HYPE OR FACT? · CHECKED 5 NOVEMBER 2026
Claude Code background subagents: 'about 99% less usage' per status update
OUR VERDICT
Partly factOur original verdict: “Fact or hype: partly fact. The behaviour change is real and announced by ClaudeDevs; the 99% figure is Anthropic's own measurement of status-update token cost, not a guarantee for every job.”
THE CLAIM
“Background subagents in Claude Code used to re-read their whole conversation every 30 seconds for the status line; now Haiku 5.5 writes the status from short notes, cutting each update by about 99%.”
What’s actually true
- Anthropic's ClaudeDevs account described the old polling behaviour and the new Haiku 5.5 short-notes status path.
- Boris Cherny amplified the post, so it is an official product note, not a random rumour.
- The ~99% saving is Anthropic's figure for each status update, not a measured cut to total bill for every long-running job.
Sources
Cost
Included in Claude Code usage; Anthropic says status updates now cost far less of the plan.
USE IT IF
People running long Claude Code jobs with background agents who were burning usage on status chatter.
SKIP IT IF
Anyone expecting a fixed 99% cut on the whole session bill without measuring their own usage.
How we knowChecked on 5 November 2026. Verdicts: Fact = the claim holds up; Partly fact = real, but with caveats or overstated parts; Hype = the headline claim isn't supported. Products change quickly, so check the maker's site before you buy. Our methods & sources ↗