Last week, Xero CEO Sukhinder Singh Cassidy made headlines when she told the Australian Financial Review that AI couldn’t easily clone her company’s accounting platform. She’d apparently tested this herself using Anthropic’s Claude Code.
The tech press had a field day. Luke Hopewell at Switzer Daily decided to test the claim — and built a working Xero alternative called “OpenBooks” in about 23 minutes.
So who’s right?
What Singh Cassidy actually said
According to reports from Xerocon 2025, Singh Cassidy’s argument wasn’t that AI can’t write accounting software. Her point was more nuanced:
- Ecosystem complexity — Xero isn’t just an app. It’s thousands of integrations with banks, payment processors, payroll providers, and third-party apps.
- Network effects — Two million users. Accountants trained on the platform. Workflows built around it.
- Deep infrastructure — Acquisitions like Melio, partnerships with Stripe and GoCardless. Years of regulatory compliance work across multiple jurisdictions.
She’s arguing that Xero is a platform, not a product. And platforms are harder to clone.
What Hopewell actually built
Hopewell’s experiment is worth examining. Using Claude Code, he scaffolded an app with:
- Invoicing
- Bank reconciliation
- Expense tracking
- Payroll basics
- Stripe integration for payments
- OCR for receipt scanning via Tesseract
Cost comparison: Xero runs $35-78/month. OpenBooks could theoretically run for $0-60 per year.
But here’s the part Hopewell was honest about: “just because you can, doesn’t mean you should.” The code worked, but maintaining it, handling edge cases, dealing with compliance, supporting real users? That’s a different story entirely.
What coding agents can actually do in 2026
The landscape has shifted dramatically. Claude Code now powers roughly 4% of GitHub public commits, with projections suggesting 20%+ by year’s end.
Modern coding agents like Claude Code, GitHub Copilot, and Cursor can:
- Read and understand entire codebases
- Plan multi-file changes with dependency awareness
- Write, test, and debug code autonomously
- Run terminal commands and iterate on failures
- Operate in parallel “agent teams” for complex tasks
With Opus 4.6’s 1 million token context window, these tools can hold an entire mid-sized codebase in memory and reason about it holistically.
Building a functional MVP of any software product is now genuinely achievable in hours, not months.
But here’s what AI still can’t do
1. Solve regulatory compliance at scale
Xero operates in dozens of countries. Each has different tax rules, reporting requirements, and banking regulations. That’s not a code problem — it’s a legal, operational, and relationship problem built over years.
2. Build trust and distribution
Software is easy. Getting 4 million subscribers to trust you with their financial data is hard. Getting 1 million accountants to recommend you to their clients is harder.
3. Handle the long tail of edge cases
Real-world accounting software deals with bizarre scenarios daily. Partial payments split across currencies. GST on imported goods with specific exemptions. Payroll for employees who work across state lines. Each edge case represents years of customer feedback and iteration.
4. Maintain and evolve a living product
Hopewell’s OpenBooks works today. But Xero ships updates constantly — new integrations, regulatory changes, security patches, UX improvements. Keeping a solo AI-built project current is a full-time job.
The real question isn’t “can AI clone Xero?”
It’s: “What happens when building software becomes essentially free?”
If anyone can build a basic accounting app in an afternoon, the moat shifts entirely to:
- Distribution — Can you reach customers?
- Trust — Will they give you their financial data?
- Ecosystem — Do you integrate with everything they need?
- Support — Who do they call when something breaks?
Singh Cassidy is betting that Xero’s answers to those questions are strong enough that code commoditisation doesn’t matter. And honestly? She’s probably right.
What this means for everyone else
For startups, the lesson is clear: your competitive advantage can no longer be “we built the software.” That’s table stakes now.
For established SaaS companies, it’s a warning: if your only moat is code complexity, you don’t have a moat.
And for developers? The interesting work is no longer writing CRUD apps. It’s building the things AI still can’t replicate: deep integrations, regulatory expertise, user trust, and genuine product insight.
Xero can be cloned in 23 minutes. What makes Xero Xero cannot.
We build operations software at EQUOS9. Like Xero, our moat isn’t the code — it’s understanding how real warehouses, freight networks, and manufacturing lines actually work. See what we’re building →