AI driven coding assistants: Personal experience

AI coding assistants are coming up as tools of the trade for the future. Most of these tools are built with similar capabilities, two of which are discussed below, of which one is a welcome feature while the other not so much.

• Automated code reviews: This is where I found the power of these tools is most useful. Unlike rules based tools like SONAR which enforce predefined rules, the code review suggestions from these AI tools cater to additional capabilities such as enforcing a consistent coding style across the organization which makes the code more readable and maintainable. The tools "understand" the code and, thus being contextual, are also capable of identifying integration bugs, which would hitherto be found during integration or complete workflow testing. In my experience with one of the tool, it even correlated and suggested setting a field that I had missed in one piece of code, while it was set at another place. That was a nice surprise.
• Coding suggestions using plugins for common IDE: This is being touted as the most useful feature for developers. "Let AI code for you", but to be honest I found this lacking in multiple aspects. First, it is truly bothersome. During development, though the tools allow accepting/rejecting the suggestions, and on accepting making tweaks, the appearance of suggestions on the IDE is a true distraction. It is akin to someone constantly pulling keyboard from your hands.Second, coding isn't just about writing code. When I write code, I have a mental picture of the flow and thus the code just streams out. Such suggestions, break that thought process. It actually led to reduced efficiency during implementation.

So while I really like the code review feature, which allows an alternative view of the implementation and identifies potential gaps, I would rather stay away from plugging it into my IDE.

Other experiences welcome in comments...