Prompt Chaining — Break Task into Steps

Example: article summarization

Step 1: Extract key points (bullet list)
Step 2: Group points by theme
Step 3: Write summary from grouped points
Step 4: Add title + polish tone

Why chain vs single prompt

Each step has smaller context → cheaper + more focused. Easier to debug (inspect intermediate). Different models per step (cheap for extraction, smart for synthesis).

Failure isolation

Fault at step 3 doesn't corrupt earlier work. Cache/store intermediates → retry only failed step.

When NOT to chain

Simple tasks where single prompt already reliable. Chain overhead > single-shot cost.