What Claude Code's 30-second agent summary call costs, read from the binary (theainewsreport.com)

Flamerix AI recently highlighted a cost analysis involving Claude Code's background agents, noting that the tool generates a brief status summary about every 30 seconds by sending the full transcript of each agent's activity to its model. This process, while efficient for real-time updates, is significant for users as it accounts for a notable share of operational costs—measured at 16.4% of a typical week's expenditure, surpassing the main conversation's 15.7%. The costs are primarily driven by the size of the agent transcripts and the frequency of cache misses, which can raise expenses significantly. The underlying mechanics show that while cache hits on the summaries might seem economical, the repeated billing for large transcripts combined with idle time not being optimally utilized leads to greater expenses. There are no supported switches to disable the summary feature directly, and existing cost management tools may fail to track these requests accurately, prompting users to utilize OpenTelemetry for precise measurement. This situation calls into question the sustainability of running multiple background agents, particularly for teams relying heavily on the capabilities of Claude Code in their workflows.