Fable's judgement

When working with AI coding agents like Fable, users achieve better results by allowing the model to exercise its own judgment rather than providing rigid, step-by-step instructions. For instance, instead of explicitly defining when to run automated tests, users can instruct the agent to determine for itself whether a specific code change warrants testing. This approach leverages the model's internal logic to handle tasks more flexibly.

This strategy is particularly effective for managing token usage and operational costs. By prompting the agent to delegate smaller or mechanical coding tasks to lower-power models, users can reserve the most capable, high-cost models for complex tasks that require deep judgment, review, and synthesis. This delegation process involves spawning subagents with specific model overrides, such as using lighter models for trivial edits and reserving top-tier models for substantive implementation.

Implementing this workflow allows for greater efficiency and cost savings without sacrificing the quality of the final output. By offloading routine work to smaller models while keeping the main agent focused on high-level decision-making, users can complete more projects while significantly slowing the depletion of their token allowances.