Kai

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.

Period

Summary

FY2027 to date

AI spend

$1,791

167 sessions · $11 each

Savings opportunity

$844

47% of spend, at list price

Practices on target

2 of 7

Focus: rtk coverage, Knowledge MCP use

Cache hit rate

72%

Team 79% · org 82%

FY2027 to date · estimated at list price

$844 of $1,791 spend

Largest saving first

Improve prompt caching$514
Cache hit rate 72% 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.
Keep context lean$155
23% of input tokens were past 100K of context.
Compact or /clear between tasks instead of carrying more than 100K tokens of context.
Right-size models$150
41% of routine spend ran on Opus: code research $207 of $527, documentation $105 of $238, ops $12 of $34, presentation $37 of $78.
Run docs, decks, ops and read-only research on Sonnet instead of Opus.
Route commands through rtk$24
62% of 1,595 Bash calls ran through rtk; 2.4M 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 Growth & SEO and the organization

Practice	Kai	Team	Org	Target	Recommendation
rtk coverage	62%Needs attention	62%	56%	≥ 80% of Bash calls	Install the rtk hook so test, build, git and search commands return compact output.
Knowledge MCP use 7 findings saved back	31%Needs attention	30%	30%	≥ 40% of sessions	Search the knowledge MCP before re-researching a codebase, and save findings back with remember.
Cache hit rate	72%Needs attention	79%	82%	≥ 85% of input	Keep CLAUDE.md and tool lists stable, avoid mid-session model switches, and resume sessions instead of restarting.
Context past 100K	23%Needs attention	18%	10%	≤ 10% of input	Compact or /clear between tasks; hand broad searches to subagents so their reads stay out of the main context.
Opus on routine work	41%Needs attention	58%	44%	≤ 25% of routine spend	Default routine sessions to Sonnet and keep Opus for harness, design and hard debugging work.
Sessions using subagents	49%On target	47%	50%	≥ 30% of sessions	On target
Tool error rate	2.8%On target	3.0%	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	–	–	45%Off target	21%Needs attention	62%Off target	36%On target	$38
Mar 2026	–	–	55%Off target	21%Needs attention	62%Off target	41%On target	$33
Apr 2026	–	–	57%Off target	21%Needs attention	62%Off target	38%On target	$65
May 2026	17%Off target	–	62%Off target	24%Needs attention	55%Off target	54%On target	$114
Jun 2026	45%Needs attention	12%Off target	62%Off target	23%Needs attention	53%Off target	40%On target	$106
Jul 2026	69%Needs attention	24%Needs attention	67%Off target	22%Needs attention	40%Needs attention	38%On target	$147
Aug 2026	65%Needs attention	35%Needs attention	72%Needs attention	24%Needs attention	38%Needs attention	48%On target	$55
Sep 2026	64%Needs attention	39%Needs attention	69%Off target	24%Needs attention	30%Needs attention	55%On target	$160
Oct 2026	74%Needs attention	34%Needs attention	78%Needs attention	23%Needs attention	27%Needs attention	57%On target	$125

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	2,326	24.9M	113.9M	8.9M	3.4M	$234
Opus 5	1,602	22.6M	66.4M	5.7M	2.3M	$239
Opus 4.8	1,464	24.6M	55.4M	5.1M	2M	$233
Opus 4.6	1,388	33.2M	54.1M	5.6M	2.2M	$284
Opus 4.5	63	1.9M	1.8M	237K	100K	$15
Sonnet 5.5	3,478	35.4M	169.6M	13.1M	5.2M	$173
Sonnet 5	3,693	50.1M	143.4M	12.4M	4.9M	$314
Sonnet 4.6	1,631	30.2M	55.8M	5.5M	2.4M	$163
Sonnet 4.5	49	1.5M	1.4M	186K	88K	$7
Haiku 5.5	5,805	58.9M	297.9M	22.8M	11.3M	$17
Haiku 4.5	3,010	45.7M	120.6M	10.6M	4.8M	$95
GPT Realtime (meetings)	125	2.2M	5.8M	510K	211K	$17
Total	24,634	331.2M	1086.1M	90.5M	38.9M	$1,791

Spend and tokens come from LiteLLM gateway logs and imported agent session history, priced at API list rates.
Savings are estimates: routine Opus work repriced at the Sonnet available at the time; uncached input moved to cache reads up to an 85% hit rate; rtk removing 70% of unfiltered command output, which stays in context for about half the session's remaining turns; and 50% of input past 100K context avoided by compacting. Levers overlap, so the total is an upper bound.
rtk and the knowledge MCP are scored from their rollout (rtk May 2026, knowledge MCP Jun 2026).
Generated Oct 20, 2026, 7:46 AM.