The AI writes the code, but the team decides the direction

For the last year, I have barely typed a single line of code at work. 👨‍💻

After about ten years away from the C# .NET ecosystem, I joined a company with a large application written using that technology. Getting back felt very close to learning a completely new language.

But instead of starting by memorizing the syntax and quirks of the language and framework, and then trying to apply that to the codebase, I have almost exclusively generated code using various LLMs and AI agent systems. GitHub Copilot, Claude Code, Codex…

Code generation, combined with a fundamental, high-level understanding of how a system could and should be designed, has had an amazing impact on my effectiveness and efficiency.

It basically removed the boring and not very productive parts, like struggling with syntax and typing boilerplate code.

It also turns out to be a great way to understand what the language and framework look like today, all while exploring a codebase that has grown using all possible patterns and styles.

Having an AI agent allows me to quickly map out the terrain and make the right changes.

I am constantly asking questions like:

🧐 Is this the modern idiomatic way to solve this problem?
🤔 Am I right to assume this pattern achieves this specific goal?
👯 How does this compare to the solution in another ecosystem?
🤷 Does this code make no sense, or am I completely lost?

Asking questions from multiple angles and cross-referencing with the documentation, I think I have really accelerated my learning to understand how to write the code using the current standards and features.

But I still feel like there is a tension. An AI can explain the mechanics and syntax but generally has little concept of the why of past decisions. And rarely fundamentally challenges my direction.

Is this unnecessarily complicated? Should this be done at all? Why is this fence here? Am I even having the right idea about where we are going?

I miss the messy, slow, but invaluable feedback loop of constant discussions with colleagues. Experimenting and learning together. Even figuring out how to use the coding assistants together.

Understanding is not something static that is created once. It is something we keep building, maintaining, and sharing over time. Something that helps us make the right decisions, both small and large.

Would I prefer an AI agent or human colleagues? Why not both?

The AI helps me see the landscape and write the code faster. The team decides which direction we should walk.