The process utilized Codex and Claude across four computers.
Amateur hour out here
Crossed a billion tokens in the last 24 hours
I think most of it was me making a sub agent for every thread I did with fable to analyize it for useful workflows or codesnippets I could re-use in the future
Got really good effect v4, svelte, and expo skills out of it for codex
Many users expressed excitement about developers hitting massive token volumes like 11B in a day using AI sub-agents for code work, while others criticized the focus on high API spending as bragging without real value.
No Digg Deeper questions have been answered for this story yet.
How did I do it? Handful of things:
1. Bumped the subagent parallelism limit in Codex from 3 to 20 2. Set up a well defined goal to port a big piece of software to Rust (more coming soon) 3. Revived an old project that needs a LOT of work 4. this
@RhysSullivan nope
@davis7 are you using https://github.com/baskduf/FableCodex?
@davis7 Had same issues. Working on this which saves a ton of tokens as well as helps agents and subagents navigate codebases better, maintain lossless context. https://github.com/ScriptedAlchemy/tracedecay
@chrislciaba Rebuilding things that already exist is easy, would have been way fewer tokens
@davis7 Congratulations!!!! You have to beat my 4B token in 24h high score before I beat it
@theo It's wild people still don't consider things like html stream when this is the amount of tokens people spend https://github.com/Alphanimble/htmlstream
@theo Are you rewriting google?
@theo I need to 10x my output 😂😅
@davis7 @RhysSullivan its physically painful watching how @davis7 somehow manages to come up with some of the most unbelievably braindead ideas imaginable.
@_virgil19 lmao no not at all
20 parallel agents means the code is being reviewed 20 times by bots before I ever have to look at it
@theo @theo Check out https://context-guard-plugin.vercel.app
It should help you to safe some of your tokens. Still in testing but helps me so far to enhance my usage limits in Codex. Also working on a Claude Code version
@orcus108 @theo This. Imagine the same situation with any other profession. Social media being full of construction workers bragging about how many hours they spent on a site without the job being done.
@theo i wonder if there’ll ever be a way to measure how productive the tokens generated are. maxxing out on tokens shouldn’t rly be a flex unless you’re actually doing something meaningful
not saying you waste it lol (maybe you do idk) but just in general
@theo 20 parallel agents just means 20 outputs to review. the bottleneck didnt move, it got 20 copies
@theo
@theo Gotcha so fable 6
@theo @grok how to get the overview from codex, Claude and cursor as well as grok build in one?
@theo I mean, this is mostly just May on my primary workstation.
At some point you parallelize enough that it's just a question of how many tokens you throw into the furnace.
@davis7 once we get to the point where subagents can be local, it's gonna be more fun