How to cut Claude Code usage lost to background-agent summary calls
Each actively working background agent in Claude Code gets a summary request about every 30 seconds, sending its whole transcript to its own model just to write a 3-5 word label for the agents panel. A video from Better Stack walks through the finding and a fix: a Bun preload script from Flamerix AI that stops the summary timer without modifying the Claude Code binary.
A video about it went up 5,512 views in 4 hours, from 32,095 to 37,607 - about 7.3 times the typical pace of other AI-tool videos SteadyVane tracks on YouTube. It has 299 likes and 40 comments, with 37,607 views about 17 hours after it was published.
Why it matters: Better Stack's OpenTelemetry test attributed 12.2% of one small session's cost to these summary requests and 8.2% in a parallel web-development test. If you run several background agents, that is worth measuring in your own usage before you change anything.
