How I use AI in my work (and where I don't)

I've been building websites for over 15 years, and AI has changed how I work more than anything else in that time. Clients are starting to ask about it, so here's an honest look at how I use it, where I don't, and what it means for you if we work together.

What I use it for

Every website or web app has a layer of groundwork that needs doing properly but doesn't need much creative thought. That's where AI earns its keep. I use it to:

• Scaffold the routine setup work at the start of a project: database migrations, models, custom WordPress blocks and Statamic blueprints

• Handle repetitive refactoring that used to take hours, like renaming and restructuring code consistently across a whole project

• Make sense of legacy code, older websites and plugins that nobody documented

• Run security reviews on my code, checking for common vulnerabilities like SQL injection, cross-site scripting (XSS) and broken access control that could expose customer data, before anything goes live

• Review every change for bugs and edge cases, as an extra pair of eyes alongside my own code review and automated tests

• Talk through a tricky architecture problem when there's no one else in the room to bounce ideas off

The security and bug checks don't replace my own judgement, but they're thorough, they don't get tired, and they regularly spot something worth a second look. Working as a freelancer, that extra check matters. And the time saved on groundwork goes back into your project: more attention on how easy the site is for your team to update, how fast it loads, and the details your visitors notice.

Knowing the tools

Anyone can open an AI tool, describe the website or app they want, and get something back. That's genuinely impressive, and it's why so many people are trying it. But there's a big gap between "it made something" and "it made something you can rely on".

I've spent a lot of time researching and working hands-on with a wide range of AI tools: coding agents that work directly in a project, the harnesses that run them, and the different large language models (LLMs) behind them. They're not all the same. Some models are better at reasoning through a tricky bug, others are quicker for simple changes. Some tools are great at working across a whole codebase, others are better kept to a single file. Each one has its own blind spots, and knowing them is what stops you trusting something you shouldn't.

Getting good results consistently comes down to how you set them up: giving the AI the right context about your project through instruction files, the coding standards it should follow and the tests it needs to pass, breaking the work into sensible steps, and reviewing what it produces properly. That's not something you pick up from a single conversation with a chatbot, and it's where my experience as a developer and my time with these tools come together.

What I don't hand over

AI is good at producing something that looks right. Knowing whether it actually is right for your website is a different question.

It doesn't know your business. It doesn't know that your property listings come from another system every hour, that your team finds a certain page awkward to update, or that a particular plugin has caused problems on other projects I've worked on. It has no idea why something was set up a certain way years ago. That understanding comes from experience and from actually knowing your project.

Writing software is a professional skill, and AI doesn't change that. It's a powerful tool, but it still needs someone who understands software architecture, security, performance and maintainability to steer it. So nothing goes live on a client's site without me reviewing it, understanding it and being happy to put my name to it. And if something goes wrong on launch day, you've got a real person who knows your site, not a chat window.

Built something yourself with AI?

I'm getting more enquiries from people who've used AI tools to build a first version of their idea themselves. That's a great way to test an idea, and often it works surprisingly well.

The trouble usually starts when it needs to handle real customers, payments or personal information. That's when missing tests, security vulnerabilities, data protection (GDPR) and a database structure that can't grow start to matter. Taking something like that, auditing the code and making it production-ready is work I'm happy to help with.

In short

AI has made it easier than ever to get something up on screen. Getting it live, secure and still working well a year from now is a different job. I think you should want your developer to use AI; just make sure there's someone behind it who understands your project and takes responsibility for the result.

Got a project in mind, or something you've built with AI that needs a second pair of eyes? Get in touch.