Claude Code Cuts Background-Agent Token Waste

This is an unglamorous but important fix: status visibility should not quietly consume a meaningful share of an agent’s budget.

• –The savings apply to each progress-summary call, not the underlying subagent’s coding work.
• –Anthropic previously acknowledged that summaries averaged about 2% of cost overall and roughly 6% in terminal usage, with heavier multi-agent sessions seeing much more. [Anthropic clarification](https://news.ycombinator.com/item?id=50284699)
• –Short-note summarization preserves the agent panel’s usefulness while avoiding repeated full-transcript processing.
• –Lower monitoring overhead makes long-running background agents more practical for parallel development and automated workflows.
• –Developers should still distinguish this optimization from the separate local status line, which does not consume model tokens. [Technical analysis](https://stackness.dev/blog/claude-code-status-line-token-usage-agent-summary-and-startup-overhead)