Post

To understand why I’m so down on agentic coding I should explain what I think LLMs are good at LLMs are good if you want *consistency* because they are statistical models with extra steps. If you ask them what comes next they’ll bias to what came before … but consistency is *BAD* for programming

17/10/26, 9:37 pm2 reposts27 likes

gabby

@fullmoon.id

I elaborate on this more in this blog post where I explain why consistency and predictability are signs of bad engineering, but the basic idea is: - repeating yourself is bad - boilerplate is bad haskellforall.com/2024/07/soft...

haskellforall.comSoftware engineers are not (and should not be) techniciansWhy predictable development work is a red flag

vqr-1tx.bsky.social

@vqr-1tx.bsky.social

I really enjoy reading your thoughts on this stuff. I also kinda think people using llms for boilerplate maybe actually just want snippets?

BenchSmoke

@benchsmoke.bsky.social

We ran the same to-do prompt three times on a local 9B model and got three different projects: a working CLI, one that stopped after package.json, and an API with no UI. Do you find the inconsistency shows up more in the code itself, or in whether it actually runs?