Burden of Reviews
Written without LLM assistance. Unreviewed, not even by me.
I previously wrote about working with LLM superusers. Since then, I’ve drastically reduced my LLM usage to near $0 as a form of resistance against this growing mound of complicated and unnecessarily complex software. But it also goes beyond a matter of principle.
Writing hand-written code, in my opinion, shows respect towards your colleagues. It demonstrates that you respect their limited time and energy and that you care. That you care about the quality of your work and your craft and perhaps most importantly, that you care about reducing the burden of cognitive load on your teammates.
When you open a pull request or write a review comment that’s unreviewed LLM output, you’re putting the burden of the knowledge work on your colleague. This is a very asymmetric work effort that is neither cooperative nor sustainable.
Writing unreviewed code and messages with LLMs, if I can even call it that, for the sake of productivity has a highly negative short- and long-term impact on the software and the team. Not only are you increasing software complexity, you are also increasing cognitive load upon your peers and thereby decreasing their productivity.
The root of the problem is the trap of constantly increasing shareholder value. Teams everywhere are given OKRs in unreasonable timelines that force individuals to use LLMs to hasten delivery at the expense of domain knowledge. The LLMs get smarter and the developer gets stupider.
I think that if you are privileged enough to work at your own pace, you should. Non-LLM users shouldn’t have to suffer because LLM users are pushing out slop. And in any case, those of us who still use our brains will probably be the ones called when things get out of hand (because they will).
–
Oh, technically, these LLM superusers are also getting compensated more in the form of AI credits. What if we incentivized bonuses for non-LLM users?