markdown documents in any meaningful sense.

It works — but it’s sloppy. And we can do better.

The .md Extension Is Overloaded

Markdown is a format with a clear purpose: human-readable structured text. Documentation. Notes. READMEs. Blog posts like this one.

But agent and skill files aren’t Markdown documents. They may contain Markdown, but that doesn’t make them Markdown-first artifacts.

An agent file:

• Has behavior
• Has constraints
• Has identity
• Has instructions intended for execution, not just reading

That’s not documentation. That’s a program — even if it’s written in plain text.

When everything is .md, you lose:

• Semantic clarity
• Tooling precision
• Visual scanning speed
• The ability to reason about files at a glance

We’re repeating a mistake the industry has made before: using a generic container because it’s convenient, then paying for it later in confusion and conventions piled on conventions.

Agents Deserve a First-Class Extension

Agents and skills are becoming primary software artifacts. They are not comments about code. They are the code.

That’s why I’m making a simple proposal:

Use .ai as the standard extension for agent and skill files.

It’s short. It’s obvious. It’s meaningful.

When you see:

customer-support.ai
researcher.ai
pricing-strategy.ai

You immediately know what you’re looking at.

No guessing. No “open it to see what kind of markdown this is.” No tribal knowledge.

“But Adobe Illustrator Uses .ai”

Yes. It does.

And that’s… fine.

File extensions are contextual. .ai files from Illustrator are binary vector graphics. .ai files for agents would be plain text, versioned, diffable, executable by AI runtimes.

No one confuses .js with .json, or .rs with .rtf, because meaning emerges from ecosystems, not monopolies.

Still, if Adobe wants something truly unique, there’s an easy solution.

A Friendly Suggestion for Adobe: Use .ail

If Adobe Illustrator is concerned about namespace purity, they could move to:

.ail  (Adobe Illustrator Layer / Language / Layout)

Or any Illustrator-specific variant they prefer.

That wouldn’t break the world — Illustrator files already embed PDF compatibility, version metadata, and proprietary structures. A new extension would actually clarify things.

Meanwhile, .ai would be free to mean what it increasingly already does in practice: artificial intelligence artifacts.

Specific Extensions Create Better Tools

When you give something a specific extension, you unlock:

• Syntax highlighting
• Validation
• Linters
• Loaders
• Runtimes
• Search and filtering
• Clear mental models

Editors can treat .ai files differently than .md.
Repositories can separate agents from docs.
Build systems can discover skills automatically.
Humans can scan directories and understand intent instantly.

This is how good software ecosystems grow: by naming things clearly and early.

This Is the Moment to Decide

We are still early.

The conventions around agents, skills, and AI-native software are not locked in yet. That’s rare — and valuable.

If we keep defaulting to .md because “it works,” we’ll be stuck with it long after it stops making sense. If we choose a purpose-built extension now, we’ll save ourselves years of awkward compatibility layers and explanations later.

So this is an open call:

• If you’re building agents: use .ai
• If you’re defining skills: use .ai
• If you’re writing behavior meant to be executed, not just read: use .ai

Let Markdown be Markdown.

And let agents have an identity of their own.