real question: do people who don't code run into limits on claude/codex plans?
lowkey always thought that coding tends to use 10-100x more tokens than other kinds of work
Andrew Gao from Cognition Labs contends that everyday non-coders seldom bump into Claude’s usage caps because their prompts chew through far fewer tokens than the sprawling, iterative sessions typical of coding work.
real question: do people who don't code run into limits on claude/codex plans?
lowkey always thought that coding tends to use 10-100x more tokens than other kinds of work
Replies quickly surfaced counter-examples: repeated literature searches, economics modeling, and philosophy threads can balloon context and exhaust the same rolling windows that coders complain about.
Exact token allowances for Pro, Max 5x, and Max 20x tiers remain dynamic and user-specific, so it is unclear how much buffer any given non-coder actually enjoys after recent limit doublings.
Users discuss token limits on Claude and Codex for non-coders, with some agreeing coding quickly consumes tokens while others dismiss non-coders using Codex entirely.
No Digg Deeper questions have been answered for this story yet.
@itsandrewgao Yeah I hit the limit on Claude with economics/philosophy stuff regularly
Codex outside of code have never hit a limit
real question: do people who don't code run into limits on claude/codex plans?
lowkey always thought that coding tends to use 10-100x more tokens than other kinds of work

@itsandrewgao non-coders hit limits too context is king
@itsandrewgao I just did with a non-coding task. Still highly technically demanding.
real question: do people who don't code run into limits on claude/codex plans?
lowkey always thought that coding tends to use 10-100x more tokens than other kinds of work

@itsandrewgao I find I hit limits a lot faster doing strategy, knowledge work, research etc, because my workflow is so much more optimized for coding with subagents, repo mapping stuff, ast-based search tools etc
@itsandrewgao It's pretty easy to hit limits if you are doing iterated literature search
real question: do people who don't code run into limits on claude/codex plans?
lowkey always thought that coding tends to use 10-100x more tokens than other kinds of work

@Hiraweb3 sorry I should be more clear - I meant rate limits / quota limits

@itsandrewgao They don't. Sam even said they lose money on power users. They make money only on those who don't hit limits.

@itsandrewgao Analysis eats like crazy. Had a couple things where I needed 1000s of reports analysed, eats through usage much faster than coding.

@itsandrewgao You could argue that it is all code - or, it is all tokens.

@itsandrewgao Yep

@itsandrewgao in periods i dont code i tend to hit limits even faster because im parsing through documents and do research which eats up the context

@itsandrewgao That's a great point! I've noticed coding can definitely rack up token usage quickly compared to other tasks.

I'm confused if you're using codex and not coding, what the fuck are you even doing with it?
if you mean chatgpt, I mean ideally no, but there are ratelimits and abuse guardrails that stop you from hammering pro models all day.
openai has not yet moved to a "shared usage across all platforms" model like anthropic or google

@itsandrewgao Basic chat does not Anything agentic, yes.

@itsandrewgao my kid plays around building mods and in-between pasting screenshots and what not, well... he is "coding" in the end, and hitting limits!

@itsandrewgao Bad coders use 100x more tokens than good coders

@itsandrewgao Hit one on my $20 ChatGPT plan today. I asked it to take a look thru all my apple notes and it took ages. No coding done

@itsandrewgao law is the only other field i'd imagine it'd happen

@itsandrewgao an afternoon of refactoring burns more than a month of chatting. it's the file reads

@itsandrewgao Research eats more