Tag: cost-optimisation

Empirical analysis of token consumption in LLM-based multi-agent systems reveals that 59.4% of tokens go to code review, not generation — and a 2:1 input-to-output ratio exposes the 'communication tax' haunting agentic workflows.