signalDEV Community2026-09-24
Claude Code Subagents Were 48% of My Bill. Their Output Was 0.9%
Subagents in Claude Code consumed 48% of the author's bill while output was only 0.9% of total tokens, because each agent request re-sends the full ~51K-token context. Analysis of session transcripts showed 45 sessions over $100 accounted for 79% of spend, and requests above 400K tokens were 54% of main-session cost. The article then offers rules to cap agents, batch small units, trim starting context, and hand off to fresh sessions.
- for who
- Developers using Claude Code subagents or agent workflows
- why now
- Subagent billing reveals hidden input token costs, so slim context to cut AI spend now.
- what changes
- Cost control now targets subagent fan-out and context trimming instead of session length
- to do
- Use the provided script to audit your own usage, then cap agents, batch units, trim context, and hand off sessions.
key points
- Subagents were 48% of bill, output tokens only 0.9% of total
- Each request re-sends ~51K-token context, cost scales with agents times requests times context
- Fixes: cap agents per run, batch small units, trim context, hand off to fresh sessions
#claude code#cost optimization#agent workflow3 sources · confidence medium
score
score 9 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source