We have written about "vibe coding": the trend of using AI tools to build websites and applications through natural language. But there is a more advanced layer in this picture: AI tools aimed at professional developers, where AI does not just suggest code but actively works inside your codebase.
Claude Code from Anthropic is one such tool. We use it in our projects. Here is what we have learned.
What is Claude Code?
Claude Code is a command-line tool: it runs in the terminal, not in a browser. You install it, open it in your project folder and can then give instructions in natural language that Claude Code translates into actual changes in your files.
Unlike GitHub Copilot (which primarily suggests code line by line as you type), Claude Code can analyse an entire codebase, understand the architecture and make broad changes across many files simultaneously.
Example of what you can say: "Add multilingual support for Norwegian and English in the blog section, using the existing i18n structure in the project." Claude Code will read the files, understand the structure and make the changes, not suggest them, but make them.
What it actually does well
Routine implementation tasks. Tasks that are clearly defined and repetitive: adding a field to a form, creating a new page with the same structure as existing pages, updating configuration files, are done quickly and precisely.
Refactoring and cleanup. Giving an existing codebase a more consistent pattern, swapping one library for another, or changing a naming convention throughout a project are tasks Claude Code handles well.
Debugging with good context. When we describe a problem and give Claude Code access to relevant files, it is often able to identify the cause and suggest or implement a fix quickly.
Documentation and explanation. Claude Code can explain what a complex section of code does, create technical documentation and help with orientation in an unfamiliar codebase.
What it does not replace
Architecture assessment and strategic technology choices. Claude Code can implement an architecture. It cannot determine whether the chosen architecture is right for the project's long-term goals and maintainability. That requires experienced human judgement.
Business logic and product understanding. What the site should achieve, who the users are and what trade-offs matter for this specific client: that is context that humans bring, not AI.
Security as a discipline. AI can implement common security practices. It cannot experience the consequences of a misconfiguration or make the deep assessment required in security architecture.
What this means for SMB clients
The simple answer: faster delivery and lower hourly cost on implementation of well-defined tasks.
The more nuanced answer: the quality of what we deliver still depends on the professional judgement steering the work. Claude Code is an efficiency tool, not a replacement for professional competence.
The analogy holds: an experienced carpenter with a more powerful tool is not replaced by the tool. But they can build more in less time, and that can reduce costs for the client.
Our conclusion
Claude Code is the most impressive developer tool we have adopted in recent years. It has a real impact on the speed at which we deliver. It has not removed the need for professional competence: it has made that competence more productive.
That is an important distinction at a time when many ask whether AI is "taking over" jobs. The more accurate question is: are professionals adopting the tools that enable them to deliver more?
Want to learn more about how we use AI in website projects, and what it can mean for cost and delivery time? Get in touch.