How I Actually Use AI in My Frontend Workflow

I want to be honest about this, because most writing about AI and development falls into one of two camps: either it's going to replace us all, or it's a glorified autocomplete that can't really think. My experience using it every day for the past couple of years is more mundane than either position, and I think that's actually worth saying.

The tasks where AI tools genuinely save me time are specific and consistent. Boilerplate: generating the skeleton of a new component, a TypeScript type, a form with validation logic I've written a hundred times before. Explaining unfamiliar code: pasting in a function I don't understand and asking what it does. Looking up API details I keep forgetting: the exact signature of a method, the right way to configure something in Next.js, how a particular CSS property actually behaves. These are real time sinks, and AI handles them well.

The tasks where AI doesn't help — or actively gets in the way — are the ones that require understanding the system. Deciding where a piece of state should live. Figuring out why two parts of the codebase are coupled in a way that makes changes painful. Knowing when a component has grown too large and what the right split would be. These aren't questions with searchable answers. They require holding the whole system in your head and making a judgment call. AI can suggest things, but the suggestions are based on what the code looks like locally, not what the system is trying to do. That context lives in your head, not in the file.

I use Claude Code for a lot of my current work, including building this site. What I've found is that it's most useful when I know exactly what I want and just need the execution done quickly. Give it a clear task with enough context and it produces good work fast. Ask it to make an architectural decision and it'll give you an answer, but you shouldn't necessarily trust it — it doesn't have the history, the constraints, the business context that make the answer right or wrong.

The thing I keep coming back to is that the developers who'll do best with these tools are the ones who have strong fundamentals. Not because the tools are going away, but because the tools amplify whatever you already know. If you understand how React renders, you can use AI to write components faster and catch the mistakes it makes. If you don't, the mistakes just end up in your codebase.

There's also something worth saying about the experience of using these tools as a learner. I'm actively building my React and Next.js skills right now, and I've made a deliberate choice to use AI assistance but to understand everything it produces. When it writes something I don't fully grasp, I ask it to explain. When it suggests a pattern I haven't seen before, I look it up. The goal isn't to ship code I don't understand — it's to move faster through the parts I already know so I can spend more time on the parts I'm still learning. That balance matters a lot.