Hi,
I just purchased the pro plan and had a really nice session until it hit the monthly token usage after only 2 hours of use.
I used the recommended settings (claude-sonnet-4-6) and it created a new script for me (only about 200 lines). I asked it to do several changes on the workflow incrementaly and to change some figures. I asked it to add more datasources from the web.
So I guess, it did some back and forth to get what I want. But still: my monthly subscription got used in only 2 hours. It's just not usable at all like this!
How can I prevent this? How can I check, what happened? Like this, there is just no way of justifying the cost for being in coding heaven for just some hours 
I can understand the frustration. Two hours of coding should not feel like it consumed an entire month's allowance.
It sounds like the repeated edits, back-and-forth context, and especially the additional web data sources may have significantly increased token usage. The 200-line script itself is not necessarily the issue the conversation history and repeated processing can add up quickly.
I'd check the usage details to see exactly which requests consumed the most tokens. For future sessions, you could also start a fresh chat when the context gets large and avoid sending the entire script repeatedly if it hasn't changed.
It would be great if the usage dashboard provided a clearer breakdown of token consumption, because otherwise it's difficult to understand why the allowance disappeared so quickly.
I've used Codex and Claude Code a lot over the past 8 months and I burned through usage, too, until I learned how to talk to them. The more precise, direct, concise, and explicit a prompt is, the better your results will be with less usage. I like to think of each letter as a token (even though it's not), so brevity is key for both parties. I make sure to include a word limit for their responses in my prompts, like, "limit your response to 200 words". A lot of back-and-forth conversation is unnecessary and can lead to drift and other bad AI behaviors. A great piece of advice I came across on Reddit is to write your prompt then ask the codebot to edit it (use a free one or put Claude to a lower model and level to limit usage).
In my time working with both codebots, I've come to accept that Claude is an overconfident loose cannon who should not be left with any standing permissions. No matter how narrow-scoped and bounded my prompts are, how explicit I am with "this is the only thing you're allowed to do", giving Claude any autonomy leads to actions that I didn't want or anticipate. I have it permanently set on "Manual" and yes, it's tedious to hit "Allow" a bajillion times, but that's better than having Claude make a mess and chew through usage.
Be careful of how much trust you place in any AI bot in general. I learned the hard way that the first response is never the best answer. Asking "Are you sure this is the best method?" or "Are there other options you missed?" kinds of questions is very important.
ETA: It's counterintuitive, but working with a higher model like Opus 5 on medium can actually reduce usage if used correctly. Sometimes lower models just aren't smart enough which wastes a ton of time going back and forth to get what you want done. One caveat is that a smarter Claude needs an even shorter leash.