Everyone’s talking about vibe coding — just describe what you want to an AI and let it build.
I think that’s going to work about as well as “just describe what you want to a developer” did before anyone invented project management.
The AI tools are real. The productivity gains are real. But the assumption that process stops mattering because the labor is cheaper? That part I don’t buy.
What I find more interesting is this: different AI models are genuinely better at different things. The model that writes the tightest specification isn’t necessarily the best at generating code. The best code reviewer isn’t the same as the best architect. And none of them are ideal for holding an entire large codebase in context simultaneously.
That’s a resource allocation problem. And resource allocation problems have been solved before.
I mapped the strengths of the major AI models against the stages of a standard development lifecycle — design, specification, architecture, implementation, QA, security, documentation — and built a methodology around the fit. Different models assigned to the stages where they perform best. A human in the loop at every gate.
I’m calling it MMMBOP. Full write-up here if you want the details: http://projects.small.to/MMMBOP.pdf
I’ll be running a real application through it this summer and documenting what actually happens.
If this is relevant to something you’re thinking about, I’d enjoy the conversation.
