How AI tools are starting to change web development

Two months since ChatGPT launched and you'd think every developer on the planet was about to be replaced by a chatbot. The hype is extraordinary. But what's the reality like when you actually sit down and try to use these tools on real client work?

I've been experimenting with ChatGPT and AI coding tools on a few internal projects over the past few weeks, and the reality is: they're useful, but they're not magic. ChatGPT is genuinely good at writing boilerplate, the repetitive stuff you'd normally copy-paste from a previous project and adapt. Setting up a basic API route, writing test scaffolding, generating TypeScript interfaces from a known data shape. It saves time on the boring bits, and that's worth something. But is it changing how I work day to day?

Where it falls over

The moment you need anything context-specific, a component that integrates with a particular CMS's content model, or a function that needs to respect a client's specific business logic, the suggestions become unreliable. You spend as much time verifying and correcting the output as you would have spent writing it yourself. And that's the trap: it feels fast, but if you're not carefully reviewing every line, you're introducing bugs you didn't write and don't fully understand.

ChatGPT is a different beast. It's brilliant for explaining concepts, drafting documentation, and rubber-ducking problems. I've used it to help write technical briefs and it's genuinely cut that work in half. But ask it to write production code for a specific framework version and it'll confidently give you something that looks right but uses deprecated APIs or patterns from two versions ago.

What this means for agencies

For me, the approach is pragmatic. I'm using these tools where they genuinely speed things up, boilerplate, documentation, research, brainstorming. I'm not using them to generate client-facing code without thorough review. The liability question alone should give any agency pause: if an AI tool generates code with a security vulnerability, you're still responsible. Think about that for a second.

The developers who'll benefit most from this wave aren't the ones blindly pasting AI output into production. They're the ones who already know what good code looks like and can use these tools to get there faster. AI makes senior developers superhuman, genuinely. One experienced developer working with AI tools can do what used to take three or four people. And yes, that does replace headcount. But it does not make junior developers into senior ones. You can't replace a decade of experience with a chat prompt. A senior developer can lead AI agents because they know what right looks like. A junior can't verify AI output because they don't have the mental model yet. That's worth being honest about.

I suspect the real impact won't be visible for another year or two, once these tools mature and we collectively work out which workflows they genuinely improve versus which ones just feel impressive in a demo. For now, I'm experimenting carefully and keeping our expectations grounded.

If you're wondering how AI tools might fit into your own development workflow, or you're an agency trying to figure out a sensible policy, drop me a line, happy to share what I've learned so far.