Harper

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.

Period

Summary

FY2027 to date

AI spend

$2,104

211 sessions · $10 each

Savings opportunity

$587

28% of spend, at list price

Practices on target

3 of 7

Focus: Knowledge MCP use, rtk coverage

Cache hit rate

83%

Team 82% · org 82%

FY2027 to date · estimated at list price

$587 of $2,104 spend

Largest saving first

Improve prompt caching$368
Cache hit rate 83% 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.
Right-size models$133
45% of routine spend ran on Opus: code research $145 of $303, documentation $78 of $196, ops $40 of $100, presentation $59 of $120.
Run docs, decks, ops and read-only research on Sonnet instead of Opus.
Keep context lean$67
9% of input tokens were past 100K of context.
Compact or /clear between tasks instead of carrying more than 100K tokens of context.
Route commands through rtk$19
65% of 1,783 Bash calls ran through rtk; 2.7M 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 Social Publishing and the organization

Practice	Harper	Team	Org	Target	Recommendation
rtk coverage	65%Needs attention	63%	56%	≥ 80% of Bash calls	Install the rtk hook so test, build, git and search commands return compact output.
Knowledge MCP use	3%Off target	4%	30%	≥ 40% of sessions	Search the knowledge MCP before re-researching a codebase, and save findings back with remember.
Cache hit rate	83%Needs attention	82%	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	9%On target	9%	10%	≤ 10% of input	On target
Opus on routine work	45%Needs attention	44%	44%	≤ 25% of routine spend	Default routine sessions to Sonnet and keep Opus for harness, design and hard debugging work.
Sessions using subagents	47%On target	46%	50%	≥ 30% of sessions	On target
Tool error rate	3.2%On target	3.7%	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	–	–	62%Off target	11%Needs attention	62%Off target	43%On target	$43
Mar 2026	–	–	67%Off target	11%Needs attention	62%Off target	26%Needs attention	$87
Apr 2026	–	–	66%Off target	7%On target	62%Off target	38%On target	$98
May 2026	24%Off target	–	67%Off target	11%Needs attention	55%Off target	35%On target	$101
Jun 2026	50%Needs attention	3%Off target	72%Needs attention	6%On target	53%Off target	39%On target	$61
Jul 2026	69%Needs attention	3%Off target	75%Needs attention	8%On target	40%Needs attention	41%On target	$70
Aug 2026	74%Needs attention	4%Off target	79%Needs attention	11%Needs attention	38%Needs attention	54%On target	$67
Sep 2026	68%Needs attention	2%Off target	83%Needs attention	9%On target	30%Needs attention	44%On target	$31
Oct 2026	76%Needs attention	2%Off target	91%On target	9%On target	27%Needs attention	69%On target	$30

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	3,217	6.6M	171.9M	11.4M	5.4M	$226
Opus 5	2,623	20.1M	119.8M	8.9M	3.8M	$312
Opus 4.8	1,858	21.5M	78.2M	6.4M	2.7M	$254
Opus 4.6	3,315	54.1M	130M	11.7M	5.1M	$536
Opus 4.5	112	2.3M	4.6M	440K	161K	$21
Sonnet 5.5	4,509	9M	270.5M	17.8M	7.5M	$165
Sonnet 5	4,087	30.5M	187.2M	13.9M	5.9M	$289
Sonnet 4.6	2,161	33.6M	88.6M	7.8M	3.1M	$204
Sonnet 4.5	110	2.1M	4.1M	395K	155K	$11
Haiku 5.5	8,194	11.7M	401.7M	26.4M	9.6M	$13
Haiku 4.5	2,600	25.7M	118.5M	9.2M	3.7M	$68
GPT Realtime (meetings)	77	634K	4M	298K	116K	$7
Total	32,863	217.9M	1579.1M	114.7M	47.3M	$2,104

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.