Skip to main content
Agent Insights metrics are diagnostic signals. Use them to locate cost and workflow friction, then inspect the repositories, projects, sessions, and team context behind the change.

​
Agent behavior and context health

Token waste causes and AI context health in Agent Insights
  • Compaction shows context-management pressure.
  • Repeated edits can indicate rework or an iterative task.
  • Failed retries can point to ambiguity, missing permissions, tests, or tool failure.
  • Interrupts may be corrections, deliberate stops, or blocked work.
  • Context pressure and floor drift help identify sessions carrying too much or stale context.
None of these signals is automatically bad. The task and repository determine what healthy behavior looks like.

​
Efficiency

Efficiency summarizes how effectively a session turns context and tool use into progress. It is not a developer-quality score. When efficiency changes, compare:
  1. The same date range and source coverage
  2. Repository and task mix
  3. Tool and model changes
  4. Retries, interruptions, and context pressure
  5. Delivery outcome and developer feedback

​
Spend

Agent cost efficiency and spend trend over time
Use spend views to compare members, tokens, sessions, average efficiency, and cost trends. Provider reporting, cached tokens, pricing changes, and incomplete coverage can affect comparisons.

​
Activity and adoption

The session activity map can be filtered by repository and tool, then grouped By Members or By Team. Capability adoption shows how widely workflows such as sub-agents, planning, skills, MCP tools, and task tracking are used.

​
Before sharing a number

  • Confirm the date range and timezone.
  • Check the last sync or provider update.
  • Resolve unmapped identities and unattributed spend.
  • Note missing agents or repositories.
  • Open the underlying view instead of copying a total without context.

​
Next step

Use budgets and outcomes to connect spend with work, or review privacy and governance.