Lead engineer · AI usage, harness practices and where spend could have gone further.
Demonstration report. This engineer and their usage are synthetic.
This page is for the engineer and their manager to find cheaper defaults and harness fixes. It is never used to rank people.
FY2027 to date
FY2027 to date · estimated at list price
$398 of $3,515 spend
Largest saving first
Cache hit rate 87% against an 85% target.
Keep stable context at the front of prompts and avoid switching models mid-session, so 85% of input reads from cache.
18% of routine spend ran on Opus: code research $72 of $434, documentation $70 of $359, ops $117 of $618, presentation $14 of $86.
Run docs, decks, ops and read-only research on Sonnet instead of Opus.
2% of input tokens were past 100K of context.
Compact or /clear between tasks instead of carrying more than 100K tokens of context.
80% of 2,156 Bash calls ran through rtk; 2.8M tokens of command output went into context.
rtk compacts test, build, git and search output before it reaches the model (about 70% smaller).
Compared with Agent Platform and the organization
| Practice | Reese | Team | Org | Target | Recommendation |
|---|---|---|---|---|---|
| rtk coverage | 80%On target | 74% | 56% | ≥ 80% of Bash calls | On target |
| Knowledge MCP use 15 findings saved back | 64%On target | 48% | 30% | ≥ 40% of sessions | On target |
| Cache hit rate | 87%On target | 84% | 82% | ≥ 85% of input | On target |
| Context past 100K | 2%On target | 6% | 10% | ≤ 10% of input | On target |
| Opus on routine work | 18%On target | 29% | 44% | ≤ 25% of routine spend | On target |
| Sessions using subagents | 57%On target | 59% | 50% | ≥ 30% of sessions | On target |
| Tool error rate | 2.5%On target | 2.5% | 3.0% | ≤ 4% of tool calls | On target |
Is the engineer moving toward the targets?
| Month | rtk | Knowledge | Cache | Context | Model fit | Subagents | Savings opportunity |
|---|---|---|---|---|---|---|---|
| Feb 2026 | – | – | 70%Needs attention | 4%On target | 24%On target | 47%On target | $52 |
| Mar 2026 | – | – | 75%Needs attention | 2%On target | 24%On target | 40%On target | $76 |
| Apr 2026 | – | – | 81%Needs attention | 2%On target | 24%On target | 36%On target | $69 |
| May 2026 | 29%Off target | – | 80%Needs attention | 3%On target | 21%On target | 44%On target | $132 |
| Jun 2026 | 63%Needs attention | 20%Needs attention | 82%Needs attention | 1%On target | 20%On target | 64%On target | $29 |
| Jul 2026 | 98%On target | 51%On target | 90%On target | 3%On target | 15%On target | 56%On target | $13 |
| Aug 2026 | 96%On target | 71%On target | 93%On target | 4%On target | 14%On target | 62%On target | $8 |
| Sep 2026 | 97%On target | 79%On target | 93%On target | 2%On target | 11%On target | 55%On target | $14 |
| Oct 2026 | 90%On target | 71%On target | 93%On target | 0%On target | 9%On target | 78%On target | $5 |
FY2027 to date
By model tier, by month
Routine types (documentation, presentation, ops, code research) rarely need Opus
Cache reads are billed at a fraction of uncached input; a high share is what good caching looks like
| Model | Requests | Uncached input | Cache reads | Cache writes | Output | Spend |
|---|---|---|---|---|---|---|
| Opus 5.5 | 1,824 | 1M | 97.6M | 6.3M | 2.2M | $100 |
| Opus 5 | 1,498 | 1.2M | 88.8M | 5.7M | 2.4M | $147 |
| Opus 4.8 | 1,297 | 5.9M | 63.4M | 4.4M | 1.8M | $135 |
| Opus 4.6 | 3,629 | 33.7M | 172.8M | 13.2M | 5.7M | $480 |
| Opus 4.5 | 110 | 1.5M | 4.3M | 371K | 153K | $16 |
| Sonnet 5.5 | 10,432 | 5.8M | 542.8M | 35M | 12.8M | $282 |
| Sonnet 5 | 13,781 | 13.5M | 728.5M | 47.4M | 20.9M | $751 |
| Sonnet 4.6 | 18,434 | 141.5M | 796.9M | 59.9M | 26.2M | $1,281 |
| Sonnet 4.5 | 710 | 11M | 32.8M | 2.8M | 1.1M | $70 |
| Haiku 5.5 | 10,608 | 6M | 561.2M | 36.2M | 16.3M | $19 |
| Haiku 4.5 | 10,591 | 51.7M | 534.8M | 37.4M | 15.2M | $228 |
| GPT Realtime (meetings) | 98 | 557K | 4.5M | 325K | 144K | $8 |
| Total | 73,012 | 273.5M | 3628.5M | 249.1M | 105.1M | $3,515 |