AI Code Reviews
One of the downsides of using AI to code is the increased need to test and do code reviews. We are talking “vibe coding”, but producing production quality code. Such code needs to be verified before going into production - same as if it were written by a human. Only with AI coding, you are producing MORE code at faster rates.
AI to the rescue. Not “eliminate the human” but give AI the first pass to reduce the human effort.
Today I had a need to write code for a client, but my client machine is not currently configured properly. While that’s being sorted out, I had Claude Code write a generic sql query to json utility. I had it developed to work locally for me with user/password and on the client machine via single signon. I directed it to use as few libraries as possible.
Out pops working code after a time. Including using Warp to set up git locally and connected to my GitHub on my new LinuxMint machine, and installing ODBC on Linux. I test it, it works.
THEN I go to Codex Cloud which is connected to my Github. I selected the repo and gave Codex the following prompt:
“You are an expert python developer, sql guru and technical educator. Starting with the productDefinition.md, do a review of the code and documentation. Look for best practices, security, and clarity”
Judge for yourself if you think this review was beneficial: https://lessoflee.pages.dev/posts/ai-code-review/
I certainly thought so. And I’ll feed it back into Claude Code to make the improvements.
I’ll have AI review that and when AI says it’s good to go — THEN — I will also review the code. This isn’t code for my home lab, this is code that clients pay me to write. It doesn’t matter that I used a tool to help write it. It does matter that I verify the quality.
Took me about an hour to work through this with Claude Code plus Codex Code review. Working code, with documentation and examples. It would have taken me at least a day, probably two to hand code, document and review. A better python programmer who does work in this exact space could have done it in less time - but not less than time than that same programmer using AI effectively.