What is prompt chaining?

Prompt chaining means breaking a task into a sequence of smaller prompts, where the output of one becomes the input to the next. Instead of asking the model to do everything in a single giant prompt, you split the work into steps — draft, then critique, then rewrite — and pass the result down the line. Each step does one clear job, which makes the whole thing more reliable and far easier to debug.

Why not just use one big prompt?

You can, and for simple tasks you should. But when a request has several distinct stages, a single mega-prompt asks the model to juggle all of them at once, and quality drops — it forgets an instruction, blends the steps together, or fumbles the hardest part. Chaining gives each step the model's full attention and a clean input.

When something goes wrong, you can see exactly which link failed and fix just that prompt instead of rewriting the whole thing.

What does a chain look like?

Say you want a polished blog post from a rough idea. A chain might run:

1. Outline — "Turn this idea into a five-point outline."
2. Draft — feed the outline in: "Write a draft from this outline."
3. Critique — feed the draft in: "List three concrete ways to improve this draft."
4. Revise — feed the draft and the critique in: "Rewrite the draft applying these suggestions."

Each prompt is short and focused, and the output visibly gets better at every hop. You control the flow, so you can inspect or edit the result between any two steps.

How is chaining different from an agent?

A chain is a fixed pipeline that you design in advance — the steps and their order are decided by you. An agent decides its own next step at runtime, choosing what to do based on what it sees, and can loop or use tools without a script. Chaining is predictable and easy to reason about; agents are more flexible but harder to control.

Plenty of solid systems are just well-built chains, and it is usually the right thing to try before reaching for full autonomy.