Claude Code Background Subagents Cut Status Update Cost by 99%

Anthropic has changed how Claude Code, its command-line coding agent, writes the short status line it shows for each background subagent, a helper agent that keeps working on a task while the main session carries on. According to ClaudeDevs, the company's developer account, a running subagent refreshes that line every 30 seconds, and until now each refresh re-read the subagent's whole conversation. Haiku 5.5, Anthropic's small model, now writes the line from short notes, which the post says makes each update cost about 99% less, so background subagents use less of a user's usage limit (ClaudeDevs on X).

The cost of those updates drew wide attention last week. Flamerix AI documented that about every 30 seconds Claude Code was sending each background subagent's full transcript to that subagent's own model just to fill a 3-5 word status line in the agents panel, and that in Flamerix's own telemetry these calls took 16.4% of Claude Code's estimated spend in one week, slightly more than the main conversation (earlier coverage). On a subscription that spend comes out of rate-limit headroom, not a bill. A commenter who said they work on Claude Code replied on Hacker News that only background agents that are actively working get these calls, and that across users who send telemetry they were about 2% of cost at API list prices the previous week, and about 6% in the terminal, where nearly all of them happen. The same comment said the team was working on "a much cheaper mode that sends short notes to a smaller model instead of the full transcript." That description matches today's post, though the post does not link the two.

The post gives only a per-update figure, so how much of a usage limit this frees depends on how much those updates were taking. An account where status updates were 2% of spend gains far less than one where they were 16%, and an update that costs 99% less still costs something every 30 seconds. The text of the post does not say how the 99% was measured, which Claude Code version carries the change, or whether it applies on every provider and plan. Users who export Claude Code's OpenTelemetry data can check the effect on their own account by watching agent_summary requests, the label Flamerix used to find these calls. Flamerix reports that the export is the only place these calls appear, because they are not written to the session logs on disk, so cost tools that read transcripts miss them.