I hit Claude’s token limit mid-project last week and my first instinct was to reach for my wallet. Upgrade the plan, problem solved, right? Turns out that’s the lazy answer. I came across a brilliant post from an AI professional who laid out 13 stupidly simple fixes that stretch your tokens way further, no upgrade required.
What I love about the creator’s approach is that none of this is magic. It’s just smart habits around how you feed Claude information and how you manage your chats. The original poster clearly spends serious time in the trenches, because every single tip targets a specific way we quietly waste tokens without realizing it.
I broke down all 13 below so you can start applying them today. Let’s get into the curated list.
The 13 token-saving fixes
- Convert PDFs to plain .md before uploading. PDFs carry tons of hidden formatting junk that eats tokens. Strip them down to clean Markdown and Claude reads the same content for a fraction of the cost.
- Crop your screenshots. The author points out a single screenshot can cost around 1,300 tokens. Crop it tight to just the part that matters and you drop under 100. That’s a massive saving for one quick trim.
- Plan in Chat, build in Cowork. Do your brainstorming and back-and-forth in regular chat, then move to Cowork only when you’re ready to build the actual file. Keeps the heavy lifting separate from the thinking.
- Tell Claude to ask first. Add the line “Read my folder. Ask me questions first.” Instead of Claude guessing and generating a giant wrong output, it clarifies upfront and saves you a full do-over.
- Redo only the broken part. Don’t regenerate the whole output when one section is off. Point Claude at the broken piece and fix just that.
- Batch tasks into one message. Got three things to ask? Put them in a single message. That’s one context reload instead of three, which the original poster notes is a quiet token drain most people miss.
- Hit ‘Edit’ instead of a new prompt. When you want to tweak a request, edit your existing message rather than firing off a fresh one. It reuses the context instead of stacking more on top.
- Start a new chat when the topic changes. The second you switch subjects, open a fresh chat. Old context keeps loading and billing you even when it’s no longer relevant.
- Summarize and restart every 15 to 20 messages. Long chats balloon in cost because every message drags the full history along. Summarize where you’re at, then start clean.
- Match the model to the job. Use Sonnet for easy tasks and save Fable or Opus for deep, complex work. No sense burning premium horsepower on simple requests.
- Turn off Search and connectors you aren’t using. Every active connector adds overhead. Switch off the ones that aren’t pulling their weight.
- Spread work across 2 to 3 sessions a day. Instead of one marathon session, break your work into chunks throughout the day to stay under the limits.
- Trim your “About Me” file under 2,000 words. This file loads into context constantly, so a bloated one costs you on every interaction. Keep it lean.
Why it matters: most of us blame the plan when we hit a wall, but the expert shows the real culprit is sloppy context management. Tokens leak from screenshots, long chats, and unused connectors, not from you doing too much actual work.
My favorite takeaways
A few of these genuinely changed how I work. The screenshot crop tip blew me away, because I had no idea one image could swallow over a thousand tokens. That’s an instant fix anyone can apply in two seconds.
The other big one for me is the “ask me questions first” line. So much waste comes from Claude confidently building the wrong thing because it filled in gaps with guesses. Forcing it to clarify upfront means you generate the right output once instead of three times.
And the model-matching advice is pure common sense that I’d been ignoring. I was running heavy models on tiny tasks out of habit. Switching to Sonnet for the simple stuff freed up a surprising amount of room for the deep work that actually needs the muscle.
The bigger picture
What this savvy professional is really teaching is a mindset shift. You’re not limited by your plan nearly as much as you’re limited by your habits. Clean inputs, fresh chats, and the right model for each job add up fast.
I think this is the kind of breakdown that pays for itself the first day you use it. Try even three or four of these tweaks and you’ll feel the difference in how far a single session stretches.
The creator shared a full free breakdown with even more detail on the setup. Check out the original LinkedIn post to see the complete walkthrough and grab the rest of the context behind each fix.