I’ll admit it. The first time I watched an AI usage bill climb faster than I expected, I blamed the model. It felt pricey, so I figured that was the whole story. Then I read a LinkedIn post from a founder that changed how I think about it, and I’ve been sharing it with everyone since.
The author makes a blunt claim: most founders are burning through Claude tokens for no reason. The model isn’t the expensive part. The way they prompt it is what wastes money.
The original poster names the usual suspects:
- 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
If even one of those sounds like you, keep reading. I saw myself in at least three.
💸 Why Every Wasted Token Costs You
A quick primer first. Tokens are the small chunks of text Claude reads and writes, and you pay for both. Here’s the catch: in a long chat, the model re-reads the whole conversation every time you send a new message. So a thread that’s been open for three days costs more with every reply, even when your new question is tiny.
The author says they built things before without leaning on AI tools this heavily, and that working this way is a whole different game. Every wasted token is a real cost line. They learned that the expensive way, watching a usage bill they didn’t expect. I think a lot of us have been through that exact moment.
✍️ The Shift: Write Complete, Not Less
This is the part that clicked for me. The obvious reaction to a big bill is to write shorter prompts. The creator argues the opposite.
“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.”
Think about it. A lazy one-liner gets you a generic answer. Then you clarify, then you correct, then you add the context you skipped. That’s four round trips, and each one re-sends the growing history. One complete prompt gets it right the first time.
🔟 The 10 Habits Working for the Author’s Team
This is the list the founder’s team runs on right now. I’ve added a short note to each so you can put it to work today.
- Start a new chat per task. Don’t run one giant thread. A fresh chat means Claude isn’t hauling around yesterday’s unrelated history, so you’re not paying to re-read it.
- State the goal, context, requirements and output format up front. This is the “write complete” idea in practice. Tell Claude what you want, why, what limits it has to work within, and what the result should look like.
- Batch related asks together. Don’t fire questions one at a time. If you need a headline, a summary and three bullet points about the same doc, ask for all of them in one message.
- Strip out context Claude already has. If it’s already in the chat or the Project, don’t paste it again. Repeated context is pure waste.
- Keep Project instructions to the rules that actually persist. Project instructions load with every chat. Keep them to the rules that apply every time, like tone, format or the tech stack, and leave out one-off details.
- Store reference docs in Projects. Stop re-uploading the same brand guide or spec every session. Put it in a Project once and let it live there.
- Only hand over relevant files or sections. Fixing a bug in one module? Share that module and the files it touches, not the entire repo. Five files usually beat five hundred.
- Match the model to the task. Not every job needs your top-tier model. Quick rewrites, formatting and simple summaries run fine on lighter, cheaper models. Save the heavy hitter for the hard reasoning.
- Ask for exactly what you need. Fewer options beats twenty you’ll never use. If you want three headline ideas, say three. Every extra option costs output tokens you’ll skim past.
- Don’t over-trim either. This is the balance point. Cut too much context and you’re back to the back-and-forth that burned tokens to begin with.
🛠️ A Simple Template to Try
To make habit number two easier, here’s a structure I put together based on the author’s advice. It’s my own example, not from the post, but it follows their formula exactly:
“Goal: Write a cold email to a SaaS founder about our analytics tool.
Context: Our tool cuts reporting time in half. The recipient runs a 20-person team.
Requirements: Under 120 words, friendly tone, one clear ask.
Output format: Subject line plus email body. Give me two versions.”
Look at what this one prompt does. It hits goal, context, requirements and format at once, it asks for exactly two versions instead of ten, and it leaves no room for “wait, what tone?” follow-ups. You’d usually need three or four messages to get there with a lazy prompt.
🎯 Deliberate, Not Cheap
What I love most is how the author frames all of this. It isn’t a penny-pinching guide.
“None of this is about being cheap with AI. It’s about being deliberate with it.”
The expert’s point is that founders who get this early move faster and spend less doing it. Cleaner prompts mean better answers, fewer corrections and shorter cycles. The savings are a nice bonus. The real win is speed and focus.
The person who shared it also says they’re building this into how they run things, not just into how they write prompts. That stuck with me. Once you start treating context as something you manage on purpose, it changes how you set up projects, how you onboard teammates to AI tools and how you plan your work.
🚀 How to Start Today
If you want a quick way in, pick three habits and try them this week:
- Open a fresh chat for every new task
- Move your most-uploaded doc into a Project
- Switch one routine job to a lighter model
Then check your usage at the end of the week. I’d bet you see a difference.
The author suggests sending this to a teammate who’s been burning through their usage without knowing why. They also asked their readers what they’d add to the list, so the comments are worth a look. Go read the full LinkedIn post and see what other people are adding.