AI Agent Partnership
In the beginning, I mostly used AI as a chat partner. I would ask questions about things I was curious about, explore ideas, get help solving a problem, or ask for some code when I got stuck. I still use AI that way all the time, but somewhere along the way, something changed. On certain projects, AI has started to feel less like something I talk to and more like someone I'm actually working with.
Sometimes it is still just a conversation. Other times, it feels much closer to a working relationship. I'm still figuring out what that relationship should look like, and I'm starting to see that the way I work with AI depends a lot on what I'm trying to build. I'm still very much at the beginning of exploring this, but so far I've found myself experimenting with two quite different ways of working.
AI Builds, I Manage
For some projects, I'm more interested in what I can build than in writing or understanding every line of code myself. I start with an idea, describe what I want, and let the agent handle most of the implementation. Then I use what it has built, see how it feels, and give feedback on what I want to change.
In that setup, my role feels much closer to that of a product owner. I focus on the direction, the features, the experience, and whether the end result is actually useful. I recently built an Anthropic Certification Preparation application to help me study for a certification. The agent wrote the code while I focused on using the app and improving it through feedback. If something didn't work the way I wanted, I explained what I was looking for, and we iterated from there.
For low risk or internal applications that I'm building for myself, I'm comfortable letting the agent handle most of the implementation. That means I can spend more of my time thinking about what I actually want the application to do. An idea can go from "this would be useful" to something I can actually try much faster than before, and for this kind of project, that feels like a very natural way to work.
AI Partner - Together as Team
For public-facing projects, more important or higher-risk work, or simply projects where I want to truly own the code, I work differently. For me, owning the code means understanding the codebase, ideally every line of it, even if I'm not always the person typing those lines.
This blog is an example. In my previous post about Website Upgrade, I was moving away from the traditional WordPress setup to something built natively with Next.js and TypeScript. I originally built this website line by line. Along the way, I used AI much like I had before: asking questions, discussing ideas, getting suggestions, and asking for help whenever I got stuck.
Today, that relationship is becoming more collaborative. The agent can find bugs, suggest improvements, refactor code, and sometimes write parts of the implementation for me. But there is one thing I still want to keep: understanding. Even when I don't write the code myself, I want to know what it does, why it is there, and how it fits into the rest of the codebase.
That is what makes the code feel like mine. Not because I typed every character, but because I understand it, review it, and make the final decision about what becomes part of the project. I really enjoy working this way too. It feels more like having another developer beside me, one I can discuss ideas with, challenge, learn from, and build with.
Still Experimenting
I'm still experimenting and learning as I go. These two ways of working are probably just the beginning. Maybe the next step is learning how to build more automated workflows with multiple agents, almost like managing a small team, where I delegate tasks and spend more of my time focusing on direction. That idea feels exciting, but I'm not there yet.
For now, I'm learning one project at a time. On one project, I might happily let an agent handle most of the implementation while I focus on the product. On another, I want to stay close to the code and understand every important decision.
What I really like about this new way of working is how much faster AI helps me move from idea to prototype. As a developer, it gives me more room to explore my own ideas, test things quickly, and build projects that might otherwise have stayed in my head for much longer. And maybe that is the part I find most exciting. AI is not only helping me write code faster. It is changing the way I think about building things in the first place. I'm still figuring out what that means for the way I work, but I'm enjoying the process.