Copy, paste, run
I'm building software again — by talking to GPT-4 and shuttling its code into VS Code by hand.
My workflow this month is absurd, and it's the most productive I've been in years.
I describe what I want to GPT-4 in a browser tab. It writes the code. I select it, copy it, switch to VS Code, paste it, run it. It breaks. I copy the error back into the browser, paste it under the code, and ask what went wrong. It apologizes — always so polite — and hands me a fix. Back to the editor. Run again.
Copy, paste, run. Copy, paste, run.
GPT-4 landed in March and it's a different animal than the one that hooked me in November. It holds a thought. It reasons through a problem instead of pattern-matching its way to something that merely looks plausible. I can hand it most of a file and it keeps the thread.
I'm not a full-time engineer. I'm a founder who has always been close to the build but leaned on a team for the actual typing. That barrier is dissolving in real time. On Sunday I sat down and built a working prototype by myself, in an afternoon, in a language I barely know.
The friction now is purely mechanical — the shuttling between windows, the context lost every time I open a fresh chat, the copy-paste tax. Someone is going to collapse that gap. The model can already write the code. It just can't see my editor.
When it can, this gets serious.