A database migration is a fairly ordinary place to notice a change in your working life. But it’s one of the clearest examples I have of AI being useful. I can describe a set of fields and relationships, get a first version of the code, and spend more of my attention on whether those decisions make sense.
I’ve used that approach in my Laravel projects, and I’ve been working with AI while rebuilding this website. Having something in front of me sooner gives me more opportunities to question it, change direction and try again. It also makes it very easy to accept something before I’ve understood it properly.
Coherent Lab’s article on AI in web development in 2026 looks at uses across coding, design, testing and maintenance. It prompted me to think about what those changes mean in my own work, beyond the promise of getting things done faster.
Small jobs add up
In my Laravel and React experiments, I’ve used AI to help draft migrations, seeders, relationships and initial validation rules. A task manager with boards, lists, cards and subtasks involves a lot of fairly similar code. Being able to describe that structure and work from a draft is useful.
There are still decisions in those apparently simple tasks. Does a card belong to a list or directly to a board? What should happen when the parent record is deleted? Who is allowed to see or change it? A generated migration can be perfectly readable while encoding the wrong answer to one of those questions.
That’s where my development experience helps. I have something to compare the suggestion against. I can follow the data through the application, question a relationship and recognise when a shortcut will make the next part harder.
I don’t have a reliable percentage for how much time AI saves me. The benefit I can describe is more practical: less time typing familiar structures, with a draft I can start examining and adapting.
This website needed a point of view
Rebuilding my own site has been a good example of the decisions that remain. I used AI during the process, but the early direction was too elaborate for what I wanted. It needed to feel more personal and give the writing room.
I kept coming back to a narrow column, black and white, a small logo and straightforward navigation. Portfolio entries could be text links with descriptions. Images and small videos could sit beside the writing without taking over the page.
Being able to produce another version quickly helped. Knowing what to ask for came from my own taste, the references I chose and what I wanted the site to do. I had to be willing to reject work that looked finished.
That matters to me as a designer. A polished layout is easy to react to, but I want to know what it communicates. Can someone understand who I am? Can they find an article? Does the page give them a reason to keep reading? Those questions gave the design a direction.
A convincing result still needs checking
The same website also needed work on performance, accessibility, metadata and the editing experience. A minimal page still has fonts to load, images to size and controls that need to work with a keyboard. Its appearance tells you very little about whether those things have been handled.
With WordPress, I also care about what happens after the page is built. I want to be able to open Gutenberg, change the content and carry on using the site. A result that looks right but is awkward to edit leaves me with another problem.
These are the questions I want answered when I’m reviewing work with AI:
- Does it solve the problem I described, including the less convenient cases?
- Can I understand the change well enough to maintain it?
- What happens when the form is incomplete, the request fails or there is no content?
- Can someone use it on a small screen and with a keyboard?
- What have we actually checked in the browser, and what are we still assuming?
I find that last question particularly useful. An explanation of why something should work is a starting point. Opening the page and trying it gives me different evidence. I want both.
What happens to the way we learn?
I’ve enjoyed helping a junior developer grow into working with WordPress and React. Part of that process was giving someone room to attempt a problem, get stuck, ask questions and gradually need less help.
I think about that when I use AI. There’s a real opportunity to explore something unfamiliar with help close at hand. There’s also a temptation to keep accepting answers until the thing runs, without building a clear picture of how it works.
I wouldn’t want someone starting out to feel they have to avoid these tools. I would want them to explain the result, change it and trace what happens when it breaks. Those are useful habits for me too. Experience doesn’t make me immune to accepting a convenient answer.
The part I worry about is whether we’ll keep giving people time to learn. If every saved minute becomes an expectation to produce more, there’s less room for the conversations and experiments that help someone become a capable developer.
Why I’m widening my skills
In 2020, I chose to concentrate on development. I wanted to go deeper into one discipline, and that time gave me a stronger foundation. In 2025, I began making room for design again, including projects such as Harbour Roastery and Wilder & Root.
AI is one reason I’m thinking carefully about where to put my effort now. I want to get better at design, accessibility and understanding the person using a product. I enjoy that work, and it helps me make better development decisions.
I’m especially drawn to application design. A useful interface depends on understanding the task, choosing the right information and making the next step clear. My knowledge of code helps me think through how it will behave. My design experience helps me notice where that behaviour will be confusing.
I see critical thinking as part of the same work. Asking a precise question, noticing an assumption and deciding what evidence would settle it are skills I want to keep practising. They matter when I’m reviewing an AI suggestion or discussing a brief with another person.
The changes reach further than code
I can be pleased that a tool helps with my work and still feel uneasy about the wider direction. I’ve written before about AI and career uncertainty. I still don’t think anyone can promise that learning a particular tool will make a career safe.
I’m interested in the possibility of more people being able to try an idea, build a small tool or understand a system they previously found inaccessible. I also worry about who gets access, who controls the tools and whose work becomes less valued. The benefits won’t necessarily be shared evenly.
For my own part, I want the time I gain to make room for better work and more learning. Sometimes that will mean exploring a new approach. Sometimes it will mean spending longer on a small detail that would otherwise have been rushed.
I still want to be able to open the code six months later, understand a decision and explain it to someone else. As AI becomes a larger part of the process, that feels like a useful standard to hold myself to.