… the next step: Spec-Driven Development for programming with AI

A few days ago (previous article ) I described my first steps with Claude Code. But knowing the tool doesn't really solve the underlying problem. When tackling a large task with prompts, the code ends up working, but usually not the way you expected, and above all without knowing very well whether the final logic is the one you had in mind.

That's the problem with vibe coding. It's fine for prototypes, but it falls short if the project grows or more than one person works on it.

It seems the next step is Spec-Driven Development (SDD): writing a clear specification first — what, why, rules, and acceptance criteria — and letting that spec be the source of truth for both us and the AI agent.

This is a summary of the essentials: what it is, what problems it solves, how it fits with TDD and BDD, the tools that are emerging, and when the effort is worth it.

As always, suggestions / criticism are welcome