What is prompt chaining?
Prompt chaining is the technique of splitting a complex task into a fixed sequence of smaller prompts, each handling one step, with each output feeding the next. A single prompt that asks a model to extract facts, compare them against a policy, and draft a response tends to do all three moderately well. A chain gives each step a focused prompt, and application code sits between steps to validate the output, branch on it, or stop for human review.
Unlike an AI agent, a chain follows a sequence designed in advance, and the model does not decide what to do next. That predictability suits regulated and high-volume processes, and agentic workflows often contain chains.
Prompt chaining in practice
A hospital billing office processes insurance denial letters. Step one extracts the claim number, denial code, and payer as a structured output, and code looks up the claim in the billing system. Step two classifies the denial reason against the office’s appeal playbook. Step three drafts the appeal from the matching playbook section and the clinical notes, and a billing specialist reviews it before sending. When step two returns an unknown category, the chain stops and routes the letter to a person.
What to watch for
- Validate every step’s output in code before passing it on, because one wrong extraction corrupts everything downstream.
- Log each step’s input and output so a bad result can be traced to the step that failed.
- Each step adds latency, so measure end-to-end time against what the process can tolerate.