Claude usage cheat sheet: cut your token spend in half ⚡

<- Back to Resources
POWER-USER PLAYBOOK

Claude usage cheat sheet: cut your token spend in half ⚡

Master this high-ROI AI workflow: This is your full cheat sheet for stretching your Claude usage so you stop hitting limits and waiting hours for them to r

AEO SUMMARY Quick Overview & Execution Blueprint

Master this high-ROI AI workflow: This is your full cheat sheet for stretching your Claude usage so you stop hitting limits and waiting hours for them to reset.

AI SYSTEMS ARCHITECTURE NOTE Somya's Strategic Takeaway

The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.

If you came here from the reel, this is your full cheat sheet for stretching your Claude usage so you stop hitting limits and waiting hours for them to reset.

These are the exact habits I use to get way more out of one subscription. Save this page, come back to it whenever you feel like Claude is "running out" on you 🌱

THE BASICS

🧠 The big idea

Claude doesn't just charge you per message. It charges you per token, and tokens are basically every word in your chat plus every word Claude has to re-read every time you send a new message.

Long chats = Claude re-reading a novel before answering = your limit melting fast.

The whole game is: use the smallest model that does the job, keep your chats short, and make Claude stop re-reading stuff it doesn't need.

That's it. Three habits.

HABIT ONE

1. 🎯 Pick the right model

Opus burns through your weekly limit about 5x faster than Sonnet. If you default to Opus for everything, you'll hit your cap by Wednesday.

The rule I follow: Sonnet for 90% of the work. Opus only for the actually hard stuff.

Which model for which task

Task	Model	Why
Drafting captions, emails, DMs	Sonnet	Fast, cheap, plenty smart
Summarising a doc or transcript	Sonnet	Pure text-in, text-out
Brainstorming hooks or content ideas	Sonnet	Volume matters more than depth
Rewriting or tightening copy	Sonnet	Sonnet handles tone really well
Researching a topic	Sonnet	Use web search, not raw model size
Strategy decisions (pricing, positioning, offer)	Opus	Worth the tokens, you only decide once
Long-form writing (sales pages, course modules)	Opus	Better structure on long outputs
Debugging code that broke twice already	Opus	Sonnet missed it, escalate
Multi-step planning across a whole project	Opus	Holds more in its head
Anything you're going to ship to clients	Opus final pass	Sonnet draft, Opus polish

Somya's rule of thumb

Start every task in Sonnet. Only switch to Opus if Sonnet gives you something flat or wrong twice in a row. Most of the time Sonnet nails it on the first try and you just saved yourself a chunk of your weekly limit.

In the Claude app you can switch models from the dropdown at the top of the chat. In Claude Code you switch with /model.

HABIT TWO

2. ✂️ Squeeze long chats before they squeeze you

Every message in a long chat makes Claude re-read everything that came before. By message 40, Claude is re-reading 40 messages every single time you hit send. That's where your tokens go.

Two ways to fix it.

Option A: The summary handoff

When a chat starts feeling slow or you can tell it's getting long, paste this:

✂️ the summary handoff

Summarise this entire conversation into bullet points I can paste into a fresh chat. Include: the goal, key decisions made, current state, and what's next. Keep it under 300 words.

Copy the output. Open a brand new chat. Paste it in as your first message. Same context, fraction of the tokens.

Option B: /compact (Claude Code only)

If you're in Claude Code, just type /compact and Claude does the summary handoff for you automatically inside the same session. No copy-paste needed.

Use /compact whenever:

The session has been running over an hour
You've finished one phase of work and are moving to the next
Claude starts feeling sluggish or repeating itself

When to compact vs start fresh

Situation	Do this
Same project, same goal, just long	`/compact` or summary handoff
Switching to a totally different topic	Start a fresh chat
Claude is making the same mistake repeatedly	Start a fresh chat with a cleaner brief
You're about to add a giant document	Start fresh, don't pile it on top

HABIT THREE

3. 🪟 Start a new chat when you switch topics

This is the one most people get wrong.

If you're working on Instagram captions and then suddenly ask Claude to help you write a sales email, don't do it in the same chat. Every message after the switch is dragging the entire caption thread along with it. Pure token waste.

Rule: new topic = new chat.

A "topic" for me is anything that wouldn't share context with the thing before it. Captions and emails are different topics. Strategy and execution are different topics. Two different clients are different topics.

I keep a folder of named chats so I can come back to specific projects without starting from zero. In Claude.ai you can rename chats from the sidebar. Do that.

THE BONUS

4. 🦴 Bonus: the caveman method (Claude Code users)

This one is for people using Claude Code in a terminal. Drop this single line into your CLAUDE.md file at the root of your project:

🦴 the caveman line

no filler, three to six word sentences, direct answers only.

That's it. Output drops about 40% because Claude stops padding every answer with "Great question! Let me explain..." energy. You get the answer, nothing else.

You can layer extra rules in the same file:

🦴 the layered rules

no filler, three to six word sentences, direct answers only.
do not summarise what you just did.
no apologies, no preamble.
ask before installing new packages.

Every project gets its own CLAUDE.md, and it loads automatically every time you open a chat in that folder. Set it once, save tokens forever.

TICK THESE OFF

✅ The checklist

Print this, screenshot it, stick it on your wall. Run through it any time Claude feels like it's eating your limit.

Before you start a task

Decided which model fits this task (Sonnet by default, Opus for the hard stuff)
Opened a fresh chat if the topic is different from the last one
Pasted only the context Claude actually needs, not your entire week

While you're working

Caught myself before pasting a 10,000 word document I could have summarised first
Switched to a fresh chat when I changed topics
Used /compact if the chat passed an hour or one full phase of work

End of session

Renamed the chat so I can find it again later
If I'll keep working on this, saved a summary handoff prompt at the bottom
Closed Opus chats I'm not actively using

One-time setup (Claude Code users)

Added a CLAUDE.md to my main project folders
Pasted the caveman line into each one
Set my default model to Sonnet in settings

DECIDE FAST

🚦 Quick decision flowchart

Is the task a quick draft, summary, or brainstorm?
→ Sonnet, fresh chat, send.

Is it a strategy call or a high-stakes write-up?
→ Opus, fresh chat, give it real context.

Is the chat already long but you want to keep going?
→ /compact or summary handoff into a fresh chat.

Are you about to switch topics?
→ Stop. New chat. Always.

GO FURTHER

💡 Three more habits I rotate in

These aren't in the reel but they save me even more.

Use Projects. In Claude.ai, Projects let you store reusable instructions and files once, instead of pasting them into every chat. Works like a permanent system prompt.
Reuse prompts, don't rewrite them. I keep a Notion page of my best prompts and copy-paste. Re-typing the same prompt every day is just paying tokens to re-explain yourself.
Batch your Opus questions. If you know you'll need Opus for three strategy questions today, do all three in one Opus chat back-to-back instead of one per day. Less context-warmup, fewer tokens.

That's the full system. Three habits in the reel, one bonus, and a checklist you can actually use 🌱

⚡ SCALING AI & GROWTH SYSTEMS?

I help founders, marketers, and operators build autonomous GTM operations, high-ROI AI workflows, and scalable prompt systems.

Book a Growth Consultation ->