# What an on going cost optimization looks like . — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/258fba39fce2fd48

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-11-19 05:46 ingested 2026-11-20 02:39

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1y01sg0/what_an_on_going_cost_optimization_looks_like/)

## retained text



```
same project ... on different 5h windows in same day .  
with continues evaluation and optimize .  I did not reach yet that perfect souse . that fit everyone but the stuff I reach and added to my general process are .  
in order of value .

1. L**ess is more :** always smaller project and code size will cost less to handle .
2. **standards :** standards and clean architecture are there for reason even if Ai can work faster than humans, still it will reach that point where it will hit the same time and cost limits that human hit . start early better than wait till later and endup with huge mess to clean . 
3. **cache :** utilize it well and keep it over 95% , never touch cold cache .
4. **context limit :** manage your context to extreme you need to reach the point that each word in your agent context is worth it by Auto compacting , context limit , work dividing , smart handoff .
5. **get your valuable bots a secretaries :** top level agent with fable should not get his context filled with bullshit like console or logs or screenshots . my director Fable agent have a Sonnat secretary that update the board for him and check the browser and logs .
6. **multilevel and true project management :** instead of fable managing direct employees . he manages project managers and for each he have team lead and they work in waves . project manager see which tasks can be done in parallel assign each task to new agent with his own tree they work finish , the reviewer review then the manager check finally tell his secretary to update board and git , take the final report and talk to fable CTO and get advised of any issue .

I am still improving the part where when each model level is at its limit and when its an over qualified. to use sonnat and haiuka more and more .

right now its 10% fable , 80% opus , 30% sonnat the issue is to get a true factual data ... I need to make a full real project like an app or an game in high quality with different sets and mixes and thats would cost days of tests and there is not $$ return of that .  
ps . I think big companies already have answers for that and exact numbers . that they will not share.

I am keep doing small step improvements to stay in my usage limit . which for my main work the 20$ plan limit is enough . for all real work and some agents running in background doing my side projects . and reaching the weekly reset at below 95% . 
```



document fetched 2026-11-20 00:41 · canonical https://reddit.com/r/ClaudeCode/comments/1y01sg0/what_an_on_going_cost_optimization_looks_like
