Soft ComputersA story published by ZDNET this week stopped us in our tracks. A writer used Anthropic's Claude Code, an AI coding assistant, to build a fully functional, paid Mac application in 11 days. No programming background. No development team. No lines of code written by hand.
The app ended up in the Mac App Store and started generating revenue. That is a real product, shipped to real customers, built by someone who would not traditionally be called a developer.
This approach has a name in the industry right now: vibe coding. The idea is that you describe what you want in plain English, the AI writes the code, and you keep refining through conversation until the product does what you need. Claude Code, built by Anthropic, is one of the tools making this possible at a level that was not realistic even 18 months ago.
Why This Matters to a Canadian SMB Owner
You may have had a software idea sitting on a shelf for years because you assumed it would cost $50,000 and six months to build. A lot of business owners in that position have either shelved the idea entirely or paid a development firm a large sum for something that missed what they actually needed.
The ZDNET story suggests the barrier to getting a working prototype is now much lower than it was. That changes the calculus on a few things.
First, proof-of-concept work is cheaper. If you want to test whether a custom internal tool would actually save your team time, you no longer need to commit to a full build before seeing something functional. A skilled developer using Claude Code or a similar tool can get you a working draft in days, not months.
Second, your own staff may already be experimenting with this. Employees in finance, operations, or sales are building small automations and internal tools using AI assistants without telling anyone in IT. That is a real governance problem we will come back to.
Third, the economics of hiring a development firm are shifting. A team that uses AI tooling effectively can deliver more, faster. When you are evaluating vendors for a custom software project, asking how they use AI in their workflow is now a reasonable question.
What the ZDNET Story Does Not Tell You
The 11-day Mac app is a compelling story. It is also a best-case scenario. The project was small and self-contained. The person building it had time to iterate, had a clear idea of what they wanted, and had no compliance requirements, no integration with existing business systems, and no team of users with different needs and edge cases.
Most software projects Canadian SMBs actually need are more complicated than that. You might need a custom quoting tool that pulls from your existing inventory database. You might need a client portal that connects to your accounting software and respects Quebec privacy law. You might need something that 15 staff members use every day, with permissions, audit logs, and a proper backup strategy behind it.
AI coding tools do not automatically handle any of that. They write code quickly. They do not architect systems, manage data security, plan for scale, or think about what happens when something breaks at 8 AM on a Monday.
The Governance Problem You Need to Address Now
Here is the issue we are seeing with Canadian businesses right now. Someone on your team reads a story like the ZDNET piece, decides to build a small internal tool using Claude or ChatGPT, and three months later that tool is handling real customer data. Nobody in IT knows it exists. There is no documentation, no backup, no security review, and no plan for when the person who built it leaves the company.
This is called shadow IT, and AI coding tools are accelerating it. We are not saying your staff should not experiment. We are saying you need a policy before the experimentation happens at scale.
At minimum, that policy should cover three things. Any tool being used to handle customer data or connect to business systems needs to go through a basic IT review before it goes live. Anything built internally needs to be documented so the business does not lose it when staff turn over. And anything that stores or processes personal information needs to be assessed for compliance with PIPEDA and any applicable provincial privacy legislation.
What We Recommend
If you have a software project in mind, now is a genuinely good time to explore it. Development timelines and costs have come down meaningfully for the right kinds of projects. We use AI coding tools in our own development work and we have seen real productivity gains.
But we recommend a structured approach rather than handing a project to an AI and hoping for the best.
- Start with a scoping conversation. Tell us what problem you are trying to solve, not what software you want built. The right solution is sometimes a configuration change in a tool you already pay for.
- Get a working prototype reviewed before you commit. If someone on your team has already built something using Claude or a similar tool, bring it to us before it goes into production. We can assess whether it is solid enough to build on or whether it needs to be rebuilt properly.
- Ask your development vendor how they use AI. A firm that is not using these tools is likely slower and more expensive than one that is. A firm that uses them without oversight is a risk. The right answer is a team that uses AI tooling within a disciplined development process.
- Put a shadow IT policy in place. If you do not have one, we can help you draft something practical that does not shut down experimentation but does give your IT environment a fighting chance of staying manageable.
The ZDNET story is real, and the tools behind it are real. The opportunity for Canadian SMBs is real too. The key is making sure the speed of AI-assisted development does not outrun the basics of good IT practice.
If you have a software idea you have been sitting on, or you suspect your team is already building things in the shadows, reach out to us. We are based in Toronto and we work with SMBs across Canada on exactly these kinds of projects.