Lately, I have been developing many applications in github.com/oonid, and AI coding agents have become a normal part of how I work. They are effective. I can explore an idea, build something, revise it, test it, and continue much faster than I could when I had to write every part myself.
For a software developer, this is exciting. I can spend more time thinking about architecture and problems instead of typing every line of code. Projects that might have taken much longer can now move forward very quickly.
After doing this for a while, however, I noticed something uncomfortable.
My repositories were growing faster than my understanding of them.
I could open a project and recognize the code. I knew what I had asked the coding agent to build, and I had reviewed many of the changes when they were created. The application worked and the tests could pass, yet sometimes I needed time to rebuild the larger picture in my head. I had to remember how the components were connected, why a certain design was chosen, what had changed during several AI-assisted sessions, and which decisions still mattered.
That moment stayed with me because I was not looking at someone else’s project. These were my projects. I had been involved from the beginning.
AI had helped me produce more software, but my ability to produce software had increased faster than my ability to keep all of that software in my head.
This is different from saying that AI writes bad code. In many cases, I am impressed by how useful coding agents have become. The problem can appear even when the generated code is good.
Software engineering has always required us to manage complexity, but AI changes the rate at which that complexity can arrive. A developer can now create, modify, and extend a system much faster, while human learning still takes time. We need to read, connect ideas, remember decisions, and gradually build a mental model of the system.
That difference between the speed of software production and the speed of human understanding became one of the reasons I started thinking about makna.
I want to explore whether we can keep the productivity of AI-assisted development without becoming disconnected from the software we are creating.
If AI helps us write more code, I think we also need better ways to keep understanding what that code becomes.