The Terminal Belongs to Designers Too
I built beautiful interfaces for my AI agents, then looked at what I'd made and took them all down. Here's why and a practical guide to getting you into the terminal.
A few weeks ago I built a UI for one of my designpowers agents. Several, actually. I spent real time on them. Clean, considered, the kind of thing I was excited to share with you. I know that many of you have said you don't love the command line whether that's Claude Code, terminal, Gemini CLI…so I wanted to honour that. Then I looked at what I'd made and felt a sort of unease. You know what I mean, that feeling when you've fallen in love with your own idea and designed the wrong thing.
I’d put the agent in a costume.
I’d taken something that was direct, alive in a particular way, that would show you its thinking if you let it, and I’d dressed it up. Made it presentable. Wrapped its reasoning inside a crisp ui. In doing that I’d introduced distance between the person using it and the thing they were actually talking to and trying to collaborate with. I’d made it more comfortable, predictable and less true. I killed the delight of using designpowers.
In a moment when AI is changing what it means to design, when the agents you’ll be working with and designing for live and breathe in this environment, the fear of the black window is a wall between you and the work that matters most.
I took the UIs down.
This post is about convincing you to give things a go in the little black window.
The GUI was scaffolding
Think about what the graphical interface actually was. Before windows, before menus, before the cursor, you spoke to a computer in its language. Most people couldn’t do that. So we built a translation layer — the desktop metaphor, files that looked like folders, trash cans you could drag things into. You needed to gesture rather than speak. The GUI was scaffolding. Brilliant, necessary, world-changing as an innovation, but nonetheless created around a fundamental limitation so invisibly useful that we forgot what it was for.
That limitation is changing. In some conversations, in some moments, the machine understands you well enough that the translation layer is already overhead. You can just say what you want.
When that’s true, the interface becomes friction.
The browser interface, even a minimal one, is curating what you see, making decisions about how to present it. The terminal doesn’t present. It gives you the thinking as it forms, unmediated. That’s quite a big difference.
Not my business
I know that many designers still feel the terminal is not their business. The terminal belongs to engineers, to people who didn’t come up through Figma and Sketch and the careful craft of making things look and feel right. That’s how it was.
That mindset needs to change.
When you work with an agent in terminal, you see the raw thinking as it emerges. Not the output, polished and placed in a response box, but the thinking itself, the reasoning taking shape in front of you, the pauses, the pivot when it reconsiders, the moment it commits to a direction. You’re watching it think. This is a new intelligence, and right now the most honest way to meet it is without the interface in the way.
Earlier this year, DeepMind researchers David Silver and Richard Sutton published a paper arguing that AI is moving from what they call the “Era of Human Data”, where systems learned by imitating human-generated content, into the “Era of Experience”, where agents learn through direct interaction with the world rather than by imitating us. This week, Google DeepMind showed exactly what that transition looks like in product form. Magic Pointer, the new Gemini-powered cursor, lets you point at anything on your screen and say "fix this" or "move that here", no prompt box, no AI window to detour into. The interface dissolves further. Which is elegant, and it will be everywhere soon.
If that’s where intelligence is going, the interface built around human imitation is already becoming a relic. We’re building another kind of interface entirely.
The terminal is where you start.
What this window is actually for
We are early enough that the agents are still legible. You can still watch one think. You can still feel the shape of how it reasons before it’s been wrapped in chrome and shipped with a logo. You can still learn, by being close to them, what they need from you to produce something good. That knowledge, the texture of how an intelligence works, it is going to be one of the most important things a designer can understand. We are at risk of designing it away too quickly.
Designers are really good at this important way of thinking already. Every time you decided how an error message should make a person feel. Every time you chose what to surface and what to hide. Every time you designed for the person who didn’t fit the assumed user, you were encoding values into a system. That work doesn’t go away when the interface does. It moves upstream, into the reasoning layer itself, and it has no canvas.
That’s design literacy for the world that’s coming.
Get a terminal worth opening
Before anything else, I want you to get Ghostty.
The default Terminal app that comes with your Mac works but I think this is better. Ghostty was built by Mitchell Hashimoto, co-founder of HashiCorp. Free, open source, released in late 2024. It’s the first terminal I’ve used that I really like.
The typography is serious. Ghostty has its own font rendering engine. It supports ligatures, which means certain character combinations display as elegant single glyphs rather than awkward pairs. If you’re going to spend time reading text on a screen, it should be beautiful text. Pair it with a font like Berkeley Mono or Monaspace and the terminal is really nice.
The colour is handled. Ghostty ships with a huge library of colour schemes built directly into the application, with a live preview interface so you can see exactly what you’re choosing before you commit. It also responds to your system’s dark and light mode automatically.
It feels native. On macOS it’s a true SwiftUI application, with native tabs, real window management, proper keyboard shortcuts, and a settings interface that feels like it belongs on your machine.
It’s clean by default. You don’t need to configure anything to get started. Open it and it already looks good.
Download it at ghostty.org. Install it like any Mac app. Then open it.
A window appears. A prompt ending in %. That’s the shell, waiting.
(On Windows: Windows Terminal from the Microsoft Store is excellent, and everything below works there too.)
Your first few minutes
I’ve put together a Terminal Starter Guide for Designers: it’s a one-page reference with every command in this post, plus the full designpowers setup. Download it and keep it open while you follow along.
Type this and press enter:
pwdThe computer tells you exactly where you are on your file system. One question, one answer. That’s the whole rhythm of this place.
lsEverything in your current folder, listed in front of you. You’re reading your file system like a map, without opening a single Finder window.
cd Documentscd means change directory. Type pwd again and you’ve moved. Type ls again and the contents are different. You’re navigating with words.
cd ..Two dots takes you up one level. That’s the core of it. You ask, it answers. No translation layer. No affordances. Just the conversation.
Meet a design agent without the costume
designpowers is a team of 10 AI agents built for design work: research synthesis, critique facilitation, brief writing, the strategic thinking that has no home in any existing design tool. You direct them. They run an inclusive design process, remember your aesthetic taste across projects, and argue competing directions when you’re uncertain.
You’ll need Claude Code installed first — npm install -g @anthropic-ai/claude-code that’s a one-time setup, and worth doing regardless of this post.
Then:
Step 1. Clone the repo.
git clone https://github.com/Owl-Listener/designpowers.gitgit clone copies a project from GitHub onto your machine. A folder appears in your current directory. You’ve pulled a whole design team onto your computer with a single sentence.
Step 2. Move into it.
cd designpowersStep 3. Start a session.
claudeWait a moment. The owl appears. That’s the first sign you’re in the right place.
Now give it something real. Don’t say “hello” or “what can you do.” Say the actual thing you’re sitting with at work. Maybe it’s:
I'm designing onboarding for a health app. My users are people who've had a difficult diagnosis. I want to understand what they need before I write a single word of copy.Or something that’s been nagging at you:
I have two directions for a dashboard redesign and I can't decide which one is best.Whatever the actual problem is. The agent doesn’t need a brief. It needs to work with you: this is the design work.
Watch what happens. The response comes word by word, live, in front of you. You’ll see the agents pause. You’ll see them reconsider. You’ll watch them commit to a direction. This is not the result of the thinking. It is the thinking, forming in real time. There’s no interface smoothing it over, no panel it appears inside.
You can go further:
I want to chat with the motion designer directly. I'm after something specific.I want to know where you all agreed and disagreed with each other during this project.Create a website to wrap up the whole project so we can present it to Tom, the Design Director.This is a new intelligence, and right now the most exciting way to meet it is without the interface in the way. It does not belong shackled behind a UI.
You’ve met the owl. Now look in the other direction.
An AI that reads your pointer, understands your context, acts on “this” and “that” before you’ve written a word. That future is going to feel effortless. It is also going to be completely opaque, shaped by decisions that were made before you arrived, assumptions baked into the reasoning long before anyone asked what designers actually needed from it. You won’t see those decisions. You’ll just work inside and alongside them.
The terminal is the last place where none of that has happened yet. Where you can still watch how the reasoning forms. Where you can still learn what the intelligence needs from you, and be the person who carries that knowledge, rather than the person who never knew where to look.
Stay a little longer and explore.




Hey Carl,
I'm glad you tried designpowers, that's awesome!
Thanks for this excellent question.
It is true that Claude Desktop gets you there for many tasks. It's powerful and has a great interface, but a few things can only happen in Terminal.
A few functional things: you can clone repos, run local code, chain tools together, pipe outputs into other commands, install and tun things that live on your machine. In the example of designpowers, the agents are on your machine, not sending everything through Anthropic's interface.
Philosophically...Claude desktop is still a costume. It makes decisions about what to show you or not, how to format things, what to surface and so on. It's beautifully designed I think, but it's still a translation layer. In terminal you see everything, the whole process and not just the output.
I really like Claude Desktop and I use it a lot. I'm aware that I'm working through an interface that was designed to feel smooth. Sometimes smooth inevitably means opaque.
Terminal literacy isn't really about the terminal. It's about understanding what's underneath any interface you're using. Once you see that, I think you can make better choices about when you want to use an interface and when you prefer not to do that.
Claude code (not the chat or co-work) lives in the terminal, they made it terminal-native by design. Their most powerful product is a CLi not a GUI app. This is because for serious agentic work, you can see the file system, watch commands execute, observe what's happening. Claude code is still a product layer on top of terminal.
I'm so glad you wrote this MC Dean. I felt the same thing when Claude Design came out. The demo was impressive, but using the tool in the browser felt limiting compared to the flexibility and capabilities of Claude Code in the terminal. It felt like the Fisher Price version of the thing.