Why you run out by Tuesday

You open your go to AI app and the menu is waiting. Thinking or instant. Pro or Flash. You pick the strongest one, because who picks the dumb one for a work task?

If that sounds familiar, some version of your week probably looks like this. It's Tuesday and you're already out. You paid the $20 and by Thursday you're rationing messages like the free plan never ended. You've stared at the $100 tier page doing math you resent. And somewhere in there you decided you must be overdoing it.

You're doing what the menu was built to make you do.

The companies hand you the menu and leave the choosing to you. Where guidance exists, it sits in docs and tutorials that almost no subscriber opens, while the app defaults you into whatever it defaults you into.

I think there's a reason it stays vague, and this part is my read, so weigh it as one. If a company prints "the small model is enough for email," every small-model failure lands on them. Keep the menu vague, and the failure lands on you. You blame yourself. You upgrade. Vague menus sell upgrades.

Even the people who study these models full time trip on the names. Ethan Mollick flagged that OpenAI had been fixing its "naming curse" with Luna, Terra, Sol, and Codex, then piled new ChatGPT-based names on top.

So if you've been reading every limit warning as proof you're doing this badly: the menu has no prices on it. Running out is what happens when a menu hides the prices.

Do you run out?

If you never hit a limit: use the everyday model for everything and close the menu. Email, summaries, rewrites, brainstorming, meeting notes. On jobs like these, the everyday model and the flagship land in the same place. Practitioners who route by task say the standard model covers the large majority of their work . On the free plan? This is you, and you don't need to pay yet.

If you run out constantly: your best model has been going to your lightest work. Flip the default. Everyday model for everyday jobs. Save the thinking or pro mode for work with real reasoning inside: multi-step logic, code, math, one long messy document.

That's the rule. It fits any plan, free included, because every plan has a limit somewhere.

Once you've flipped the default, one thing still drains your limit faster than anything else: the chat you never close.

Every message you send, the model re-reads the whole conversation before it answers. The thread accumulates, so every new turn in a long chat costs a little more than the last. I measured a single Claude thread and found message 30 cost roughly 30x message 1. Some tools soften it with caching, but the month-old thread you keep alive because "it already has the context" is still the most expensive habit you have.

The full fix is in the web post. It shows you how to:

Stop wasting your strongest model on email, summaries, rewrites, and other everyday work.

Leave a bloated chat without starting from zero, using a short handoff that carries the decisions and unfinished work into a clean thread.

Cut the extra turns that make a conversation more expensive, including when to edit an earlier message instead of adding another correction.

Keep large PDFs and first-pass research from chewing through your paid limit before the real thinking begins.

Test a smaller model on your own repeat work, then trust it only after it earns the job.

You can use any one of these today. Together, they turn the model menu and the usage meter into something you rarely need to think about.

I changed one habit that afternoon: Haiku for repetitive batches, Sonnet for recurring automation like daily morning brief, ChatGPT for personal life assistant & image generation, and Opus/fable5 only when the job earns it. It took 30 seconds. I haven’t seen the limit warning since—and the annoying part is how long I waited to do it.

Which small job has been getting your biggest model? Hit reply. I read every answer.

Dan Rice · AI Signal · Every Tuesday Read once. Use AI better all week.