Design Openbook

Choosing an AI Model for Each Task

How do you choose which AI model to use for different tasks? Use a small model for fixed-procedure tasks, a mid-size model for generation with existing standards, the largest model only for design decisions, and two large models independently for final review.

You do not need the most expensive model for everything. Separate the work of setting the standard from the work of carrying it out.

How to use it

  • Give simple execution (lookup, copy, replace, anything with a fixed procedure) to a small, fast model.
  • For general execution (generating or editing when a spec or design standard already exists), a mid-size model is enough. With a standard in place you do not need a large model.
  • Use the largest model only for design: the work of creating the standard itself, such as structure, strategy and judgment.
  • Give final review to two large models separately, and do not show them each other's results. A model that reviews its own work misses its own mistakes.
  • When running several agents, set the model for each sub-agent yourself. If you do not, the expensive model is copied to all of them.

Prompts to copy

Break down the work

Split this task into steps, and assign each step a model tier using the criteria below.
- Simple execution with a fixed procedure: small model
- Generation or editing with an existing standard: mid-size model
- Design or judgment that creates the standard: large model
- Final review: two large models reviewing independently without knowing about each other
Output as a table: Step / Task / Model tier / Reason
Task: (describe the task here)

All cards (27)