Click a bar to drill into agent breakdown
| # | Action | Agent | Cost | % Total |
|---|---|---|---|---|
| 1 | Code generation | CC Claude Code | $148.20 | 18.5% |
| 2 | Spec authoring | Ma Maren | $78.20 | 9.8% |
| 3 | File operations | CC Claude Code | $72.40 | 9.1% |
| 4 | Test execution | CC Claude Code | $58.30 | 7.3% |
| 5 | Pattern scanning | Br Brue | $52.40 | 6.6% |
| 6 | Verification checks | Ma Maren | $41.50 | 5.2% |
| 7 | Code generation | CC Claude Code | $38.40 | 4.8% |
| 8 | PR review | Br Brue | $38.90 | 4.9% |
| 9 | Git operations | CC Claude Code | $38.10 | 4.8% |
| 10 | Decision capture | No Nova | $32.80 | 4.1% |
| 11 | KB search | Ma Maren | $32.10 | 4% |
| 12 | Task routing | No Nova | $28.40 | 3.6% |
| 13 | Deployment | CC Claude Code | $25.60 | 3.2% |
| 14 | Claim verification | Br Brue | $22.10 | 2.8% |
| 15 | KB consolidation | Tu Tuck | $18.90 | 2.4% |
MCP Gateway trending 23% over monthly budget
At current daily spend rate ($2.52/day), MCP Gateway will exceed its $60/month budget by Apr 28. Primary driver: increased Claude Code deployment cycles during gateway hardening.
Maren's spec-authoring costs up 8% month-over-month
Spec authoring cost increased from $172.40 (March) to projected $186.40 (April, annualized). Investigating whether increased spec complexity or model pricing changes are the driver.
Utility model savings: $312 this month
Routing 1,240 background tasks to GPT-4o-mini instead of Opus saved an estimated $312 in compute costs. Tuck's context pack compilations account for 64% of these savings.
| Model | Tasks | Total Cost | $/1K input |
|---|---|---|---|
Claude 4.6 Opus Anthropic | 186 | $436.80 | $0.0150 |
Claude 4.6 Sonnet Anthropic | 420 | $148.20 | $0.0030 |
GPT-5 OpenAI | 98 | $124.80 | $0.0120 |
GPT-4o-mini OpenAI | 1240 | $18.40 | $0.0001 |
Gemini 2.5 Pro Google | 45 | $42.30 | $0.0070 |
Route reasoning-heavy tasks (spec authoring, decision capture, complex code generation) to Claude 4.6 Opus.
Route high-volume routine tasks (context compilation, freshness audits, simple file operations) to Claude 4.6 Sonnet.
Route quality review tasks to GPT-5 for adversarial model diversity.
Fall back to GPT-4o-mini for background utility tasks.