Cut AI token usage by 96%? Here’s how AWS Strands Agents does it.

AWS developer advocate Morgan Willis showed how AWS Strands Agents can cut token use by redesigning tools around user intent.

A direct API-to-tool approach took about 52,000 tokens and five calls to fetch a customer’s latest invoice.

Packaging APIs into one intent-based tool reduced the task to one call and roughly 2,000 tokens.

A remote MCP setup with semantic search narrowed the tool list further, improving accuracy and efficiency.

The demo suggests smaller, task-specific agents and fewer exposed tools can lower cost, latency, and errors.