Hitting your Claude weekly limit on a Wednesday afternoon ranks near the top of modern tech frustrations. You are deep in a project, the momentum is there at last, and then every thread goes dark until the reset clock ticks over. I have slammed into that wall more times than I want to admit, often on the exact day a deadline was breathing down my neck.
So a blunt list making the rounds on LinkedIn stopped me cold: 22 rules for running Claude without torching your token budget. An AI professional opened with a dare, claiming you should not touch the tool again until you have read every one of them. The second line softened the blow, framing the whole thing as a rescue mission for people bleeding tokens on habits they never notice.
The frustration behind the post is obvious, and the fixes are almost too small to take seriously. I went through all 22 and added the mechanism behind each, so you can see why they move the needle. A handful surprised me, and one changed how I open every chat now.
*Ad
Granola is the AI notepad for people with back-to-back meetings. You jot down quick bullets while Granola transcribes straight from your computer, then it turns them into a clean write-up you can send to the team before your next call, so everyone leaves aligned.
Use code CYBERCORSAIRS for your first month on us.
1. Pick the right model and effort
The first cluster targets model choice and effort level, and the author is blunt: these two settings move your token burn more than anything else. Low effort for grammar and translation, Medium for the rest, High for the ambitious work. Never Max.
More effort means more thinking tokens, and most tasks do not need the extra churn. That is the whole insight: you are paying for reasoning you will never read. Default to Medium and feel the difference.
A rule I had not considered is to never switch models mid-chat. Change the model between turns and the new one reprocesses the entire conversation from scratch. You pay for the same context twice.
There is also a quiet discipline around resets. If you get a free limit reset, do not blow it on a routine Tuesday. Save it for the day everything breaks at once.
2. Run your chats smarter
Here is where the real leak lives. Every message you send ships the full conversation history again, so long and messy threads cost a fortune. That reframed the whole thing for me.
The author says it plain. Never reopen an old chat, because adding one question forces Claude to reread everything above it. Start fresh and paste in only what matters.
A few tactical moves stack up fast. Tell Claude 'no web search, no files' when you want a quick answer, so it stops dragging extra content into the thread. Plan in plain text first, then ask for the finished file last.
My favorite from this batch is dead simple. Highlight the broken paragraph instead of asking for a full redo, because a full rewrite regenerates everything, including the parts you liked. Point at the one line that is off.
Rules like these only pay off once they run on autopilot, and reading a list rarely gets you there. The claude track in 3 Minute AI builds that reflex in three-minute lessons, each ending with a task you run in a built-in chat lab. Daily lessons and two full courses are free.
In partnership with
Built for the unstoppable.
The Dell Pro powered by Intel® Core™ Ultra with Intel vPro® adjusts power usage based on how you work, so your battery life never slows you down.
3. Trim what you feed it
Inputs are the next quiet drain, and this is the part that sold me. A full-screen screenshot is stuffed with pixels the model does not need. Crop it to the part that matters, and convert PDFs to plain text when you can.
Then there are the background costs nobody watches. Scheduled reports you never open still burn tokens every time they run. Pause the ones you do not read.
The rule that surprised me most: start every task with 'ask me questions.' Letting the model clarify before it drafts kills wrong first drafts, and wrong drafts are where most waste hides. Batch all your asks into one message instead of five.
Two more worth stealing: load connected apps only when a task needs them, since every live tool pads the context. And speak your prompt instead of typing it. Talking lets you dump richer context up front, which means fewer correction rounds later.
4. Keep the habits clean
The last batch is pure hygiene, and it separates the people who run dry on Wednesday from the ones who never notice a limit. Never correct a derailed chat: start over with a sharper prompt. A clean restart costs less than piling corrections onto a confused thread.
Keep custom instructions under 2,000 words, since they load on every single turn, so a bloated block is a tax you pay forever. Stash reusable files in a Project instead of uploading the same brand guide again and again. One topic per chat.
In Claude Code, the move is simple: finish one task, run /clear, start the next one clean. The command wipes the session context so you are not hauling old weight around. Splitting your day into three sessions also plays nicer with how the usage windows reset.
Here is why I find this worth your time: most token waste does not come from hard problems. It comes from small habits. Long old chats, uncropped screenshots, model switching, endless corrections.
Unify Your Teams and Tech Stack With HubSpot
Connect your customer data, teams, and tools without the hassle of complex integrations or lengthy setups. One easy platform gives marketing, sales, and service teams a unified customer view and the tools to turn it into growth.
Why HubSpot and what's new
• Generate leads and automate marketing with Marketing Hub
• Build your pipeline and close more deals with Sales Hub
• Scale customer support and drive retention with Service Hub
• Keep customer data clean, connected, and actionable with one, unified platform
Join 306,000+ in over 135 countries using HubSpot to grow their businesses.
See what a more connected approach to growth can do for you and your team. Get setup quickly and start checking off your hardest tasks.
*Ad
Your next chat, done right
Open your next Claude task on Medium effort, start it in a fresh chat, and type 'ask me questions' before anything else.
If you want the other 19 moves spelled out, here is the full list of 22 rules.
Worth 10 minutes if you keep hitting your weekly limit before the week is even half over.
Poll: Be honest. Where do most of your Claude tokens actually go?
Hit reply and tell us why.
