Skip to main content
Budget shows where AI spend is going and whether projects are staying within their monthly limits. Outcomes shows what kinds of work that spend supported. Use the two views together: find a cost change in Budget, then inspect the work behind it in Outcomes.

Open Budget

Track monthly budgets, run rate, untracked spend, and project-level burn.

Open Outcomes

Break spend into work outcomes and drill into repositories, projects, members, teams, themes, and tasks.

Before you compare the views

Use the same date range, timezone, team, member, repository, project, and agent filters in both views. A different filter or stale sync can make a budget change appear unrelated to its outcome mix.
Spend is attributed from supported coding-agent sessions and available project context. Totals can change after identity mapping, project mapping, provider updates, or late-arriving session data.

Budget

The Budget view answers three questions:
  1. How much are we spending now?
  2. Which projects are likely to exceed their monthly budgets?
  3. How much spend is not governed by a project budget?
Budget dashboard overview with spend breakdown, weekly pulse, and spend distribution

Read the spend summary

The Spends Breakdown cards provide a fast health check. Two additional cards expose cost that can be easy to miss:
  • Untracked Spend shows spend that is unallocated or belongs to projects without a budget. The percentage indicates how much of total spend is outside budget coverage.
  • Weekly Spend Pulse shows the latest weekly spend, its week-over-week change, and the project that moved the most.
A large untracked percentage is a governance gap, not necessarily waste. Map the sessions to the correct project and set budgets before deciding whether the spend is excessive.

Inspect spend distribution

Spend Distribution shows total spend divided across projects. Use the donut and ranked project list to identify concentration: a project can dominate overall spend while remaining within its own budget. The distribution includes all visible projects, including projects without budgets. Compare the largest slices with Untracked Spend to distinguish high intentional investment from missing budget coverage.

Use the project breakdown

The Projects Breakdown table is the operational view for budget owners. Filter it by:
  • All — every visible project
  • Over budget — projects above their configured monthly budget
  • On track — budgeted projects that have not crossed their limit
  • No budget — projects receiving spend without a configured budget
Search for a project or sort the table by spend. Each row can include:
Projects Breakdown table with spend, budget status, members, sessions, and activity
Projects without a budget show Set budget instead of budget variance. Add a budget only after confirming that the project mapping and owner are correct.

Drill into a project

Expand a project row to compare:
  • total project spend;
  • the percentage of budget consumed;
  • weekly actual spend;
  • the configured budget line; and
  • the recommended weekly pace or reduction shown in the chart.
Use the weekly chart to distinguish a one-time spike from a sustained change. If Slack delivery is configured, Set Reminder can send a cost digest so the owner does not have to monitor the dashboard manually.
Expanded project budget detail showing total spend and budget versus actual by week
1

Confirm attribution

Check the project name, members, sessions, and task counts. Resolve unmapped or misattributed sessions first.
2

Find the change

Compare weekly bars with the budget line and locate the week where spend accelerated.
3

Open Outcomes with matching filters

Keep the same date range and project filter, then inspect which outcome classes and themes grew.
4

Choose an intervention

Correct mapping, add a missing budget, reduce retries, improve context, adjust routing, or change the budget only after the cause is clear.

Outcomes

The Outcomes view connects spend with the work performed in supported coding-agent sessions. It is intended for understanding the shape of work, not for measuring individual performance.

Read the outcome breakdown

Outcome Breakdown starts with total spend and divides it into four classes.
Outcome Breakdown with total spend, outcome categories, task counts, and feature trend
Each card shows spend, share of total spend, and task count. A high task count with low spend can mean many small tasks; a low task count with high spend can point to a few expensive themes that deserve inspection. The trend chart shows how the selected outcome share changes over time. Use it to determine whether a shift is persistent or limited to a release, migration, incident, or evaluation period.
Outcome classes are derived from available task and session context. They are directional analytics, not accounting codes or a quality score. Validate representative tasks before using a category in financial, staffing, or performance decisions.

Find where spend happened

The Where it happened? section can group the same outcome mix:
  • By Repos
  • By Projects
  • By Member
  • By Team
Outcome spend grouped by repository with stacked outcome bars and total spend
Outcome spend grouped by project with stacked outcome bars and total spend
Each row combines a stacked outcome bar with total spend. The bar explains the mix; the spend column explains the scale. For example, two repositories may have similar percentages but very different financial impact. Use the grouping that matches the decision:
  • choose By Repos for codebase or platform questions;
  • choose By Projects for budget ownership and delivery planning;
  • choose By Member for enablement conversations and data-quality checks, not ranking;
  • choose By Team for operating-model and investment discussions.

Inspect themes and tasks

Open a theme to view Theme Details. The detail panel can include:
  • a generated theme name and summary;
  • total cost and change in activity;
  • repositories represented;
  • total tokens;
  • task count;
  • member count;
  • the theme date range; and
  • a session-activity timeline.
Theme Details panel with cost, tokens, tasks, members, activity, and tasks ranked by cost
Below the summary, tasks are ranked by cost. Each task can show its title, generated description, agent or model source, token count, and spend. This is the most useful level for validating whether a category or theme accurately represents the underlying work. When a theme is expensive, review several tasks rather than only the highest-cost one. A single long evaluation run, repeated retry loop, or generated report can distort the theme total.

Investigate a spend change

1

Locate the budget signal

Start with Projects Over Budget, Weekly Spend Pulse, or a project whose run rate is accelerating.
2

Match the scope in Outcomes

Apply the same time, project, repository, team, member, and agent filters.
3

Identify the outcome shift

Compare outcome shares, task counts, and the trend chart. Look for a category that grew faster than total activity.
4

Find the source

Switch among repository, project, member, and team groupings to isolate where the change occurred.
5

Validate representative tasks

Open the dominant themes and inspect several high-cost and typical tasks, including their descriptions, tokens, and session activity.
6

Act on the cause

Improve project mapping, repository context, task scoping, review policy, agent configuration, model routing, or budget coverage as appropriate.

Common interpretation mistakes

  • Treating untracked spend as overspend. It may reflect missing project mapping or a project with no budget.
  • Comparing a trailing four-week actual with a calendar-month budget without noting the difference.
  • Using outcome percentages without total spend or task counts. A category share can rise while its dollar amount falls.
  • Assuming fixes and rework are always negative. Planned maintenance and incident response can legitimately dominate a period.
  • Ranking people from member-level spend. Role, task complexity, tooling, data coverage, and model choice all affect cost.
  • Changing budgets before validating themes and tasks. Budget variance is a prompt to investigate, not proof of inefficiency.

Pair with Model Router

Agent Insights explains how supported coding agents are used. Model Router controls routing for supported traffic. After a routing change, compare spend, sessions, outcome mix, task quality, and developer feedback before deciding whether the change worked.

Next step

Review Agent Insights metrics and data for metric definitions, members and skills for adoption workflows, or the metrics glossary for cross-product terminology.