- 00:00readcheckout/CartSubmit.tsx$0.02
- 00:41editpayments/retry.ts$0.31
- 01:12retrynpm test payments$0.18
- 01:40retrynpm test payments$0.21
- 02:05editpayments/retry.ts$0.44
- 03:30verifytests passed$0.09
Agent Insights
AI agent analytics across spend, usage, waste & outcomes.
Track spend, token waste, outcomes, adoption, budgets, and recommendations across Claude Code, Codex, and Cursor.
Problems with token spend
AI agent bills report totals without engineering context.
01Spend breakdown
Claude Code, Codex and Cursor bill separately. Provider totals do not show spend by developer, project or repository.
- Claude$2,640
- OpenAI$920
- Cursor$676
- Codex$412
02Token wastage
Compaction, interrupts, repeated edits, and failed retries are not itemized in provider bills.
- Compaction1.2M tokens
- Interrupted860K tokens
- Repeated edits640K tokens
- Failed retries120K tokens
03Budget visibility
Project overruns become visible after billing unless agent spend is tracked against project budgets.
Payments API143%Over budget$860 over project budget
Agent Usage Visibility
Know where your agents spend, work and improve.
From token usage and budgets to outcomes, adoption, and workflow recommendations.
- 01Waste
Token wastage
100 tokens spent - Compaction
- Interrupts
- Repeated edits
- Failed retries
Measure compaction, interrupts, repeated edits, and failed retries separately.
- 02Spend
Spend attribution
spend by developer · Sep$1,184 - Emily C.$412
- Ryan B.$318
- Megan T.$264
- Jake R.$190
by modelby projectby repositoryGroup the same spend by developer, model, project, or repository.
- 03Outcomes
Outcome tracking
293 agent tasks · 30 days - Features48%
- Fixes & rework21%
- Tests & reviews13%
- Exploration18%
Classify agent tasks by work type and link them to pull requests and tickets.
- 04Adoption
Team adoption
sessions per day - Emily
- Ryan
- Dana
- Megan
- Jake
- Paige
Track session activity by member or team.
- 05Budget
Project budget
payments-api · Sep budget $1,500 75%$1,190Near limitAlert emailed · crossed 75%
Set monthly budgets for Linear and Jira projects and receive an email alert at a chosen threshold.
- 06Skills
Skill and workflow recommendation
- ECAdd a repo map to CLAUDE.md in web-appCLAUDE.md
- RBUse plan mode before multi-file editsworkflow
- MTTry the team’s /test-plan skillskill
Generates recommendations from each dev’s sessions. Developers see only their own recommendations.
Process
Receive actionable insights from your first session.
- 01Sync
- Claude Code412
- Codex138
- Cursor96
Sync agent sessions
- 02Attribute
- developer
- model
- repo
Attribute cost
- 03Classify
- Features48%
- Fixes & rework21%
- Tests & reviews13%
- Exploration18%
Classify waste and work type
- 04Recommenddo this next - Emily C.Add a repo map to CLAUDE.md in web-app9 sessions compacted twice before the first edit.CLAUDE.mdquick
Generate recommendations
Product tour
Track agent usage from the team to the individual session.
See spend, adoption, efficiency, and outcomes across developers and projects, then drill into the sessions behind them.
Agent Spend Health Overview
Track where every AI dollar goes across models, developers, and projects.
Org Maturity72Proficient
Avg Efficiency68
Cost / developer$70212/15 devs
Error rate4.2%
Agent Session Activity Map
By MembersBy TeamAsk Ellie
- EEmily
- RRyan
- DDana
- MMegan
- JJake
- PPaige
6/12Jul 20Aug 17Sep 14This week
Budget
Track AI spend per project against monthly budgets, and stay ahead of where every dollar goes.
Total Monthly Budget$12,0008 projects
Actual (Month)$8,42170% of budget
Projects Over Budget2Needs attention
Untracked Spend$1,24015% of spend
Projects Breakdown
All 8Over budgetNear limitOn trackAsk Ellie- LPayments API5 members · 214 sessions$2,860143%
- JGrowth Web6 members · 168 sessions$1,42095%
- LData Pipeline3 members · 82 sessions$61051%
- LCheckout2 members · 50 sessions$340Not tracked
| Project | Spent | Budget | Usage | Activity |
|---|---|---|---|---|
| LPayments API5 members · 214 sessions | $2,860 | $2,000 | 143% | |
| JGrowth Web6 members · 168 sessions | $1,420 | $1,500 | 95% | |
| LData Pipeline3 members · 82 sessions | $610 | $1,200 | 51% | |
| LCheckout2 members · 50 sessions | $340 | — | Not tracked |
Outcome
See where your coding agent’s time goes, broken down by work type, repo, and theme.
Features created$4,04248% · 310 tasks
Fixes & rework$1,76821% · 136 tasks
Tests & reviews$1,09513% · 84 tasks
Exploration & overhead$1,51618% · 116 tasks
Where it happened?
By ReposBy ProjectsBy MemberBy TeamAsk EllieRepositoriesSpend
- web-app$3,105
- payments-api$2,470
- checkout-web$1,712
- infra$1,134
| Repositories | Work-type split | Spend |
|---|---|---|
| web-app | $3,105 | |
| payments-api | $2,470 | |
| checkout-web | $1,712 | |
| infra | $1,134 |
Team members
Session sync, efficiency, and proficiency for every developer.
Sessions, last 30d6462,310 all time
Synced so far6406 arriving
Sync readiness99%landed
Sync SLA2hfrom session end
Team members
Sort by Sync Progress ▾Ask Ellie- ECEmily CarterExpertEff.45Today
- RBRyan BrooksProficientEff.41Today
- MTMegan TaylorProficientEff.38Oct 0396%
Spend analysis
See where AI spend is concentrated
Compare spend by developer, model, or work type, then inspect the sessions behind each pattern.
1 dot ≈ $70 of spend$8,421 · 646 sessions
- claude-sonnet$4,631 - 355 sessions
- gpt-5.6$1,965 - 151 sessions
- claude-opus$1,193 - 92 sessions
- cursor$632 - 48 sessions
Setup
Install the CLI and sync sessions.
The first sync imports up to 30 days of Claude Code, Codex, and Cursor sessions from each machine. Sessions sync daily after the initial import.
- 01Install
Install & authenticate
$ curl -fsSL https:// cli.entelligence.ai/ install.sh | sh $ ellie auth login - 02Sync
Sync your sessions
$ ellie insights sync
Token Waste
Token waste by category.
Compaction, interrupts, repeated edits, and failed retries are measured separately.
13%of tokens used by AI agents are wasted, according to the Token Maxxing Report.
- CompactionContext rebuilt after the window filled
- InterruptsWork stopped before the agent finished.
- Repeated editsThe same files changed again and again.
- Failed retriesTool calls repeated after an error.
Watch a cause’s band shrink to see a fix land. Here, compaction after a context rule was added to the repo.
Research
Research on AI agent cost, behavior, and code quality.
Benchmarks and guides on AI agent cost, engineering behavior, code quality, and reducing waste without sacrificing reliability.
See where your agents spend, from the first sync.
Track spend, token waste, outcomes, adoption, budgets, and recommendations across Claude Code, Codex, and Cursor.
- Claude Code, Codex, and Cursor
- Up to 30 days imported on first sync
- Sessions sync daily