LANDMARK · 8 MIN · USE
Break work into steps
After this landmark, you can decompose a large request into ordered stages with checkpoints, instead of asking for the whole thing in one shot.
Ask a model for a big, tangled deliverable in a single prompt and errors compound silently — a wrong assumption in paragraph one quietly poisons everything after it. Breaking the work into stages — outline, then draft, then check — gives you a place to catch a mistake while it’s still cheap to fix, and lets each step work from a verified starting point instead of a guessed one. It also makes the model’s reasoning legible: you can see where it went off, not just that the final answer feels wrong. The cost is a few more exchanges; the payoff is that a fault shows up at the stage it started, not baked into the end.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
For anything multi-part, ask for the plan or outline first, correct it, then let the model fill it in. Fixing a bad outline takes seconds; fixing a finished draft built on it takes far longer.
MAKE A DECISION
You need a research memo with analysis and recommendations. What gives you the most reliable result?
CARRY THISTake one big ask you’d normally send in one prompt and split it into “plan → do → check.” Approve the plan before the doing.