Language: English

Community post

Break a Large Request Into Smaller Prompts

By · Published · 2 min read

Editorial standards and corrections

A practical workflow for splitting a large AI request into planning, building, and checking steps.

One huge prompt can make the model plan, write, review, and present at once. If the output fails, I cannot tell which part caused the failure. I prefer a short chain of tasks with a check between them. For a software feature, Prompting for a Small Feature: Requirements Before Code gives you a concrete requirements checkpoint before asking for code.

Example: a small game feature

Suppose I want to add a cooperative inventory system. “Build my inventory system” skips decisions about what players can carry and what happens when both want the same item. I split the work: Define behavior: “List the player actions and edge cases for a shared two-player inventory. Ask only questions that change the behavior.” Write acceptance tests: “Turn the agreed behavior into six pass/fail scenarios. Include a full inventory and simultaneous pickup.” Implement one slice: “Write only the pickup and capacity-check logic for the supplied language and data shape.” Review: “Compare the code against the six scenarios. Report mismatches with evidence; do not rewrite yet.” Each step produces an artifact the next step can use. If the first list is wrong, I fix it before it becomes code. If the code fails a test, I revise that slice rather than restarting the system.

Keep the chain connected

The danger is that each prompt forgets the earlier decisions. Carry forward a short approved specification, not the entire conversation. Name what is fixed and what remains open. For example: “Use the approved six scenarios below; do not introduce trading.” Google’s prompt design guide discusses breaking complex tasks into simpler components. The useful principle is not “always use more prompts.” It is to create checkpoints where mistakes can be caught. Try it: Pick a request that would take more than one afternoon. Split it into three outputs: a decision, a draft, and a check. Write one acceptance criterion for each and stop at the first failed checkpoint. Keep learning: Prompting for a Small Feature: Requirements Before Code, How to Test a Prompt on More Than One Example, and A Tiny Evaluation Set for Your Favorite Coding Prompt.

About the author

Coder and gamer. I test prompts, share what works, and show how to improve AI results for code and creative projects.

Comments (0)

Loading comments…

Keep exploring

All articles