I nearly drowned over Christmas 2024. Got sucked into a river mouth while surfing, proper scary stuff. So instead of getting back in the water, I spent the next few weeks drinking beers and wine late at night, watching YouTube like a degenerate. That's how I accidentally stumbled into AI coding.
Not "AI" as in asking ChatGPT to write you a birthday message. I mean AI agents that live in your terminal, write entire platforms, and push code to production while you sit there talking to your screen like a madman.
Fast forward to now: I've got two Mac Minis, two portable screens so I can build from anywhere in the world, and AI agents with full terminal access coding and pushing straight to GitHub and Vercel. No fucking hands. I just talk.
I can't code. Never could. Don't need to. And that's the whole bloody point.
This article is the progression I wish someone had laid out for me six months ago. Where to start, what to skip, and how to go from poking around in ChatGPT to running autonomous coding agents that do the building for you.
Get Off ChatGPT. Seriously.
ChatGPT is not the best AI anymore. Full stop.
Look, they were first to the race, and credit where it's due — they made AI mainstream. But everyone I talk to still thinks AI is ChatGPT. It's like thinking the internet is AOL. You're stuck in 2023, mate.
Here's what non-coders need to understand: when you can't code yourself, you don't know whether the code the AI spits out is right or wrong. And quite frankly — you don't need to. That's the whole point. What you do need is to use the right tool, because some AI models are language models designed for thinking and conversation, and some are coding agents designed purely to build.
The difference is massive. A language model will give you code in a nice little chat window that you then have to copy, paste, debug, and figure out where it goes. A coding agent will just... build the thing. In your actual project. With actual files. That actually work.
Different tools, different jobs. Stop using a hammer to screw in a bolt.
Stage 1: Play Around With the No-Code Builders
Start here. No shame in it. I did.
Replit, Bolt, Lovable, Vercel v0 — these are your training wheels. They let you describe what you want in plain English and they generate a working app. You can see it, click around, and start to understand the relationship between what you ask for and what gets built.
The beauty of these platforms is there's nothing to install. No terminal. No IDE. No Git. Just you, a browser, and a text box. Type "build me a landing page for a surf school with a booking form" and watch it appear.
You'll hit limits fast. The customization gets clunky, things break in weird ways, and you start wanting more control. That's the point. That frustration is your signal to move to Stage 2.
But don't skip this step. It builds your intuition for how AI interprets instructions, and that skill — prompting well — is the one thing that matters at every single stage after this.
Stage 2: Get Yourself a Proper IDE
An IDE is just a fancy text editor where code lives. And the one you want is Cursor.
Cursor is the most popular AI-powered IDE right now, and for good reason. It's basically VS Code (the industry-standard editor) but with AI baked into every corner. You can highlight code, ask it questions, tell it to refactor something, or just say "make this work" and watch it try.
For a non-coder, this is where you start to feel the power. You're not copying and pasting from a chat window anymore — the AI is working inside your actual project. It can see all your files. It knows the context. It makes changes and you see them happen in real time.
Install it. Open a project. Start talking to it. Break things. Fix them. Break them again.
This stage is about getting comfortable with the environment. The terminal will still scare you. Git will make no sense. That's fine. Just keep prompting and let Cursor do the heavy lifting while you build your intuition.
Stage 3: Make Friends With the Terminal
Here's where most non-coders tap out. Don't.
The terminal is just a text-based way to talk to your computer. That's it. Instead of clicking buttons, you type commands. Instead of dragging files, you move them with words. It's not magic — it's just different.
Start simple. Learn cd (change directory), ls (list files), mkdir (make folder). That's enough to navigate around. Then learn git add, git commit, git push — that's how you save and share your code.
Here's the thing: once you're comfortable in the terminal, you unlock an entire category of tools that don't exist anywhere else. The most powerful AI coding agents live in the terminal. If you never learn to use it, you're locked out of the best stuff.
You don't need to be a wizard. You just need to not be scared. Open Terminal on your Mac or Linux machine, or install Windows Terminal on PC. Poke around. Google what you don't understand. It gets easier fast.
Stage 4: Unleash the Terminal Agents
This is where it gets wild.
Claude Code, Codex CLI, Aider, OpenCode — these are AI agents that run in your terminal and have full access to your codebase. You talk to them like a colleague. "Refactor the auth system." "Add a dark mode toggle." "Fix whatever's breaking the build." And they just... do it.
Not like ChatGPT where you get code back and have to figure out where it goes. These agents make the changes directly. They create files, delete files, modify files — all while you watch. Some of them can even run your code and debug it themselves.
I call my main agent the Dumpling Bot. It's based on Kimi Moonshot — a Chinese AI that absolutely rips. One night I pointed it at a blank folder and said "build me a recruitment platform." Three hours later, I had a working MVP with auth, database, UI, the lot. I just sat there drinking wine and occasionally answering questions.
This is the unlock. This is what you've been working toward. An AI that doesn't just help you code — it codes while you supervise.
Stage 5: Upgrade to Warp
Once you're living in the terminal, you want a better terminal. That's Warp.
Warp is a modern terminal built for people who actually use it all day. It's got AI built in, so you can ask it questions right there. It's got blocks, so your output is organized instead of an endless scroll. It's got autocomplete that actually works. And it looks good, which matters when you're staring at it for hours.
The free tier is enough to start. Once you go Warp, regular Terminal feels like notepad.
Stage 6: Final Form — No Hands
My current setup: two Mac Minis running 24/7, two portable monitors so I can work from anywhere, and AI agents with full terminal access pushing code to GitHub and Vercel while I talk.
I use voice-to-text to give instructions. I barely touch the keyboard. I'll be walking around the house, talking to my screen, saying things like "add a cron job that checks the recruitment queue every hour" — and by the time I sit back down, it's done and deployed.
This isn't science fiction. This is available right now. You just have to progress through the stages.
The tools I use daily: Claude Code for complex thinking, Cursor for when I need to see the code visually, Warp as my terminal, Vercel for instant deploys, Supabase for the database. Everything connects. Everything flows.
Could I have gotten here faster? Maybe. But the stages matter. Each one builds the intuition for the next. Skip too many and you'll be lost when things break — and things always break.
The Honest Bit Nobody Tells You
You're going to waste money. You'll pay for tools you don't need. You'll start projects you abandon. You'll spend hours debugging something stupid. That's the cost.
You're going to feel dumb. Real developers will say things you don't understand. You'll Google basic concepts for the hundredth time. You'll wonder if you're even allowed to be doing this. Ignore it.
You're going to build things. Real things. Things that work. Things you couldn't have built six months ago because you didn't know where to start. That's the payoff.
The gap between "knows nothing about code" and "ships actual products" has never been smaller. You don't need a CS degree. You don't need to understand algorithms. You just need to talk to the right AI, in the right way, with the right tools around you.
Start at Stage 1. Don't skip steps. And one day, you'll be talking to your screen while robots build your empire.
FAQ
Not really. You need to understand what code does at a high level — what a function is, what a variable is, how files connect to each other. But you don't need to write it yourself. The AI handles that. Your job is to direct, review, and decide.
Cursor is $20/month. Claude Pro is $20/month. Warp is free to start. You could be up and running for $40/month. Compare that to hiring a developer.
For coding: Claude (Anthropic) is the current best. For general thinking: ChatGPT, Gemini, and Claude are all competitive. For autonomous agents: Claude Code or Codex CLI. Try them, see what clicks.
Yes. Everyone is at first. The trick is to just use it anyway. Start with three commands and build from there. Within a week, it'll feel normal.
Yes. I'm doing it. Multiple platforms, multiple revenue streams, no developers on payroll. The tools are that good.
Bottom Line
Six months ago I couldn't code. Now I deploy production apps by talking to my screen.
The path: ChatGPT → No-code builders → Cursor → Terminal basics → Terminal agents → Warp → No hands.
Each stage makes the next one possible. Don't skip. Don't rush. The tools are there. The gap is closeable. You just have to start.
Now get off ChatGPT and go build something.