# Newbie here: anyone feel like Code is burning tokens way more quickly? (Opus 5.5) — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/2c379291b282dbf7

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-11-14 09:55 ingested 2026-11-15 02:34

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xvyof2/newbie_here_anyone_feel_like_code_is_burning/)

## retained text



```
\*\*Leaving the original post so you can learn from my mistakes but TL;DR; start new sessions for each change/phase. Don't yap too much. Keep convos short and focused. Use the /context command. Thanks for the help!

\--------------------------------------------------

  
So for some context, I'm working on a VST MIDI plugin, which was previously two separate plugins that were designed to talk to each other and keep in sync, but it meant updating both in parallel as they were in two separate folders (all of which I was able to do quite comfortably within my 5 hourly blocks).

Decided to merge the plugins, but that was also a bit of a rookie mistake as there were conflicting UI elements and just a big tangle of code and clutter. But, it worked. It was just very hard to use.

So I set about a total redesign and rebuild. Made a UI kit, designed it to be modular so that individual parts of the UI could be previewed and tested without full plugin rebuilds. I thought I was being more economical.

And then this morning...25% in 19 minutes, just doing a couple of UI tweaks and like one small new feature, on ONE part of the plugin.

So I asked Claude what was up and what it told me was;

\- Our conversation was long (900 lines, 6 phases)  
\- It included many, many high resolution screenshots that I never asked for and was never explicitly told "hey, look at this"  
\- It was re-reading (partially or in whole) all of this, plus code previews, with every new line of conversation

SO I've given it some new explicit rules;

\- Don't produce these high res images unless I need to see something  
\- Alert me when the conversation gets too long  
\- Give me the session prompt to continue next phases in a new chat  
\- Read code by line range, don't read whole files if you don't have to

I guess my question is, does anyone else feel their usage is higher? Do you feel like Claude is making these kinds of inefficient decisions "on purpose"? Claude would've known the conversation was long, and token heavy. It could've prompted me to start a new session...

OR is it more likely that as my understanding of what I'm making has deepened, I'm naturally hitting limits faster because I'm working faster, and more thoroughly? A bit like when you get a new computer and it feels like the fastest thing ever until you get used to it.
```



document fetched 2026-11-15 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xvyof2/newbie_here_anyone_feel_like_code_is_burning
