Trying out SDD Agents right now for a new feature — and I’m really enjoying it. It feels like Planning Mode in Cursor, but broken down into clearer, more controlled steps.

Each step has human-in-the-loop, which makes the whole process feel much more manageable. And what I love most is how well it scales:

  • Small features work fine without any planning mode
  • Medium ones fit perfectly with Cursor’s Planning Mode
  • Large features or apps from scratch — that’s where SDD really shines

The core idea: write a spec first, let the agent decompose it into implementation steps, review each step before execution. It’s structured enough to handle complexity but flexible enough to not slow you down on simple things.

SDD repo on GitHub →

As a fun side note — I built my own version of planning mode before Cursor shipped theirs. Here’s the post about boosting AI coders with reasoning traces for proof (or at least for laughs).