Prompt chaining
A fixed sequence of model calls where each step works on the previous step’s output, with checks in code between the steps.
Prompt chaining splits one task into a fixed sequence of model calls. Each call gets the previous call’s output as input and does one smaller job: extract, then classify, then draft, then translate. Between steps your code can run a gate, a programmatic check that stops or retries the chain when an intermediate result is wrong. It is the simplest agentic workflow pattern.
Anthropic’s Building effective agents names the trade: you give up latency for accuracy, because each call becomes an easier task. A chain takes more calls and more time than one big prompt. In return every step can be tested on its own with evals, logged, and moved to a cheaper model where the step is routine (model routing). The gates carry most of the value: a JSON schema check, a length limit, a required field, a rule that the summary mentions every invoice number. Without gates, an early mistake flows silently into every later step.
Use a chain when the steps are known in advance and the same for every input, such as document intake or a content pipeline. If the number or order of steps depends on the input, you need an orchestrator or an agent. If some steps don’t depend on each other, run them in parallel (fan-out and fan-in).
Sources
- Anthropic: Building effective AI agents anthropic.com
Related terms
- Agentic workflow (workflow vs agent): A process where your code fixes the steps and a model does some of them, as opposed to an agent that decides its own steps.
- Fan-out and fan-in (parallelization): Splitting work into parallel model calls or subagents (fan-out) and merging their results in one step (fan-in).
- Evaluator-optimizer (maker-checker): One model drafts, another checks the draft against stated criteria and returns feedback, in a loop until it passes or a round limit stops it.
- Orchestrator-workers pattern (supervisor): One model plans the task, hands parts of it to worker models or subagents, and combines their results.
- Plan-and-execute: A planner model writes a step-by-step plan, executor models carry out the steps, and the planner revises the plan when a step fails.
- Evals: A repeatable test set that scores an AI feature, so changes can be measured instead of guessed.
- Model routing and fallback: Sending each request to the model that fits it, and to another one when the first fails.
