Multi-agent AI systems waste tokens in two specific, fixable ways. A technical breakdown of phase-scheduled activation, context pruning thresholds, and the patterns I actually use to keep a fleet of subagents cheap.
Opus 5 shipped a five-level effort parameter, and Claude Code now lets every subagent set its own model and effort in frontmatter. A practical guide to mapping your agent fleet to a cost curve instead of a flat rate.
Google's new Flash model jumped from 34.4% to 43.6% on FrontierCode 1.1 at half the price of its predecessor. A technical lead's take on when to route agentic coding work to the fast, cheap tier — and when not to.
Get notified when I publish new posts on AI, .NET, cloud architecture and more.
