Marlo

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

$2,284

169 sessions · $14 each

Savings opportunity

$543

24% of spend, at list price

Practices on target

3 of 7

Focus: rtk coverage, Knowledge MCP use

Cache hit rate

82%

Team 79% · org 82%

FY2027 to date · estimated at list price

$543 of $2,284 spend

Largest saving first

Improve prompt caching$300
Cache hit rate 82% 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$156
44% of routine spend ran on Opus: code research $200 of $462, documentation $85 of $191, ops $68 of $152, presentation $27 of $65.
Run docs, decks, ops and read-only research on Sonnet instead of Opus.
Keep context lean$70
8% 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$17
64% of 1,655 Bash calls ran through rtk; 2.5M 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 Web Platform and the organization

Practice	Marlo	Team	Org	Target	Recommendation
rtk coverage	64%Needs attention	46%	56%	≥ 80% of Bash calls	Install the rtk hook so test, build, git and search commands return compact output.
Knowledge MCP use 8 findings saved back	32%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	82%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	8%On target	8%	10%	≤ 10% of input	On target
Opus on routine work	44%Needs attention	42%	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	51%	50%	≥ 30% of sessions	On target
Tool error rate	3.0%On target	2.7%	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	–	–	60%Off target	6%On target	62%Off target	18%Needs attention	$53
Mar 2026	–	–	66%Off target	9%On target	62%Off target	34%On target	$46
Apr 2026	–	–	67%Off target	10%Needs attention	62%Off target	31%On target	$80
May 2026	29%Off target	–	72%Needs attention	8%On target	55%Off target	35%On target	$112
Jun 2026	42%Needs attention	16%Needs attention	74%Needs attention	10%Needs attention	53%Off target	50%On target	$68
Jul 2026	66%Needs attention	27%Needs attention	80%Needs attention	10%Needs attention	40%Needs attention	48%On target	$61
Aug 2026	73%Needs attention	38%Needs attention	81%Needs attention	9%On target	38%Needs attention	58%On target	$67
Sep 2026	75%Needs attention	37%Needs attention	86%On target	10%On target	30%Needs attention	55%On target	$37
Oct 2026	74%Needs attention	31%Needs attention	88%On target	6%On target	27%Needs attention	65%On target	$21

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	3,301	11M	149.2M	10.2M	4.1M	$208
Opus 5	4,262	24.8M	190.3M	13.7M	5.7M	$448
Opus 4.8	2,008	19.6M	87.5M	6.8M	2.9M	$257
Opus 4.6	3,058	45.9M	121.8M	10.7M	4.6M	$472
Opus 4.5	162	3.3M	5.9M	584K	228K	$29
Sonnet 5.5	3,841	13.1M	190.9M	13M	5.3M	$131
Sonnet 5	5,955	37M	283.5M	20.5M	8.7M	$403
Sonnet 4.6	2,203	31.2M	97.6M	8.2M	3.5M	$206
Sonnet 4.5	130	2.7M	4.7M	472K	199K	$14
Haiku 5.5	4,826	16.5M	256.9M	17.4M	7.1M	$10
Haiku 4.5	4,167	33.3M	188.2M	14.1M	5.4M	$97
GPT Realtime (meetings)	104	819K	5.3M	391K	165K	$10
Total	34,017	239.1M	1581.8M	116.2M	47.9M	$2,284

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.