Stop learning AI tools.

Every week there’s a new AI tool. Every week everyone acts like you should already know what it does, how it’s different, why it matters and how you’re falling behind.

Should I learn Cursor? Do I need Claude Code? What about Lovable, v0, Paper, Figma Make? What do I even do? I’ve heard these questions from a bunch of designers. I’ve asked them myself.

You absolutely do not need to learn every tool.

The naming is inconsistent. The categories blur. The hype cycle resets every Monday. One tool is “the future of coding” on Tuesday and irrelevant by Friday. The conversation moves so fast that even people who work in AI full-time can’t keep up.

So what do you do with all of this?

STOP trying to learn every tool or feeling like you have to. Learn the fundamental layers instead, which makes learning ALL the tools much easier as they come and go.

Before we get to the framework, something more useful to think about is where are you right now?

Every person engaging with AI falls somewhere on a spectrum. Not necessarily a skill ladder, but a spectrum of control versus convenience.

There are three positions on this spectrum, and none of them is better than the others. They’re just different relationships with the technology.

1) Pioneer
​You’re in the terminal. You run Claude Code, Codex, or OpenClaw directly. You choose your own model, configure your own tools, and build your own workflows from scratch. You see everything, control everything, and carry the full cognitive load.

This is where a lot of the turbulence is happening. You’re likely a developer, or at least developer-adjacent, or just super interested in the latest and greatest tech. You’ve been experimenting with AI agents since before most people knew what an agent was.

2) Builder
​I believe everyone now lives at the builder level, both developers and designers, and the continuum between them is shrinking fast.

You’re making things with AI through a surface that handles complexity for you. Maybe Cursor or Lovable or Figma Make. You’re actively creating, but you’re leveraging opinions someone else baked into the tool. That’s not weakness. That’s speed.

Developers are faster at writing code than ever. Designers are building UIs faster than before. But solid developers are still needed to make strong decisions. Solid designers are still needed to sweat the details and add polish. AI is the rocket ship that gets you to 75% done. Humans fill in the last 25%.

We used to have a VERY distinct split between developer and designer. We still do, but it's rapidly becoming one category with a very broad continuum with traditional developers on one end and designers on the other.

I think we're now done with developer-only or designer-only. People are simply becoming builders. Developer-centric builders and design-centric builders.

For this reason, I think it's critical for designers to start experimenting as much as possible with all of these new tools.

3) Consumer
​This is the person using ChatGPT instead of Google. AI is in your tools and you use it.

Gmail suggests your sentences, Notion summarizes your notes, and Figma removes backgrounds of your photos or renames your layers.

You didn’t opt into “using AI.” The software just got smarter one day, and you appreciated it.

You’re not building with AI, you're simply benefiting from it.

I don’t believe we are moving toward a future where my mom is building her own grocery delivery app because it’s “faster and easier.”

There may be apps that let some tinkering consumers build custom things, but I'm not so sure that will be the new norm. This is where the vast majority of people on earth live, and there’s nothing wrong with that.

Take a second to think through this.

Which one are you?

The same way there are three personas...

the framework I've been using explains why they exist.

Every AI tool ever built sits on a stack of three layers.

These layers won’t change.

The tools on top of them will change every week, but the layers are structural. They’re the foundation underneath all of it.

Learn the layers once and you’ll never be confused by a new tool launch again.

1) Intelligence

This is the raw AI brain. The model. It’s what everyone means when they say “Claude” or “GPT” or “Gemini.”

Intelligence is what the AI knows and how it thinks. It processes text, images, code, and audio. It reasons, generates, and predicts. It’s incredibly powerful and completely inert. On its own, it just sits there waiting for input.

The Intelligence layer has its own spectrum…

You don’t interact with the Intelligence layer directly unless you’re firmly in the Pioneer category.

But every AI product you’ve ever used has an Intelligence layer underneath. It’s the engine and everything else is built on top of it.

2) Harness

This is the layer that makes Intelligence useful.

A harness wraps the raw model with tools, memory, file access, and goals. It turns a text predictor into something that can act.

Without a harness, the AI can answer questions. With a harness, it can write code across your entire codebase, manage your email, reorganize your files, generate designs, or run background tasks while you sleep.

The harness is opinionated.

Every harness reflects the decisions and priorities of whoever built it.

Claude Code is Anthropic’s harness. It’s opinionated about deep codebase reasoning, autonomous file editing, and long-running agentic loops. Anthropic decided those capabilities mattered and built accordingly.

Codex is OpenAI’s harness for coding and background execution. You fire off a task and come back later to review the pull request.

OpenClaw is an open source harness that’s opinionated about being personal, lightweight, always-on, and connected to everything... your messages, calendar, smart home. Different philosophy entirely.

I bit the bullet and bought a Mac Mini and got OpenClaw 🦞 set up to continue experimenting with AI, but still tinkering and learning. Will share more on this later...

The Harness layer breaks down into a few categories.

If the Intelligence layer is the engine, the Harness is the drivetrain. It decides what the engine can actually do.

This matters because when a tool feels wrong, it’s usually the things layered on top. The harness + prompts.

3) Surface

This is where you meet the AI. The interface. The thing you see, touch, and interact with.

Surface is where the majority of people’s experience of AI begins and ends. You open an app and type something and something happens. The Intelligence and Harness layers are running underneath, but you might not even realize they’re there.

The Surface layer is also a spectrum, ranging from high visibility to low visibility.

Two products can use the same exact Intelligence model, but the Harness and Surface FEEL completely different.

The surface contains different choices about what to show you and what to hide.

Six months ago, a developer in Cursor or Claude Code and a designer in Lovable or v0 were in completely different worlds. Different tools, different skills, different languages. The developer wrote prompts and code. The designer wrote prompts and pushed pixels. They needed each other.

Those lines are FADING FAST.

The gap is closing.

The designer in Lovable is shipping real React applications without knowing React. The developer in Cursor is generating polished UI without thinking about design systems.

AI is absorbing the hard parts that used to separate them. The harness layer keeps getting thicker, the surfaces keep getting more capable, and the skill that used to be required to move between layers keeps shrinking.

More people are becoming Builders every month. Tasks that required a Pioneer six months ago now only need a Builder. Tasks that required a Builder will soon be achievable by a Consumer. The barrier to entry keeps dropping, and it only goes in one direction.

This is the convergence.

AI is compressing the spectrum, especially for the builder crew.

It’s making everyone on this spectrum more capable of operating at higher layers than they ever could before.

The Pioneer still has more control. The terminal still offers more power. That’s real and it’s not going away.

But the opinions baked into the surfaces are getting good enough, fast enough, that more people can build more things without needing that control.

Real products don’t sit neatly in one box. Most span multiple layers, and that’s by design.

Cursor is a harness AND a surface. It has its own agentic capabilities. Things like tab completion, codebase indexing, multi-file editing. That’s the harness in action.

Cursor also gives you a workspace with a file tree, editor, preview. That’s all surface things.

But Cursor’s real innovation is the harness. The surface is a VS Code fork. They are betting on the harness and surface together with swappable intelligence as a single product.
​
Whereas Claude Code is mostly fixed intelligence on an opinionated harness, with only a terminal surface.

But then on the opposite end of the spectrum, they're working hard on a very opinionated Surface, via Claude Cowork Desktop app, etc., geared towards light builders and consumers.

When you see these layers, it becomes interesting to see how each of these companies are focusing on one or multiple points on this AI continuum across these 3 personas and 3 AI layers.

v0 and Lovable look like a surface, but the harness is thick.

It’s making a thousand opinionated decisions with React, Supabase, Tailwind, how to scaffold authentication, how to structure components. You don’t see those decisions because the surface hides them. But they’re doing the heavy lifting.

This is one of the most useful things the framework reveals. It shows you where a product actually competes.

Every product spans layers. But most products win on one. Understanding which layer carries the real value tells you what the product is actually about, regardless of how it’s marketed.

You’ve probably already found yourself on this spectrum.

Now you just need to decide which direction you want to go.

If you’re a Consumer and you’re curious, pick one Builder surface this week.

Open v0 or Lovable and build something small... a component, a landing page, anything.

See what 75% done feels like. That’s the starting line.

If you’re a Builder who wants more control, experiment with a harness.

Move beyond just prompting little things here and there.

Install Claude Code or Cursor and start actually building real things.

See what it feels like to have the AI operate without a curated surface between you and the work. You might find that extra power is worth the discomfort.

The framework is to create some calm in the storm of AI.

The three layers are:

The three personas are:

The tool names will change next week. I don’t think the layers will.

Find your layer.

Decide your direction.

Ignore the noise.

Happy Sunday,
MDS

PPS: I've been heavily exploring AI tools over the last year and we're gearing up for some serious AI-native curriculum at Shift Nudge. We're already exploring, sharing, and discussing these things on our regular live calls. Apply now if this sounds interesting to you.