One of the things I love most about working with Claude and Codex is that they make it much easier to step outside my comfort zone. I don't just use them to build apps I don't fully understand. I use them to explain the reasoning behind what they do, so I come away actually understanding how a platform works, from the front end all the way down to the backend.
That distinction matters to me. Over the past while I've effectively been learning backend development on top of my existing front-end experience, and the agents have let me build real platforms while learning a lot faster than I would have on my own.
The agency question
There's a lot of talk right now about losing agency. About losing touch with the craft because you're not doing the thing yourself. About trusting the agent too much.
I felt that fear too, early on. When I first started leaning on AI to write code, I worried I'd slowly hand over the part of the work that actually made me a developer.
What changed wasn't the tools. It was how I used them.
From vending machine to parallel partner
Instead of treating an agent like a vending machine, where I type a request and accept whatever comes out, I started treating it like a coding partner sitting next to me. Its job isn't only to implement. It's to help me understand why something is built the way it is.
So when it sets up a database schema, I ask why those tables, why that relationship, why JSONB instead of separate columns. When it writes an API route, I ask what's happening at each step and what the alternatives were. The implementation is still useful, but the explanation is the part that compounds.
That framing is what keeps my sense of agency intact. I'm not outsourcing understanding. I'm borrowing momentum.
Why this works for me
For context: I've been designing websites and products for over a decade. I have half a computer science degree, where I picked up a lot of front-end fundamentals, and I've spent the last five years building React sites and small applications.
So I wasn't starting from zero. I had a strong front-end foundation and a real gap on the backend. That gap is exactly where an explaining agent shines. I already had the mental hooks to hang new concepts on, so when the agent walked me through how a request flows from the client to the server to the database and back, it connected to things I already knew, and it stuck.
It's a back and forth
The critics aren't wrong that this can go badly. If you accept everything blindly, you do lose touch. You can end up with a platform you can't reason about or maintain. Too much trust is a real failure mode.
But the answer isn't to avoid the tools. The answer is to make it a conversation. Ask why. Push back. Ask for the alternative approach and the trade-offs. Sometimes I disagree with what the agent suggests, and working out why is often where the real learning happens.
Used this way, AI hasn't made me a less capable developer. It's made me a faster learner, and it's opened doors to parts of the stack I used to avoid.