Yep — asking the model to reason before it does something makes a big difference.

→ View original post on X — @mattshumer_