A Claude developer outlines four types of ambition for AI use: speed, altitude (stating desired outcomes rather than individual tasks), depth (applying field-expert tools and methods), and range (drawing on multiple disciplines).

The post uses a quarterly business review deck as a running example, showing how prompts escalate from filling in template numbers to redesigning the review process entirely. It also recommends insisting on raw data, adversarial assumption-checking, outside validation, and setting performance targets the model initially considers unreasonable.