Devin vs Claude Code (2026): Autonomous AI Engineer or Steered Pair-Programmer?

The short version: These two tools answer the same question — "can an AI do real engineering work?" — in opposite ways. Devin takes a whole ticket and runs it unattended, like a junior developer you delegate to and check on later. Claude Code sits in your terminal and works with you, step by step, never changing a file without your say-so. The split between them is trust: are you willing to let an agent run on its own, or do you want to review every move? For the typical Australian small team, Claude Code is the pick — it scored 20/25 to Devin's 13/25, it's cheaper, and it has the only credible local story in this category. Devin is the more autonomous machine, but its seat price and offshore-only data are the gate that keeps it out of most small Australian shops.

The tension: delegate the whole job, or stay in the loop

Most AI coding tools assist. These two do the work.

But they do it for two different people. Devin is built for the team lead who wants to hand over a clearly scoped ticket — "fix this failing test," "add this endpoint," "bump this dependency" — and walk away. The agent spins up its own sandbox, writes the code, runs the tests, debugs its own failures, and opens a pull request. You come back to a finished branch and review the result, not the process.

Claude Code is built for the developer who wants to stay in the loop. It lives in your terminal, reads your codebase, and proposes coordinated edits — but the vendor states it never modifies your files without explicit approval. You watch it work. You steer it line by line. You catch the wrong turn before it happens, not after.

That is the whole decision in one sentence. Devin asks you to trust an agent running unattended. Claude Code asks you to supervise one that never gets ahead of you. Everything below — price, fit, reliability — flows from which of those two working styles suits your team, and how much you can spend to find out.

We've reviewed both in full: Devin (13/25) and Claude Code (20/25). This piece is about choosing between them.

What they cost in Australia

Anthropic now bills Australians in Australian dollars, with its pricing page stating "Prices include 10% GST". Cognition still bills Devin in US dollars only, so the Devin figures below convert at approximately ×1.40 and are shown ex-GST. That difference matters for a small Australian business claiming input credits.

Sources: Devin pricing (Wayback, 25 Jun 2026); Claude pricing (Wayback, 28 Jun 2026). Same-day archive of the Claude page was captured on review; if the snapshot link lags indexing, the live vendor page is the primary source for every figure above.

The seat-price gate. This is where Australian small teams should pay attention. To put a three-developer team on Devin's Teams plan, you pay the US$80 base plus US$40 per developer — roughly A$280 a month before any overage, billed in foreign currency. The same three developers on Claude Code's Team Standard plan cost A$34 each, or A$102 a month all in with GST included, with Claude Code bundled. Devin's autonomy is real, but you are paying a premium for an agent that runs unattended, and for a small shop that premium is the deciding number.

GST. Anthropic adds around 10% GST at checkout for Australian subscribers, so an Australian business can claim the input tax credit on a tax invoice. Devin's pricing page makes no mention of GST, VAT, or sales tax, and Cognition's billing documentation addresses US, EU, and UK customers without any Australian provision. Whether 10% is added — and whether you get a compliant tax invoice — is not disclosed on Cognition's own pages. Check your first invoice before committing to an annual Devin plan.

Overages. Both tools meter usage. Devin charges API-rate overages once your quota is spent. Claude Code runs on a rolling five-hour window plus weekly caps, and heavy users on the Opus model burn quota fast enough to need a Max plan or the metered API. Neither cost is fully predictable at the heavy end.

Feature diff: what each one actually does

The honest summary: Devin's autonomy is its headline and its risk. It can take a GitHub issue and finish it without a human touching the keyboard — genuinely impressive, and our Devin review scored Feature Depth 4/5 for exactly that. But an agent running unattended makes confident mistakes, and Devin's output has been shown to introduce security vulnerabilities you only catch on review. The autonomy you bought is the oversight you now owe.

Claude Code is the most capable agent we've tested for the in-the-loop style. It reads a whole codebase, makes coordinated multi-file edits, runs your tests, and opens a PR — but it shows you each move and waits for approval. You trade Devin's "walk away" promise for the safety of watching the work. For most developers, that trade is the right one.

Neither tool is for non-coders. Both assume you can scope an engineering task, read code output, and review a pull request. If you can't yet code, neither of these is your starting point.

Australian fit: residency, support, GST

This is the criterion The AI Directory weighs most heavily, and it's where the two tools separate hardest.

Data residency. Neither tool hosts data in Australia. Cognition's privacy policy states Devin's data may be transferred, stored, and processed globally — including the EU, UK, and US — with no APAC region and no Australian localisation option. Anthropic is the same offshore picture today, with one difference that matters: it is the only vendor in this category actively investing locally. A Sydney office opened in 2026, there's an Australian Government MOU, and named local customers include Canva, Quantium, and Commonwealth Bank. For anything covered by APRA, the Privacy Act's sensitive-information provisions, or the CDR, get sign-off before either tool touches production data.

Support. Devin has no Australian support team and no AEST business-hours commitment; Teams gets priority support, Enterprise adds an account manager, but all of it is offshore. Claude Code has no dedicated AU support line either, but Anthropic's local presence and government engagement make it the more credible partner for an Australian business that wants a vendor it can actually reach.

GST. Covered above — Anthropic invoices with GST; Cognition's treatment is undisclosed. For a small business that claims input credits, a clean tax invoice is not a nicety, it's the difference between a deductible expense and an accounting headache.

That gap is why the scores land where they do.

Scores carried from our existing reviews; we have not re-scored either tool here. Devin's Australian Fit of 2/5 means it could not be named a category Top Pick even on a higher raw total — our methodology gates the top picks on local fit, and a tool with offshore-only data and undisclosed GST does not clear that bar.

Who buys which

Buy Devin if you lead a team, you have a backlog of clearly scoped, repetitive tickets, and you genuinely want an agent that runs unattended while you do something else. The case is strongest when you can keep several Devin sessions working in parallel on well-defined background tasks, and you have the engineering discipline to review every PR it opens. You are paying a seat premium and accepting offshore data for one thing — autonomy — so only buy it if that autonomy earns its keep across a real, steady workload.

Buy Claude Code if you are a developer or a small team that wants to stay in the loop, see each change, and keep control of what ships. It's cheaper, it's bundled into a Claude Pro subscription you may already pay for, and it's the only tool here with a real Australian story — local office, government MOU, named local customers, and GST on the invoice. The catch is the usage caps; budget for a Max plan if you'll lean on it daily.

The verdict

For the typical Australian small team, the winner is Claude Code.

It scores higher (20/25 to 13/25), costs a fraction of Devin's per-seat price, and clears the Australian-fit bar that Devin cannot. Devin's unattended autonomy is the more ambitious machine, and a well-resourced engineering team with disciplined review habits will get real value from it. But for most Australian small shops, the seat price is the gate — and behind that gate sits offshore-only data and an undisclosed GST treatment that a careful business owner shouldn't wave through.

Trust is the real decision here. If you're ready to let an agent run on its own and you can pay for the privilege, Devin is the more autonomous tool. If you'd rather steer the work and stay in control — which describes most Australian developers and nearly every small team — Claude Code is the better buy, and it isn't close.

Reviewed and compared by Adam. Last reviewed 20 September 2026. Claude pricing read from claude.com/pricing as rendered to Australian visitors on 19 September 2026 (native AUD, GST included); Devin pricing verified 28 June 2026 in USD with AUD estimates at ≈1.40, ex-GST. Neither tool replaces a developer's judgement — both can ship insecure code, so review every change before it merges. For the full breakdowns, read our Devin review and Claude Code review.