Live demo · sample data

A fictional org’s spend, reconciled to what actually shipped.

The full spend-review loop on sample data: what merged, what is in flight, and what never tied to a PR. Review and scan actions are disabled here. Your own workspace shows your real org.

Org overview

Capture 0/5 · 5 awaiting

AI coding spend review

See what shipped, what is still in flight, and what did not tie back to a PR before you drill into team and engineer diagnostics.

5 expected engineers have never sent capture (joined over a week ago).

Captured AI spend

ⓘ

$48,200

6138.5M tokens · list-price estimate

Shipped: $28,400
In flight: $7,100
Didn’t tie to a PR: $12,700

Shipped

$28,400

59% of captured spend · 483 PRs

Didn’t ship

$9,800

Past attribution window: review

Where the spend went

Spend states stay aggregate-first. Open a team or engineer only when you need diagnostics.

Merged PRs

59%

3616.9M tokens · $28,400

Tied to merged PRs

In flight

15%

904.2M tokens · $7,100

Open PR or inside attribution window

Didn’t tie to a PR

26%

1617.4M tokens · $12,700

No matching PR20% · $9,800
Non-PR work4% · $1,800
No git context0% · $200
Other repo2% · $900

Cache reuse

Share of input-side tokens served cheaply from cache, not re-paid as fresh input or writes.

37%

Reads per write

0.8×

Cache writes

$26,984

Cached vs. fresh

3797.6M / 6406.1M

Opus 4.829%$24,151

Haiku 4.554%$318

Sonnet 4.656%$2,515

Cache churn.Context is being re-paid rather than reused. Check for long idle gaps (>5 min expires the cache), volatile context, or heavy MCP payloads.

Spend over time

$48,041Last 30 days

Daily AI spend (estimated + metered), by model version, against the spend that shipped.

Opus 4.8Sonnet 4.6Haiku 4.5Spend ($, left)Yield (% shipped, right)

Spend by model

Which Claude model versions the spend ran on, how much of each shipped, and the right-sizing headroom.

Opus 4.8

$42,091

87% · 5456.4M tokens

49% shipped · tied to merged PRs

Sonnet 4.6

$5,440

11% · 1097.1M tokens

54% shipped · tied to merged PRs

Haiku 4.5

$669

1% · 408.1M tokens

53% shipped · tied to merged PRs

Right-sizing headroom: up to $16,837. Flagship-model spend ($42,091) would be ~$25,255 at Sonnet rates. Review which flagship tasks could run on a smaller model.

Actionable insights

What the agent surfaces for review, with the dollars at stake: unshipped spend, budget spikes, or cap risk. 3 open flags.

Reviewed

45%

Actioned

36%

Legit non-PR

18%

False-pos / attribution

Budget anomalyPlatform$35,462
Platform spend is 12.7× its baseline
Platform: $35,478 in the last 30 days vs ~$2,801 per 30-day baseline. Check whether shipped output rose with the spend.
AI spend tied to merged PRs (PR-only scope) · billing: metered API · 30-day attribution window · ≥3 contributors (k-anonymous) · metadata only: no prompts, code, or transcripts stored
Period 5/16/2026 – 6/20/2026 · last scan 6/20/2026
Details
- 74% of captured spend this period ($35,462)
- Set a monthly budget in Settings to see this as a share of budget.
Cache hygieneRecurringPlatform$21,283
$21,267 on cache writes at 19% reuse (Platform)
Platform reused only 19% of its input-side tokens from cache in the last 30 days while spending $21,267 writing context into cache. Low reuse means context is re-paid instead of amortized. Check for long idle gaps (>5 min expires the cache), volatile context, or heavy MCP payloads.
AI spend tied to merged PRs (PR-only scope) · billing: metered API · 30-day attribution window · ≥3 contributors (k-anonymous) · metadata only: no prompts, code, or transcripts stored
Period 5/9/2026 – 6/20/2026 · last scan 6/20/2026
Details
- 44% of captured spend this period ($21,283)
- Set a monthly budget in Settings to see this as a share of budget.
- Recurring: the same flag was raised in an earlier period.
Model right-sizingRecurringPlatform$12,810
$12,857 model right-sizing headroom (Platform)
Platform ran $31,997 on flagship models in the last 30 days; the same usage at Sonnet rates would be ~$19,237. Review whether these tasks needed the flagship model.
AI spend tied to merged PRs (PR-only scope) · billing: metered API · 30-day attribution window · ≥3 contributors (k-anonymous) · metadata only: no prompts, code, or transcripts stored
Period 5/9/2026 – 6/20/2026 · last scan 6/20/2026
Details
- 27% of captured spend this period ($12,810)
- Set a monthly budget in Settings to see this as a share of budget.
- Recurring: the same flag was raised in an earlier period.

Reviewed history (11)

Unshipped: review$640 unshipped to review$6404/20/2026
Outcome: Action taken
Budget anomalyPlatform spend is 2.1× its baseline$4305/23/2026
Outcome: False positive
Budget anomalyBilling spend is 2.4× its baseline$3804/24/2026
Outcome: Expected adoption spike
Unshipped: review$260 unshipped to review$2606/9/2026
Outcome: Action taken
Model right-sizing$210 model right-sizing headroom (Platform)$2105/1/2026
Outcome: Action taken
Cache hygiene$190 on cache writes at 22% reuse (Platform)$1905/13/2026
Outcome: Action taken
Unshipped: review$150 unshipped to review$1505/7/2026
Outcome: Legitimate research / exploration
Unshipped: review$120 unshipped to review$1205/18/2026
Outcome: Incident / support / debugging
Unshipped: review$110 unshipped to review$1106/1/2026
Outcome: Attribution issue
Model right-sizing$95 model right-sizing headroom (Billing)$95.005/28/2026
Outcome: Action taken
Cache hygiene$85 on cache writes at 31% reuse (Billing)$85.006/10/2026
Outcome: Legitimate research / exploration

Teams

2 teams

Team	Engineers	Tokens	Avg tokens/PR	Est. spend	Shipped %
Platform	3	4137.4M	6.8M	$43,866	48%
Billing	2	2001.1M	6.6M	$4,334	67%

Each row is team totals · Shipped % = share of tokens tied to merged PRs · per-engineer diagnostics are admin-only in the full product.