Dev.to
6/30/2026

The Spec Was Never the Good Part
Short summary
Specs-first workflow treats AI as a feature-typing machine, but its real strength lies earlier—in real-time planning conversation where it pushes back on incomplete ideas. The author argues critical errors get baked in during batch spec generation; conversational planning catches them at the branch before cascading. They introduce an adversarial-thinking rule: AI actively challenges risky assumptions during design, raises one objection at a time, and stops once the user commits to a decision.
- •Specs are records of thinking, not the thinking itself
- •AI's real value is in real-time debate during planning, not spec validation
- •Adversarial rule: AI challenges risky assumptions during design, then stops when user commits
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



