How Do You Run Out of Tokens so Fast?
Seems that lots of posts on social media complain that even the expensive ($200/mo) paid tiers for coding tools like Claude Code and Codex don't provide sufficient output to sustain any serious efffort: "One day in and I've reached my weekly token limit." Others pick up on that with glee, declaring that AI coding was just a brief fad while AI was heavily subsidized and that now that the bill's coming due, it's just not viable anymore.
I've come to believe that this is mostly posturing, based on three data points:
Myself. Of course I want to be careful about extrapolating, hence the additional data points. Even with running Claude Code in high-effort thinking mode on just about any programming task, ideally with multiple agents working together, I don't get anywhere near my usage limit.
This guy I chatted to during Web Summit who coded his own trading agent and uses Claude Code for all that. Asked him about token usage and he's well within the limits of even one of the cheaper plans.
Luca Rossi, writer of the Refactoring newsletter, chronicling his agentic coding workflow to build the note-taking app Tolaria with a coding agent running 24/7, pulling in issues from a task board and working on them. According to his latest update on the coding workflow, he stays well within the limits of Codex's subscription.
The third point in particular tells me that there's a way of coding with AI that produces great results without burning all the tokens. If you find yourself running out halfway through the billing period, take a look at Luca's writeups or hit reply and tell me about how you're using AI and I'm sure there's a more productive way forward.
