ECO-AIM-AI-001 Examples

Pattern

A service sends large repeated prompt boilerplate and unnecessary context on every inference request.

Why this is inefficient

• more tokens than needed
• higher latency
• higher cost
• unnecessary compute

Better

Cache or template stable prompt sections and send only the variable content where possible.

Tradeoffs

• caching adds some implementation complexity
• prompt trimming should not remove important context