you are viewing a single comment's thread
view the rest of the comments
[โ€“] 95 points 7 months ago (6 children)

Reviews (as an addition to human reviews) is actually one thing that AI does pretty well. It's not good for large architectural issues but it can point out nuanced issues in single files that often wouldn't be caught otherwise

I keep saying this but painting any use of AI at all as the same as vibe coding just harms the real complaints against it, ESPECIALLY in this case where it's subtracting from another real issue (the age gating)

  • source
  • hideshow 6 child comments
  • [โ€“] 22 points 7 months ago (1 child)

    I agree AI in the hands of someone competent is just a speedup. Stupid stuff like making serializer for endpoints is tedious work most of the time

  • source
  • parent
  • hideshow 1 child comment
  • [โ€“] 7 points 7 months ago

    Exactly, yeah. I actually really like AI for line completion and the occasional use of it for debugging. It really just enhances existing IDE features.

    The problems with it come with large unreviewed chunks of code generated by LLMs being thrown carelessly into a codebase. A developer asking for an AI code review or a developer letting AI complete a line they were likely writing anyway to save time is so far removed from the problem that it just screams mindless anti-hype or overly-confident inexperience by juniors

  • source
  • parent
  • [โ€“] 5 points 7 months ago

    I'm personally pretty skeptical of the value of AI code reviews - there's tooling for it where I work, where they will auto-submit a review on any PR I open. I've never actually received a comment of any value from it, while my peers can find things that need resolving without any issue.

    A poor supplement, and absolutely not a substitute.

    And this is coming from someone not entirely opposed to all kinds of AI tooling

  • source
  • parent
  • [โ€“] 2 points 7 months ago

    It's not good for large architectural issues but it can point out nuanced issues in single files that often wouldn't be caught otherwise

    Yeah I agree. It's sometimes good at code smells, though sometimes it can be straight-up wrong in ways that are actually surprising, so it always requires a human in the loop. It's not good at larger-scale architectural decisions, and I'd also add that it's usually not capable of understanding the intent behind business logic.

  • source
  • parent
  • load more comments (1 reply)