There are some caveats to coding agents. While they can speed up many tasks, understanding what they generate, verifying correctness, and maintaining control of the overall architecture is still essential. #AI

The LLM models behind coding agents have evolved a lot. They now have a much better understanding of context and can reason about code in ways that feel closer to human thinking. This makes them far more useful for real coding tasks, and it’s interesting to see how quickly they've improved. #AI

In the beginning, I mostly used prompts to generate boilerplate code. Over time, I've started using agents and do more complex tasks from start to finish, including implementing features, suggesting optimizations and even helping with testing. It's opened up new ways to think about development. #AI