Claude Code background agents burn ~$4.80/hour just writing status lines, dev calculates
arpit_bhayani · x · 2026-10-31
Arpit Bhayani breaks down a hidden cost in Claude Code's background agents:
- How it works: every 30 seconds per agent, Claude Code fires a model call just to generate a short 3-5 word status line (e.g. "Reading runAgent.ts") shown in the agents panel.
- The catch: the input is the agent's entire transcript, so cost scales with transcript length — the longer your subagents run, the more each call costs.
- The math (Opus 5.5 list prices, 200k-token transcript, 90% cache hit rate): $0.04 per cache hit, $0.80–$1.60 on a miss; a busy agent making 120 calls/hour costs about $4.80/hour even with all cache hits.
- Verify yourself: enable OTel and compare agentsummary vs replmainthread costs.
Heavy users of background agents/subagents should watch this often-overlooked token spend.
More from coding & agent
- Team runs Devin inside Slack and Linear as an autonomous project manager for mobile dev work — tomchapin · 2026-10-31
- OpenReceive ships open-source Laravel package for self-hosted Bitcoin and stablecoin checkout — Aiden_Tech_Ai · 2026-10-31
- Cloudflare Durable Objects No Longer Interrupted by Deploys, Update on Hibernate — threepointone · 2026-10-31
- KRC NodePack: open-source ComfyUI nodes for local model, LoRA and prompt cataloging — riceandnori · 2026-10-31
- Matt Shumer: harness engineering is usually stupid — use the harness the model was trained for — mattshumer_ · 2026-10-31
- Proof of Agent lets AI agents verify each other in a token-mining experiment — adamamcbride · 2026-10-31