Elektrine lite

← Feed

@hugo@social.treehouse.systems

Post #2114769

2026-02-27 20:32 UTC

It is staggering to me that I'm seeing people just now realizing that "wait, if the coding agents produce more code, that means we need to spend more of our time in review, either as the person running the coding agent, or externalizing that cost onto others, or otherwise it means we're just trusting code without anyone reviewing or understanding it!" How the hell are some people only seeing that now? That is literally one of the most basic premises in place here, and immediately obvious on spending even a few seconds of thought on this.

Replies (6)

  • @swelljoe@mas.to 2026-02-27 21:02

    @hugo@social.treehouse.systems most vibe coders are not reviewing the code. If they're at all concerned about quality they're having other models review it. I suspect this puts upper bounds on what they can build, even ignoring quality. But, a theme that comes up a lot when people talk about it is "trust" and how much code one can write with LLMs is determined by how much trust you can put in them. I don't trust them all that much, but I can still offload a lot of boring stuff.

    Open ##3110176

  • @lunareclipse@snug.moe 2026-02-27 21:20

    @hugo@social.treehouse.systems b-but line of lines of code go up?

    Open ##3110181

  • @sabik@rants.au 2026-02-27 22:52

    @hugo@social.treehouse.systems Also each stage (writing code, review) that's done by agents means one less person who understands the code, which is often already not a big number to begin with, so ops will be interesting

    Open ##3110184

  • @M0CUV@mastodon.radio 2026-03-01 09:36

    @hugo@social.treehouse.systems @mos_8502@studio8502.ca Agreed - a perspective I heard recently as to why it takes longer to review LLM-generated code: with human-generated code, you’re starting from a position of trust - you know your co-worker and how they write, what they’ll pay attention to (and maybe miss)… with an LLM, you have to assume a complete lack of trust.

    Open ##3110185

  • @dl2jml@mastodon.radio 2026-03-01 10:54

    @hugo@social.treehouse.systems Once you think about it, poor code is not a bug, it is a feature. One can sell more fixes that way.

    Open ##3110187

  • @satsuma@dice.camp 2026-03-01 12:23

    @hugo@social.treehouse.systems it’s the same effect as trying to speed up a project by employing a load of junior programmers who have no experience of the real world. Sure, they’ll churn out a lot of code (and work overtime for free) but you’ll spend more time fixing their mistakes than doing it properly, and you won’t have the benefit of a better trained junior at the end of it.

    Open ##3110194