I stared at the “you’ve reached your usage limit” message three times last week. Same subscription, same work, and yet somehow I kept smacking into the wall by mid-afternoon. My first instinct? Upgrade the plan. Throw money at it.
Then I ran across a post from a LinkedIn creator who flipped that instinct on its head, and I’m glad I read it before I paid for more.
The core idea from this savvy professional is blunt and a little uncomfortable: your tokens are leaking, and your habits are the leak. You’re not hitting limits because your plan is too small. You’re hitting them because of how you’re using the tool. The expert laid out seven fixes anyone can copy, and I want to walk you through each one because the reasoning behind them is where the real value sits.
🔧 The 7 fixes the original poster shared
- Match the model to the task: Haiku, Sonnet, Opus, Fable. This is the biggest one and most people skip it entirely. Running a quick formatting job or a simple classification through the heaviest model is like renting a moving truck to pick up a coffee. Lighter models handle the small stuff fast and cheap. Save the big guns for the work that actually needs deep reasoning.
- Paste text instead of dropping files. When you attach a file, the whole thing gets processed, headers, boilerplate, formatting artifacts and all. Paste the relevant chunk instead and you’re only paying for the part you care about. Sounds small. Adds up brutally fast across a day.
- Make Claude plan before you let it build. A plan is cheap. A wrong build followed by three rounds of corrections is expensive. Ask for the approach first, approve it, then let it run. You catch the misunderstanding while it costs you almost nothing.
- Stop stacking follow-ups, edit one message instead. Every follow-up drags the entire conversation history along with it. Ten follow-ups means you’re re-sending the whole thread ten times. Editing your original message keeps the context lean.
- Batch every ask into one prompt, not five chats. Five separate conversations means five separate context loads. If the asks are related, stack them into a single well-structured prompt and let the model handle them together.
- Turn off web search and unused connectors. Every active connector is potential context injected into your session, whether you use it or not. If you’re not searching the web for this task, that capability is just sitting there eating budget. Switch off what you don’t need.
- Keep your “about me” file under 2,000 words. That file loads into every single conversation. Every. Single. One. A bloated 6,000-word profile is a tax you pay on every message you send. Trim it to what actually changes the model’s behavior.
Why it matters: the creator’s framing is that upgrading is no longer enough to win in the token economy. You can buy a bigger bucket, but if the bucket has holes, you just get to watch more water leak out. Fix the holes first.
🎙️ The bonus tip that I think is quietly the best one
The post’s author tossed in an extra tip at the end, and honestly it might be my favorite of the bunch: speak your prompts instead of typing them, using a voice tool like Wispr.
The logic is beautiful. When you type, you’re lazy. You write a short prompt because typing is work. When you speak, you naturally give richer context in one take, you explain the background, you mention constraints, you describe what you actually want. Richer context means fewer follow-ups. Fewer follow-ups means fewer tokens.
More answers from the same plan. As the original poster puts it, never a loss.
💡 How to actually apply this today
Reading a list is easy. Changing habits is the hard part. Here’s how I’d sequence it if you want results this week:
- Start with your “about me” file. It’s a one-time fix with a permanent payoff. Open it, count the words, cut the fluff. This costs you fifteen minutes and saves you tokens on every message forever.
- Audit your connectors next. Another one-time toggle. Go through what’s enabled and switch off anything you haven’t deliberately used in the past week.
- Then work on the model-matching habit. Before you type, ask yourself: does this need deep thinking, or does it need speed? Rough rule I use: rewrites, formatting, simple extraction, quick questions go to the light models. Architecture decisions, complex debugging, nuanced writing go heavy.
- Finally, break the follow-up reflex. This one’s the hardest because it’s muscle memory. Next time you’re about to type “actually, also add…” stop and edit the original message instead.
🎯 Who this really helps
If you’re a solo user, this is maybe an hour of setup that buys you back a chunk of your daily capacity. But I think the bigger win is for teams. If five people on your team are all hauling around bloated profile files, leaving connectors on, and stacking follow-ups, you’re collectively burning through a plan’s worth of tokens on pure waste. The original poster even framed it as a team-level fix, and I think that’s right.
There’s a broader trend underneath all this too. As AI tools get baked into daily workflows, the skill isn’t just prompting well, it’s using compute efficiently. The people who figure out how to get the most out of a fixed budget are going to have a real advantage over the people who just keep upgrading. Context management is turning into a genuine professional skill, and posts like this one are how you pick it up.
📋 The quick version
If you remember nothing else from this, remember these three:
- Right model for the right job
- One rich prompt beats five thin ones
- Trim what loads on every message
What I appreciate most about this contributor’s approach is that none of it requires you to be technical. There’s no API, no config file, no scripting. It’s all habit-level stuff you can start doing in your next conversation.
The full LinkedIn post has the complete breakdown along with a video walkthrough. Go check it out, and maybe fix your “about me” file while you’re at it.