Two People. Same Job.

AI Is About to Separate Them.

You're not behind on tools. You're behind on the system that makes them work — and it's already compounding.

The Gap Is Hardening

Anthropic just published its Economic Index: Learning Curves — the most detailed look yet at how people actually use AI — and the findings are uncomfortable for anyone who thinks fluency is just a matter of time.

The study examined over one million conversations on Claude last month. The headline: experienced users — those six months or more in — achieve a 10% higher success rate in their AI interactions than newcomers. The gap isn't explained by which tasks they're doing, which country they're in, or which model they're using.

"The real divide isn't between people who use AI and people who don't. It's between experienced AI users and newcomers."

— Axios, Behind the Curtain: America's next class war: AI fluency

That's not an abstract data point. Think about what it means in practice. Two people, same role, same company. One has been building with AI for eight months. The other started two months ago. They're not just using different prompts — they're operating at different levels of leverage. The experienced user is producing more, faster, at higher quality. The gap isn't visible yet. But it compounds every week.

Two workflow categories doubled in prevalence between November and February: automated sales and outreach, and automated trading. The Anthropic data is blunt about the implication: people doing those tasks manually are at growing risk of being outpaced — not by AI, but by the colleagues who learned to use it systematically.

This is already showing up in hiring, performance reviews, and who gets trusted with more responsibility. The fluency gap isn't a future problem. It's a present one — it's just invisible to the people on the wrong side of it.

This is what Peter McCrory, Anthropic's head of economics, described as the core insight: you can develop skills that make you better at getting value out of AI. The fluency gap is learnable. But only if you stop treating AI as a chatbot and start treating it as infrastructure.

It's Not a Tool. It's a System.

Here's the mental model most people are still running:

Open Claude. Type a prompt. Get an answer. Repeat tomorrow.

That works. But it doesn't compound. Every session starts from zero. You re-explain your voice, your format, your audience, your process. You're not building anything — you're just having a series of very smart conversations that evaporate the moment you close the tab.

The shift that separates high-fluency users from everyone else is architectural. They've stopped asking Claude to help with tasks. They've started building systems that run those tasks automatically.

Claude's recent evolution — Projects, Knowledge, Skills, MCP — isn't a feature list. It's a blueprint. Each layer serves a specific function:

Miss any layer, and the system is incomplete. Most people are missing at least two.

The Layer Most People Skip: Claude Skills

I want to spend time here, because Skills are the single most powerful feature the majority of Claude users have never set up — and the one that most directly closes the fluency gap the Anthropic data describes.

The Moment It Clicked

A few weeks ago I caught myself doing something I've done dozens of times. I had Claude open. I was drafting a structured report. And I started typing — tone, format, structure, output expectations.

Again.

Same instructions. Same setup. Same rebuild from scratch.

Halfway through I stopped. Because this wasn't an AI problem. It was a system problem. If I have to explain the work every session, I haven't actually built anything.

Retyping the same instructions every morning is like onboarding the same employee every single day.

What a Skill Actually Is

A Claude Skill is a packaged set of instructions — a SKILL.md file — that teaches Claude exactly how to perform a specific task, in your voice, to your standards, without re-explanation. Anthropic defines them as reusable instruction systems that Claude loads dynamically to improve performance on specialized tasks (read: What are Skills? →).

In plain terms: it's an SOP for your AI. You build it once. Claude follows it automatically every time that task comes up.

Without a Skill, every conversation starts from scratch. With a Skill, you do the thinking once — and then it compounds.

Why This Matters More Than It Did Six Months Ago

When Skills launched, creating one required writing a SKILL.md file by hand, packaging it into a zip, and uploading it manually. It was tedious, error-prone, and easy to avoid.

That changed. Now you simply ask Claude to turn a working prompt into a Skill. It generates the properly formatted file. You install it with one click. From that point forward, Claude recognizes when the Skill applies and uses it — no commands, no setup.

The process in practice: identify a task you repeat, ask Claude to turn your best prompt into a Skill, install it. The whole thing takes under five minutes. The payoff compounds every time you use it.

The Portability Factor

There's one more dimension worth understanding. Anthropic published the Agent Skills format as an open standard — and the industry followed. The same SKILL.md specification now works across Claude, ChatGPT, Perplexity, GitHub Copilot, Cursor, and a growing list of other platforms.

That means every Skill you build today isn't locked to one tool. It's portable expertise you carry across the entire AI ecosystem. The time you invest now becomes more valuable as adoption grows.

What High-Fluency Users Do Differently

The Anthropic data makes a distinction worth internalizing. There are two modes of AI use: automation (do this task) and augmentation (think with me, build with me, stress-test with me).

Using AI as a search engine or copy editor is what Axios CEO Jim VandeHei calls dumb AI. The people pulling ahead are using it as a thought partner, a builder, a system operator. The Anthropic research found that skilled users are getting better at collaborating with Claude across a wide variety of work — not just automating specific activities.

But here's what the data doesn't show you: how they got there. The answer, almost universally, is that they stopped starting from scratch.

What It Looks Like in the Real World

Consider two sales reps at the same company, same quota, same tools. One still writes prospecting emails from scratch — adjusting tone, researching the account, formatting the follow-up manually. It works. He's competent.

The other built a Skill three months ago. She drops in the account name and two context notes. Claude pulls from her messaging playbook, mirrors her voice, formats it to her standard, and outputs a ready-to-send sequence. The whole thing takes four minutes instead of forty.

Same AI. Same model. Completely different leverage. The gap between them isn't skill. It's architecture. And it widens every week.

The pattern looks like this: pick one recurring workflow, build a Project around it, add your documents, create a Skill, refine outputs over time. Now it runs better every week. That's compounding. That's what the fluency gap is actually measuring.

What to Do This Week

Here's exactly how I built the system I use every week — and how you can replicate it in a single session.

My primary Skill is a newsletter drafting system. Every issue of Signal starts the same way: I drop in my notes, sources, and angle. Claude pulls from a Skill that knows my voice, my structure (TLDR → thesis → example → action), and my formatting standards. It doesn't ask what tone I want. It doesn't need to be told what Signal sounds like. It already knows. The first draft is 80% there before I touch it.

Layered under that is a weekly reporting Skill — same logic, different output. When I need to summarize performance, surface insights, or brief a client, I'm not starting from scratch. I'm refining something that already understands what I'm trying to say.

That's the system. Here's how to build yours:

One workflow. One Skill. One Project. You'll feel the difference the first time you run it — and you'll understand exactly why the fluency gap exists. The people on the right side of it aren't smarter. They just stopped rebuilding from scratch.

If it still feels like work, you haven't built the system yet.

The Bottom Line

The Anthropic data landed quietly last week, but its implications are significant. A two-tier workforce is already taking shape — not between AI users and non-users, but between people who have built systems and people who are still having conversations.

Washington isn't addressing it seriously. Most companies aren't either. But you can.

The infrastructure is already there: Projects, Knowledge, Skills, MCP. You don't need to understand all of it today. You just need to start building — one workflow, this week.

The gap between the two groups isn't talent. It's architecture.

AI doesn't just help you work. Eventually, it starts working for you. But only once you've built the system underneath it.