# Subagents

Compare delegated agents by issues, tool failures, orchestration, run volume, and duration.

The **Subagents** dashboard answers two questions: which delegated agent needs attention first, and what is going wrong inside it. Select a trace service and a date range to compare the named subagents Brizz observed.

A subagent appears here after its work has been [marked as a subagent](/docs/instrument/subagents.md). Unmarked work stays with the main agent and is not assigned to a subagent row.

## How to read it

The page opens with a suggested starting point, then shows the evidence behind it:

- **Top issues coming from a subagent** lists issues whose findings were detected inside that subagent's conversation items, ordered by severity and reach.
- **Issues by subagent** compares blocking, major, and minor issue counts with distinct affected sessions and users, tool-call failure rate, and run volume.
- **Which tools are failing, and for whom** compares the same tool across subagents. If one subagent fails a shared tool much more often than another, investigate that caller before blaming the tool itself. Dot size represents call volume.
- **Delegation** reconstructs who handed work to whom. A subagent can appear beneath more than one parent when the selected traffic contains both relationships.
- **How long a run takes** ranks subagents by p95 duration, with p50 alongside it, so a slow tail is not hidden by a fast typical run.

Click a subagent anywhere on the page to open its drilldown. The drawer shows the issues detected inside it, its tools with call and error counts, and recent runs with outcome, duration, start time, and session reference.

## Important limits

**Inside does not mean caused by.** An issue is attributed when its finding lands on a conversation item belonging to the subagent. The dashboard is showing containment, not claiming that the subagent caused the issue.

Tool failure rates use only calls with a recorded outcome. Fewer than 20 attributed calls is shown as **—**, because the sample is too small to rate. Cost and token totals are omitted: only some LLM spans can currently be attributed to a subagent, so presenting them would undercount by an unknown amount.

All results are scoped to the selected service and date range. Changing either can change the subagent list, issue attribution, delegation tree, and run statistics.

## Availability

The dashboard is off by default and must be enabled for your organization. Super admins can still open it while reviewing the feature. If **Subagents** is not in your navigation, [contact us](mailto:support@brizz.ai).

## See also

- [Instrument subagents](/docs/instrument/subagents.md) — mark runs and preserve parent-child relationships.
- [Issues](/docs/platform/issues.md) — investigate the issues attributed on this page.
- [Tools](/docs/platform/tools.md) — inspect one tool across all sessions that used it.
- [Sessions](/docs/platform/sessions.md) — read the full conversation and expand each subagent lane.