• 1 Post
  • 9 Comments
Joined 2 years ago
cake
Cake day: April 26th, 2025

help-circle
  • My personal theory is this is a natural outcome of pushing LLMs for coding too hard. The chatbot said the PR is good, I don’t need to review it myself.

    If the majority of devs are no longer fully reading and understanding their code, the bigger the chance that their PR introduces some new edge case or breaks something minor unintentionally. These small issues slowly compound until there is a critical failure.

    Hell doesn’t even need to be the majority of devs doing this. The worst least capable lazy devs in the office can crank out feature after feature that is “good enough” faster than humans can review.

    I’m betting most companies that have impraced AI coding are going to experience longer and more frequent critical failures or outages.

    Even if chat bots are 99.99% as good as a human, multiply 0.9999 by 0.9999 a thousand times and it’s 90.48% as good as a human. 10,000 times and its 36.78%.