token optimization

BMAD achieves 70-85% token reduction through persona/role-based context structuring—specific evidence of token management as CE technique

Anthropic's definition centers on 'optimizing utility of tokens passed into context window'—this is the measurable optimization problem at the heart of context engineering

Core insight is that agent-native interfaces reduce token overhead. This is direct token optimization through interface redesign.

Moves beyond simple compression metrics to timing-aware compression, optimizing for coherence not just token count

Explicitly mentions saving context space through queuing and avoiding repeated permission dialogs, which directly optimizes token usage.

Parallel processing of summaries and single synthesizer call is a token-efficient pattern for handling 100 sessions without balloonng context.