The first time I really looked at an AI usage dashboard, I couldn’t figure out where all the tokens had gone. Nothing seemed unusual. It was just chatting, pasting things in and asking follow-ups. That’s why this LinkedIn post stopped me mid-scroll.
The author, a founder who builds with AI every day, says most founders are burning through Claude tokens for no reason. In their words: “Not because the model is expensive. Because the way they’re prompting it is wasteful.”
The original poster learned this the expensive way, after a usage bill they didn’t expect. They list the usual suspects, and I’d bet you’ll spot at least one of your own habits here:
- Long chats that never get closed
- Vague asks that need three follow-ups to land
- Entire codebases pasted in when five files would do
- Documents re-uploaded every single session
The expert also admits they used to build things without leaning on AI tools this seriously. Now it’s a different game. When AI sits at the center of how you work, every wasted token is a real cost line.
Write complete, not less
This is the insight that stuck with me. You might think the fix is shorter prompts, but the creator says that’s wrong. “It’s not about writing less. It’s about writing complete.”
A tight prompt with the right context beats five short lazy ones every time.
This makes sense once you know how chat works under the hood. Each new message in a thread usually carries the earlier conversation with it. So five lazy messages cost more than five small chunks of text. You pay to resend everything that came before, again and again. One complete prompt avoids that.
Here’s what’s been working for the author’s team, with my notes on why each habit matters.
10 habits for leaner Claude usage
- Start a new chat per task. One giant thread grows heavier with every message. A fresh chat for each task keeps the context clean and cheap. My tip: when a long thread reaches a milestone, ask Claude for a short summary and paste it into a new chat.
- Put goal, context, requirements and output format up front. This one habit kills most follow-ups. Instead of “write an email,” try something like “Goal: re-engage churned users. Context: SaaS tool, they left after the free trial. Requirements: under 120 words, friendly tone. Format: subject line plus body.”
- Batch related asks together. Don’t fire three questions in three separate messages. Each one resends the whole history. Group related requests into one well-organized prompt.
- Strip out context Claude already has. If the draft or data is already in the conversation, don’t paste it again. Point to it with something like “use the outline above.”
- Keep Project instructions to the rules that actually persist. Project instructions load into every chat in that Project. Any bloat there costs you every single time, so keep only the rules that apply across the board.
- Store reference docs in Projects. Brand guides, product specs and style sheets belong in a Project, uploaded once. Stop dragging the same PDF into every new session.
- Only hand over the relevant files or sections. If the bug lives in five files, share those five, not the whole codebase. A focused input often gets you a sharper answer too.
- Match the model to the task. Not every job needs your top-tier model. Quick rewrites, formatting and summaries run fine on lighter models. Save the heavy hitter for hard reasoning and complex builds.
- Ask for exactly what you need. Fewer options beat twenty you’ll never use. If you only need three headlines, ask for three. Output tokens count as well.
- Don’t over-trim either. I think this caveat makes the whole list work. Cut too much context and you’ll just get more back-and-forth, which wipes out your savings.
How I’d put this into practice this week
Ten habits at once is a lot, so here’s a simple way to start based on the author’s list:
- Day one: Close your old mega-threads and commit to one chat per task.
- Day two: Write your next five prompts using the goal, context, requirements and format structure.
- Day three: Move your most-used reference docs into a Project and trim the Project instructions.
- End of week: Check your usage and compare it to the week before.
I expect most people will see the biggest drop from the first two habits alone. Closing long threads and writing complete prompts together cover a big share of the waste.
This isn’t about being cheap
The expert closes with a point I really agree with. “None of this is about being cheap with AI. It’s about being deliberate with it.”
The founders who get this early move faster and spend less doing it. The author is building this into how they run the whole business, not just how they write prompts. That makes sense to me. Clear thinking shows up in a clear prompt, and a clear prompt gets a better result for less money.
The post’s author also suggests sending this to a teammate who’s been burning through their usage without knowing why. They’re asking readers what’s missing from the list, too, so it’s worth adding your own trick.
Check out the full LinkedIn post to see the original list and join the conversation.