The Burden

This short post is a coalescing of two thoughts in my brain at the moment:

1. The burden of responsibility when reviewing an AI generated change set from a colleague.
2. The surface level understanding of codebases when AI is heavily involved.

You could write essays upon essays for those statements. But I’m going to be quick.

Burden of responsibility

I happened to land on the AI Policy for Dash, the GitHub CLI Dashboard extension. It hit the nail on the head for me with this paragraph:

Every discussion, issue, and pull request is read and reviewed by humans. It is a boundary point at which people interact with each other and the work done. It is rude and disrespectful to approach this boundary with low-effort, unqualified work, since it puts the burden of validation on the maintainer.

Time may prove us wrong on this point, but I think I’m currently very much aligned to this thinking. Historically, you had some level of confidence that the author had written, and understood, the code being presented for review (Give or take some Stack Overflow involvement). You roughly had an understanding of the skillset from the author too etc. With AI, more and more I wonder if the author has read the change set before asking for a review. Even worse, do they understand it themselves? This seemingly, puts even more burden on the reviewer to validate everything about the change. Is this slowing a team down overall?

AI is a great tool, when used appropriately, but I’d love to have the confidence back that changes are understood before being sent for review. It’s unclear how this can be achieved, with the current tooling/processes. Maybe annotating the changeset with some rationale is a good start?

Surface level understanding

I’ve been vibe coding a little CLI tool for myself to read RSS articles. I’ve reviewed each change the robot has created. The other day I wanted to hand code something myself, and it dawned on me that I just didn’t have the solution in my head. Many other projects I have, I can dip in and out and remember exactly how it all works. That wasn’t the case with this AI driven project. It took an awfully long time to be in a position to code the feature I wanted to implement. Overall, I started to wonder if using AI has really been saving me time in the long run.

And that is all I have to write at the moment. The journey continues!