you are viewing a single comment's thread
view the rest of the comments
[–] 116 points 20 hours ago (7 children)

Here's a really good article- with illustrations- about something that theoretically can and so definitely will go wrong with "agentic coding" on any sufficiently large project. The Decoherence Trap

It warms my heart to read first-hand accounts of this starting to happen at real companies. I hope the financial AI crash holds off just a little longer to make sure they've all had enough time to dig themselves into holes past the point of no return.

  • source
  • hideshow 7 child comments
  • [–] 21 points 15 hours ago

    Locked into a development model that slowly degrades our systems no matter how much effort we put into fixing them.

    This is the ultimate dream of the ai bros. We are trapped forever, using their service at an ever increasing rate.

  • source
  • parent
  • [–] 6 points 13 hours ago

    Great article, thanks for sharing!

    Really articulates the issues with vibe coding. If you don't spend the tokens focusing on coherence you're lost. And if the model can't increase coherence you're fucked!

  • source
  • parent
  • [–] 5 points 5 hours ago

    I once rewrote a large component in our codebase, which was the right move because it was unmaintainable. When I was done, I was quite proud of myself and I also think that i designed it nicely and extensible (coherent pipes) while keeping the constraints of the rest of the system in mind. Then the slop commits came in and you could watch it deteriorate so fast it is insane and sad. And additionally, I stopped caring, which is an additional loss

  • source
  • parent