> ## Content Index
> Fetch the complete content index at: https://claudefox.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Update Claude Code: background subagent status lines now ~99% cheaper
- URL: https://claudefox.com/update-claude-code-background-subagents-now-cost-about-99-less-to/
- Published: 2026-11-06T08:57:00.000Z
- Updated: 2026-11-20T03:25:30.000Z
- Description: Claude Code used to re-read a background subagent's whole conversation every 30 seconds to write its one-line status. Per a QbitAI report, Haiku 5.5 now writes it from short notes: about 99% cheaper per update, roughly 5% off a session's usage.
- Author: hupicake
- Tags: Tools

*Updated November 20, 2026: Expanded with what the saving means for usage limits, the follow-up changes in later releases and how to turn progress summaries off.*

If you run background subagents in Claude Code's terminal app, there is a small change worth updating for. While a subagent works, Claude Code shows a one-line progress summary for it, refreshed about every 30 seconds, according to a QbitAI report. Until early November, that report says, each refresh re-read the subagent's entire conversation to write that single line. On long tasks, the context can run to tens of thousands of tokens, read again and again for a status message.

## What changed

According to Claude Code team posts reported by QbitAI (republished by 36Kr on November 5), the status line is now written by Haiku 5.5 from short notes instead of the full conversation, which makes each status update about 99% cheaper. We could not find the team's posts themselves, and the changelog confirms the short-notes change but not the figures, so treat the numbers as reported. Three details from the team's follow-ups matter:

- Only the status line changed. The subagent's real work still runs on whatever model it was using.
- Across a whole session with a background subagent, the team put the saving at roughly 5% of your usage limit, or about 6% more prompts on the same quota. The 99% applies to each status update, not to your bill.
- It affects the command-line tool only. Lydia Hallie, who works on Claude Code developer experience, said the desktop app never generated these subagent status lines.

The team said the change needs Claude Code v2.1.316 or later, so check which version you are running first. The changelog refers to a built-in updater command:

```
claude update
```

## The follow-up releases

The public changelog does not mention the change in the 2.1.316 entry itself, but later releases show the same idea being rolled out and refined:

- 2.1.318: on Amazon Bedrock, Google Vertex AI, Microsoft Foundry, or with telemetry off, a progress summary on a cold prompt cache can wait up to five minutes for the subagent's pending response.
- 2.1.321 and 2.1.323: fixes for summaries that ran a request to its end after a tool call, and for summaries skipping an update when a reply took 10 to 20 seconds.
- 2.1.323: a new setting, CLAUDE\_CODE\_DISABLE\_AGENT\_PROGRESS\_SUMMARIES, available since 2.1.318, turns off the summaries and the requests that write them.
- 2.1.327: summaries on Bedrock, Vertex AI and Foundry are now written from short notes rather than the whole conversation.
- 2.1.328: on the Anthropic API, summaries are written by the default small fast model from short notes even with telemetry off.

Read together, the changelog shows the short-notes approach reaching Bedrock, Vertex AI and Foundry sessions in 2.1.327, and Anthropic API sessions with telemetry off in 2.1.328\. 2.1.328 also fixes a regression from 2.1.327 in which effort could be turned off for a subagent's model for the rest of the session if Bedrock, Google Cloud's Agent Platform or Foundry refused a progress summary's effort setting. If you use a cloud provider or run with telemetry off, 2.1.328 is the earliest release the changelog shows with the change for your setup.

## Turning summaries off

If you do not watch the status lines at all, you can skip them and their requests entirely: set CLAUDE\_CODE\_DISABLE\_AGENT\_PROGRESS\_SUMMARIES to 1 before launching Claude Code. In a macOS or Linux shell that looks like this; add the export line to your shell profile if you want it to persist:

```
export CLAUDE_CODE_DISABLE_AGENT_PROGRESS_SUMMARIES=1
claude
```

The changelog says this turns off the progress summaries and the requests that write them; it does not describe any other effect. Expect to lose the live one-line view of what each background subagent is doing.

## Why this matters

Background subagents run concurrently while you keep working, and Claude Code surfaces their permission prompts in your main session, naming the subagent that is asking; foreground subagents block the main conversation until they finish. Every bit of overhead per background subagent multiplies with how many you run and how long they take. A 5% saving on a usage limit is not dramatic, but it is free, and it came from fixing housekeeping rather than the main task.

## Our take

The lesson travels beyond Claude Code: measure what your agent setup spends on bookkeeping, not just on the work. Status updates, summaries and progress checks are ideal jobs for a small, cheap model working from compact notes, and handing them a full transcript is an easy mistake to make in your own agent harness too.

Source: [Claude Code re-read the entire conversation every 30 seconds to write one status line (36Kr, via QbitAI)](https://eu.36kr.com/en/p/4056255491183745?ref=claudefox.com), [Claude Code changelog](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md?ref=claudefox.com), [Create custom subagents (Claude Code Docs)](https://code.claude.com/docs/en/sub-agents?ref=claudefox.com)

This story first appeared in the [claudefox daily briefing for November 6, 2026](https://claudefox.com/daily-2026-11-06/).