Ladies and gentlemen, I challenge you to a vibe coding challenge, so you can find out for yourself whether this artificial intelligence is something we can actually use in our craft, or whether it’s all nonsense.
First, take a project you shelved and left sitting for months or years. The side project. That thing you always had in your head but never built. Take that idea. Now make three separate folders. Grab one of those agents you can run straight from the terminal, Claude Code, Codex, Open Code, one of these. We’re going to try to build this same project three different ways.
One shot, then a conversation, then a spec
The first step is to do it with a single prompt. One-shot the side project. Go into the first folder, unleash the beast, and tell it what you want. The prompt can be as long as you like, but it has to be one shot.
In the second folder we do it conversationally. Put the agent in the folder, start explaining what it has to do, and correct it as many times as you want inside a single conversation.
Step three is spec driven. Go into the folder and tell the agent: now let’s plan the project, let’s break it into pieces. For each piece we write a markdown file with the specs. We write the spec. If you want you can use something like spec kit, but there’s no need. You have it write the markdown, and then you let it work through the pieces, chatting along the way to keep things on track.
After that, compare the three results and come tell me how much you enjoyed programming with this artificial intelligence. Whether it’s all garbage, or the breakthrough of the century, or maybe whether we need to have a slightly smarter conversation about how we work.
