Show HN: Spec-Driven Development Workflow for Claude Code

A new GitHub project is pitching a simple promise: if you make an AI coding helper work from written step-by-step plans, clear its memory between stages, and break big jobs into smaller ones, you’ll supposedly get better results for less money. Sounds neat. The catch? The community immediately went into "show me the proof" mode.

That’s where the real action is. One commenter compared this whole workflow trend to using power tools on wood: flashy, fast, and still somehow leaving you with a rough mess that needs a ton of sanding at the end. Ouch. The complaint wasn’t even that the plan was bad — it was that the AI still acts, in their words, kind of lazy about following it. In other words: the instructions may be fancy, but the robot intern still needs supervision.

Others weren’t against the idea, just deeply skeptical. The biggest hot take was basically: where are the benchmarks? If this method really “boosts performance” and “keeps costs low,” commenters want numbers, comparisons, and receipts — especially versus other planning tools and modes already out there. Another user asked how it stacks up against rivals like OpenSpec and Superpowers, turning the thread into a mini showdown of competing AI productivity rituals.

So the vibe is clear: interesting idea, cool packaging, but the crowd is not handing out applause until the data shows up. In true internet fashion, the comments turned a workflow demo into a trial by skepticism, jokes, and side-eye.