Some thoughts on using AI for coding
I’m not the biggest fan of coding with AI. I’m lucky enough to work at a company that is rolling out AI to those who want to work with it but not forcing it onto people who don’t.
I have tried out agentic coding and I just don’t get the same joy out of it that I do from hand coding things. I don’t find it satisfying to prompt an AI and wait for it to write code which I then feel I need to go in and change anyway as it never manages to do things quite how I like them to be.
What I am using it for, at least at work, is reviewing code. I have found it to be quite successful, specifically in finding security issues. I have configured a review agent with 2 subagents: a principal engineer which looks at general code quality and consistency with the existing codebase, and a security engineer which is focused on the security aspects of the review. My workflow looks like this:
- Check out the branch and spin up
opencode - Ask it to review the diff (takes roughly 5 minutes)
- Manually review the PR in Github
- Check the AI review output and add any relevant findings to the review
I think some other concerns also keep me from jumping on the AI band wagon.
- The unbearable hype that AI CEOs pump out. This causes headlines to be dominated by meaningless AI headlines and the output of my favourite YouTubers has shifted to AI focused news which I’m just not interested in.
- The environmental impact that the data centers are causing. It’s quite depressing that companies can just throw money at things like that and seem to have no repercussions.
Overall, I understand that AI is enabling people to be more productive, I just question if this is a good thing if what they’re producing is of worse quality and harder for humans to understand. The current pricing of AI isn’t sustainable and when it reaches the tipping point where regular people can no longer afford to use it, they will be stuck with their mountain of unmaintainable code.
/rant