Claude Code guide: pricing, setup, 10 habits | satori.

Not a developer? Stay anyway. The first half explains Claude Code in plain English, and the pricing works either way.

It's three in the morning. You've been staring at the same file for three hours.

Something's broken, but not where. The stack trace is 80 lines long. You've tried four different fixes. Nothing bit.

You paste the whole trace into the terminal and type four words:

claude "diagnose root cause first"

Thirty seconds later it has found an off-by-one in a completely different file. Writes the fix. Runs the tests. All green.

You go to bed.

This isn't a someday thing. It works today, on a ThinkPad in a flat in Marrickville. And it's what this guide to Claude Code is about: what it is, what it costs in Australia, and the ten habits that separate someone who plays with it from someone who actually gets out of it what it's worth.

---

What Claude Code actually is

Not a VS Code extension. Not autocomplete. Not ChatGPT in an input box.

Claude Code is an agent that lives in your terminal.¹ You start it with one command: claude. Then you talk to it like a colleague who happens to have read your entire codebase in thirty seconds and remembers every architecture decision you've made.

You say:

"Refactor the authentication logic into a separate service."

It reads your files. Understands the structure. Rewrites code across six files. Runs the tests. Fixes what failed. Asks whether it should commit.

It's not an assistant that suggests code. It's an agent that does the work.

And it's Anthropic's own tool, not a third-party app glued onto an API.

---

Three things that make Claude Code different

1. Massive context. It reads your whole codebase at once (up to 1 million tokens on the top tiers). No "read this file, now that file". Everything, together.

2. Real agency. It runs npm. Runs pytest. Runs git. Commits. You're not copying commands between terminal windows.

3. CLAUDE.md. A markdown file in your project root it reads automatically, every session. Your coding standards, architecture decisions, "never do this" rules live here. Write it once. It follows it forever.

It's the three together that change the game. Not any one alone.

---

A normal Thursday with Claude Code

You start the day with coffee and a ticket: "Add password reset to the auth flow."

You open the terminal, run claude, hit Shift+Tab, and ask for it, pointing at the existing login code as the pattern. Claude reads it and spits out a plan: three files change, two new ones, a database migration. You read the plan. Adjust one thing. Approve.

It starts coding. You commit every five minutes. At the end you ask for a diff with a one-line explanation per change. Two changes are in files you didn't ask about. "Roll back those two, keep the rest." Done.

Time: 45 minutes, for a feature that would have taken three hours to write and one to review.

---

The 10 habits that actually make a difference

Each solves a concrete problem. Each has saved me at least a full day. They're not tips. They're the habits.

Plan Mode in Claude Code (Shift+Tab)

Ask for a feature and Claude starts coding immediately, sometimes on a wrong assumption. Press Shift+Tab first and it delivers a plan instead: nothing gets written until you say go. This one habit prevents more bugs than anything else on this list.

/init and the CLAUDE.md context file

Tired of repeating "we use TypeScript strict, never commit without tests" every session? Run /init, let Claude generate a CLAUDE.md, then top it up with your standards and your "no exceptions" rules. It's read automatically every session. The biggest lever in the whole tool.

/clear: one task, one conversation

Start on a database refactor, drift into a frontend bug, and Claude starts mixing contexts. /clear wipes it. Rule of thumb: one conversation per task, always. Two seconds that save hours.

The reference-file technique

Describing code in words gets you something 70 per cent right and 30 per cent generic. Point at a real file instead: "Look at src/auth/login.ts. Implement password reset following exactly the same pattern." Claude picks up your naming, your error handling, your structure.

Incremental build

"Build the whole feature" gets you something that works 80 per cent of the time, and finding the other 20 takes longer than writing it yourself. Break it down: schema, endpoint, validation, form, wiring. Test after each. When something breaks, you know exactly where.

Test-first

Ask for tests after the code and they mostly verify what the code already does. Flip it: describe the edge cases, have Claude write the tests first, then implement until everything passes. TDD, Claude Code style.

Error-paste (plus four magic words)

Paste the full stack trace, not a summary, and add: "Diagnose root cause first." Without those four words it jumps on the first plausible fix and patches the symptom. With them, it reasons before it fixes.

Diff review

After a big change: "Show a diff of every file you changed. Explain each change in one sentence." Thirty seconds of reading. Saves your evening from the change it made "for consistency" in a file you never mentioned.

Undo checkpoints

Before every bigger change: git add . && git commit -m "checkpoint". Sounds trivial. But nine times out of ten when someone says "Claude wrecked my code", they hadn't committed for two hours. Commit often. Commit cheap. Free insurance.

Recovery mode

You and Claude are looping: fixing A breaks B, fixing B breaks A. Break the loop. Tell it to stop, read the last working version from git, restate the goal in one line, and start over with a different approach. Starting fresh often beats untangling accumulated fixes.

---

What does Claude Code cost in Australia?

Anthropic prices in US dollars; approximate Australian dollars below (July 2026).²

Pro costs about A$30 a month. If Claude saves you 30 minutes on a Wednesday, it's paid for itself. The rest of the month is profit.

For teams, Claude Code comes with Team Premium seats: US$100 per user per month annual (about A$150), or US$125 monthly (about A$190).

Worth knowing: your usage is shared between Claude Code and regular Claude on the same account. If the whole team hammers the chat app, that counts against the same limit as the terminal work.³

And if you want the rollout done for you: our satori-claude package is A$850 setup plus A$15 per user per month, licences from A$42 per user. Fixed prices.

---

How is Claude Code different from autocomplete tools?

You've probably tried GitHub Copilot for autocomplete or Cursor for IDE work. Claude Code is a different discipline: an agent that lives in the terminal, reads your whole codebase and actually carries out tasks across multiple files. Not a completer. A colleague.

Most teams that take Claude Code seriously keep an autocomplete tool for boilerplate too. It's not either-or.

---

Which model should you pick?

Sonnet is the default. Fast. Good at everything everyday.

Opus is heavier. Better at architecture, security reviews, multi-file changes.

Rule of thumb: run Sonnet for 90 per cent of the work. Switch to Opus when you're planning something big or Sonnet is stuck in a loop. Think builder and architect. You need both.

---

Setup in five minutes

# 1. Install (macOS, Linux, WSL)
curl -fsSL https://claude.ai/install.sh | bash

# 2. Start it inside your project
cd your-project && claude

# 3. Log in (Pro or Max)
# 4. Generate the context file
/init

# 5. Ask for something concrete
claude "write tests for user-service and run them"

Windows? irm https://claude.ai/install.ps1 | iex in PowerShell instead.

That's it. You're away.

---

Is Claude Code safe for Australian company data?

Claude Code is a US product, so this question lands on every privacy officer's desk. The short version is friendlier than you'd expect.

When you run Claude Code locally, your code stays on your machine. Only prompts and selected snippets are sent to Anthropic, and the commercial terms exclude training on business data. For most dev teams that's a comfortable position under the Australian Privacy Principles.

For stricter data-residency requirements, Claude also runs on AWS Bedrock in Sydney (ap-southeast-2), so nothing needs to leave the country. The full picture is in our Privacy Act guide for Claude, or see how we set Claude up for Australian businesses.

---

Common questions

What is Claude Code?

An AI agent in your terminal, built by Anthropic. It reads your whole codebase, writes code across multiple files, runs tests, and asks before it commits. Not autocomplete: an agent that does the work.

How much does Claude Code cost?

Pro US$20/month (about A$30), Max 5x US$100 (about A$150), Max 20x US$200 (about A$300). Teams get it via Team Premium seats at US$100 per user annual (about A$150). Usage is shared with regular Claude on the same account.

Is Claude Code worth the price?

Pro costs less than a couple of coffees a week. Save 30 minutes once a week and it's paid for itself. For seniors on Max, the return is three- to twentyfold.

What is Plan Mode?

Shift+Tab. Claude analyses your codebase and proposes a plan without writing a line of code. You review, approve, then it executes. The single best bug-avoidance habit.

How do I get started?

curl -fsSL https://claude.ai/install.sh | bash. Run claude in your project. Log in. Then /init. Five minutes, done.

What's the most common mistake?

Skipping Plan Mode. People jump straight to execution, Claude assumes the wrong thing, and three hours later 200 lines need rewriting. Plan Mode solves it in thirty seconds.

---

The close

Nobody gets good at Claude Code because they installed it.

You get good at it when you start thinking in Plan Mode, committing often, and knowing when to say stop, start over. The ten habits above aren't tips. They're what separates someone playing with a terminal from someone shipping faster than their colleagues.

And look. This isn't a someday thing. It works today.

---

Want help rolling it out across the team?

Learning it yourself is one thing. Getting 15 developers pulling in the same direction, with shared CLAUDE.md patterns and the same review flows? That's another.

That's where we come in. satori. helps Australian teams roll out Claude Code properly, not just install it. We set up your CLAUDE.md files, train the team in the ten habits, and bake the patterns into your CI/CD. Claude training runs from A$490 per person for a two-hour session to A$990 for a full day, and a hands-on working sprint is A$700 per person.

Get started or drop us a line.

---

Sources

Footnotes

1. Anthropic (2026). Claude Code overview. https://code.claude.com/docs/en/overview ↩

2. Anthropic (2026). Claude Plans & Pricing. https://claude.com/pricing ↩

3. Anthropic Support (2026). Using Claude Code with your Pro or Max plan. https://support.claude.com/en/articles/11145838-using-claude-code-with-your-pro-or-max-plan ↩