AI Ops Weekly Digest — Week of July 27, 2026 ============================================================ COST STATUS: ON TRACK MTD: $662.58 | Projected: $760.74 | Budget: $1,500.00 (51%) 7-day avg: $27.94/day MODEL EFFICIENCY: Opus ratio: 100% | Wasteful sessions: 5 | Est. waste: $151.81 COST ALLOCATION: RGDM (internal): $18,437.94 (28 sessions) SESSION QUALITY: HEALTHY Avg score: 98/100 | Cache hit: 93.5% | Flagged: 0/28 OVERHEAD TREND: STABLE Latest avg input: 14,495,140 tokens/session | WoW: +39.0% | Consecutive growth: 2 weeks RECOMMENDATIONS: 1. [HIGH] Route trivial sessions (<5 min) to Sonnet/Haiku instead of Opus (est. $607.24/mo savings) ALERTS: [MEDIUM] (efficiency) Opus ratio 100% exceeds 25% threshold [MEDIUM] (efficiency) Estimated $151.81 waste from 5 trivial Opus sessions [LOW] (allocator) RGDM (internal) accounts for 100% of AI spend — consider cost optimization OUTCOME TRACKING: [-] Opus Ratio: 100% (was 93%, +7.9% — regressed) [+] Est. Waste: $151.81 (was $152.66, -0.6% — improved) [-] Budget Pacing: 51% (was 48%, +5.0% — regressed) [+] Avg Quality: 98/100 (was 94/100, +4.9% — improved) [+] Cache Hit: 93.5% (was 85.2%, +9.7% — improved) [-] Overhead: 14,495,140 tok (was 10,429,828 tok, +39.0% — regressed) 4-week trends: Budget Pacing: 24% → 20% → 22% → 48% (worsening) Opus Ratio: 100% → 100% → 100% → 93% (improving)

• › Route trivial sessions (<5 min) to Sonnet/Haiku instead of Opus