As a software developer, I've used AI coding agents on a daily basis.
I started using coding agents about six months ago, and I've used a variety of services. Mainly, Claude code was my best; however, I don't use it anymore.
The reason I unsubscribed from Claude code was not about its productivity or service quality. It's just that I don't code as much as I used to.
I'm studying English harder than coding, so the quota of Codex CLI is sufficient for me.
There are pros and cons to using coding agents.
First, absolutely, the coding agent is robust coding assistance. We can use it in a lot of development processes, like debugging, reviewing, refactoring, planning and so on.
It not only creates codes but also thinks with us to plan more complex ideas or sketch out projects. With these agents, people who even aren't experts in coding can create their own projects for monetizing or just fun.
One of the real advancements of using it is that it has reduced the gap between our creative ideas and implementations. Nowadays, people say that coding is not a top priority, structural thinking and planning details are.
In spite of that, I still think a minimum level of coding skills and knowledges for getting self-confidence are required.
The cons are obvious. Coding agents changes code too quickly to grasp what has been changed. It causes a lot of problems in real-world projects. Additionally, developers don't review their code in projects and don't know functions as much as before. This is a serious problem called technical debt. The more code in your project that you don't know the reason for, the less likely you are to clarify what caused errors in some picky debugging situations.
In my perspective, as you know, AI coding agents are a double-edged sword. It's a trivial metaphor, but it's real.
Furthermore, the major concern of developers today is how to organize multiple coding agents at once, called the subagent or parallel programming. They want to build their own AI agent teams. It's fantastic, and I found a project named "oh-my-opencode" that helps us manage multiple agents easily.
Is this the era of productivity? What do people want exactly? I was also surprised by what a coding agent created for me in a minute. I heard development cycles don't have huge improvement, though using AI agents has been more popular. What are we gaining from AI? We have to introspect critically more than ever.