If you write code in Australia in 2026, odds are you’ve already let an AI finish a function for you this week. Maybe it nudged you through a tricky Pandas join, maybe it scaffolded a React component, maybe it talked you through a Terraform stanza that you couldn’t be bothered Googling. The question is no longer whether to use an AI coding assistant — it’s which one, and how much of your client’s repo you’re comfortable shipping off to a US data centre while you’re at it.
Our AI desk lead Priya Raman has been switching between the big three — GitHub Copilot, Cursor and Claude Code — for months across a mix of Python data work, a Next.js side project and some grumpy legacy .NET. Josh on the PC desk has been pair-programming with them on his home lab, and Dale keeps stress-testing them on documentation rewrites. This is what we’ve landed on, with Aussie pricing, privacy and the realities of mid-size SaaS work front of mind.
The three tools at a glance
GitHub Copilot is the incumbent. It lives inside VS Code, Visual Studio, JetBrains, Neovim and Xcode, and in 2026 it ships with a mix of models — OpenAI’s GPT-5 family, Anthropic’s Claude Opus 4.x, and Google’s Gemini 2.5 Pro — that you can swap between for chat and agent mode. It’s the default at most Australian enterprises already on GitHub Enterprise, which is to say, most of them.
Cursor is an opinionated fork of VS Code built around AI from the ground up. Same keybindings, same extensions, but the AI is woven into the editor rather than bolted on. It also routes between frontier models — predominantly Claude Opus 4.x and GPT-5 — and has become the cult favourite among front-end and full-stack devs at places like Canva and a long tail of Sydney and Melbourne SaaS startups.
Claude Code is Anthropic’s terminal-native agent. There’s no IDE — you run it in your shell, point it at a repo, and it reads, edits, runs tests and commits on your behalf. It’s powered by Claude Opus 4.x and is the tool of choice when Priya wants to hand off a multi-file refactor and go make a coffee. It also now plugs into VS Code and JetBrains via official extensions, but the CLI is still where it sings.
Pricing in AUD
All three publish in USD and bill in USD, so what you actually pay depends on the day’s exchange rate and your card’s foreign transaction fee. Approximate AUD figures at the time of writing:
- GitHub Copilot: around A$15/month for Pro, A$30/month for Pro+, and roughly A$60/user/month for Business with admin controls and IP indemnity. Free tier still exists with monthly completion and chat caps.
- Cursor: a free Hobby tier, Pro at about A$30/month, and Business around A$60/user/month with SSO, audit logs and zero-data-retention guarantees.
- Claude Code: bundled into Claude Pro (~A$30/month) and Claude Max (A$150–A$300/month depending on tier). Usage-metered API access is also available if you’d rather pay per token.
For a freelancer billing $150/hour, any of these pays for itself in roughly one saved coffee break per month. For a 40-dev shop, the calculus is more about which one keeps everyone shipping — and which one your security team will actually sign off on.
How they feel to use day-to-day
Copilot is still the king of incremental work. You type, it greys out the rest of the line, you hit Tab. It’s quiet, fast, and rarely intrusive. The chat panel and the new agent mode have closed a lot of the gap on bigger tasks, but Copilot’s sweet spot remains the steady drip of completions while you stay in flow. If your day is mostly editing existing services — patching an Express route, adding a column to a Django model, tweaking a Power Automate flow — Copilot is the lowest-friction option.
Cursor wins on IDE fluency. Its Tab model is genuinely uncanny: it predicts where you’re going next in the file and offers multi-line jumps that feel like the editor is reading over your shoulder. Composer (its agent mode) handles multi-file edits without breaking a sweat, and the inline diff UI makes reviewing AI changes far less stressful than scrolling through a chat window. If you live in VS Code anyway, switching to Cursor is a half-day exercise. Em on our mobile desk swapped over for a React Native rebuild and hasn’t looked back.
Claude Code is the agentic outlier. You give it a task — “add OAuth to this Fastify app, write tests, run them” — and it spelunks the codebase, edits multiple files, runs your test suite, and reports back. It can use the terminal, hit the network if you let it, and chain tool calls in ways the other two are still catching up to. For Python data pipelines, Go services and any work where the answer requires reading a lot before writing a little, Priya’s take is that Claude Code is currently in a class of its own.
Language and stack coverage
All three are strong on the bread-and-butter languages most Aussie devs touch: JavaScript, TypeScript, Python, Go, Java, C#, SQL. Where they diverge:
- Python: All three are excellent, but Claude Code’s ability to actually run scripts and inspect dataframes gives it the edge for data work. If you’re wondering why everyone reaches for Python by default, our earlier piece on why the Python programming language is so popular still holds up.
- TypeScript / React / Next.js: Cursor’s Composer is hard to beat here.
- Rust: Growing fast in AU infra teams. Claude Opus 4.x (so Cursor and Claude Code) handles borrow-checker gymnastics with notably more patience than GPT-5.
- .NET / C#: Copilot’s Visual Studio integration is still the most polished experience for the .NET shops dotted around Canberra and the big four banks.
- Go: A wash. All three are competent; Claude Code’s agent loop suits Go’s “write small, test often” rhythm.
- Swift / Kotlin: Copilot via Xcode and Android Studio remains the path of least resistance for mobile teams.
The honourable mentions
The big three aren’t the only show in town. Sourcegraph Cody is a serious option for enterprises that need on-prem code search wired into their assistant — popular in regulated AU industries. Windsurf (formerly Codeium) is a credible Cursor alternative with a more generous free tier and is gaining traction with students and indie devs. Tabnine still leads on fully self-hosted, air-gapped deployments, which matters if you’re working with classified or health data. And JetBrains AI Assistant is the path of least resistance if your team lives in IntelliJ, PyCharm or Rider and doesn’t want to leave.
Privacy, data handling and Aussie client code
This is where most of our reader emails land, so let’s be direct. All three providers offer paid tiers that promise not to train on your code. Copilot Business and Enterprise, Cursor Business, and Claude Code on the Team/Enterprise plans all contractually guarantee zero data retention and no model training on your inputs. On the free and personal tiers, the picture is murkier and you should assume your prompts can be reviewed for abuse and may, in some cases, contribute to model improvement.
For Australian developers handling client code, the practical questions are: where does the data go, who can see it, and what happens if it leaks? All three route prompts through US infrastructure by default. None of them are currently hosted in Australian regions for the consumer tiers. If you’re handling personal information under the Privacy Act, the OAIC’s guidance on cross-border disclosure under APP 8 is essential reading before you paste a customer record into a prompt window. For broader threat modelling — including supply-chain risk from AI-generated dependencies — the Australian Cyber Security Centre has been publishing increasingly practical guidance for development teams.
Our rough rubric for AU teams:
- Open-source or your own side project? Use whichever you like.
- Client work for an Aussie SMB? Business tier of any of the three, with zero-retention enabled, is fine for most engagements — get it in writing in your contract.
- Government, health, finance or anything covered by the Notifiable Data Breaches scheme? Enterprise tier, SSO, audit logs, and a documented DPIA. Consider Cody or Tabnine for self-hosted options.
- Defence or anything classified? None of the cloud tools. Full stop.
Where each one shines
- Pick Copilot if your team is already on GitHub, you want low-friction inline completions, and you need enterprise procurement to be easy. The Microsoft account managers will turn up to your office; the others won’t.
- Pick Cursor if you live in VS Code, you do a lot of front-end or full-stack TypeScript work, and you want the AI to feel like part of the editor rather than a sidebar.
- Pick Claude Code if you’re comfortable in a terminal, you do a lot of multi-file refactors, codebase exploration or agentic tasks, and you want the tool to do the boring parts end-to-end while you review the diff.
Plenty of our team run two of them. Priya keeps Copilot on for inline completions and fires up Claude Code in a second pane when she needs to hand off a bigger job. Josh runs Cursor as his daily driver and uses Claude Code for one-shot CLI tasks. There’s no rule that says you can only pick one — and at A$30 a month each, two subscriptions is still cheaper than a single billable hour lost to yak-shaving.
The remote-work angle
One underrated benefit of agentic tools like Claude Code: they’re brilliant for working from a laptop on a slow regional connection, because the heavy lifting happens server-side and you’re really just streaming text. We covered the broader kit question in our guide to reliable mobile tools for work-from-home success — pair a decent 5G hotspot with a CLI agent and you can genuinely ship from a caravan park in Esperance.
Storage matters too, because all three tools encourage you to keep more code and notes in cloud-synced folders. If you’re juggling personal and work accounts, our explainer on why Microsoft’s OneDrive name isn’t exclusive is a useful reminder that “the cloud drive” your assistant reads from is rarely as private as you think.
Final thoughts
The honest answer is that all three are good enough that the wrong choice won’t ruin your year. GitHub Copilot is the safe, sensible default. Cursor is the editor you’ll enjoy using. Claude Code is the colleague that does the unglamorous work while you make a flat white. Our team’s current split is roughly a third on each, and we expect that to keep moving as the models leapfrog one another every few months.
The bigger decision isn’t which tool — it’s how seriously you take the privacy posture around it. Get the Business tier. Turn on zero data retention. Write the policy. Tell your clients. And if you’re not sure whether you’re allowed to feed a particular repo to a US-hosted model, ask before you paste, not after. The tools will keep getting better; the obligations on you as the developer aren’t going anywhere.
