Working with AI coding assistants is fundamentally changing how we write code. Here are some early observations:
The Good
- Rapid prototyping: Especially when combinations of tools are leveraged. I’ve had a lot of success using Lovable for design, and the GitHub Copilot Agent functionality for the actual functionality.
- Instant documentation lookup: I’ve been using it like having a partner smarter than me explain documentation to me.
- Pattern recognition across large codebases: My most common personal use case so far here has been reuse of styling and functionality between different apps or contexts.
The Challenges
- Over-reliance on suggestions: While I feel like I am still learning rapidly, I find when I am tired I will overrely on suggestions. Instead of seeking to understand what it did, I will accept and then immediately test what it did. I am not sure that is an actual problem, but it is something to consider.
- Need for careful verification: There have been plenty of suggestions that either didn’t work, they introduced new problems, or similar. My lesson here so far is working to better judge what things need careful verification, and what doesn’t.
- Context understanding limitations: With the recent limitions on number of files removed from Copilot, I might take this off the list. It used to be something low, like less than 15 files. At the moment I right this the only limitation seems to be around maxing API/AI limits. If this functionality is kept I will likely remove this from the list.