Sabitlenmiş Tweet
MBV
2.2K posts

MBV
@mbvlabs
fullstack dev contractor building https://t.co/w6412rsEJY to make web dev in Go faster and easier stream on twitch: https://t.co/T9w8F6xJnC
Katılım Kasım 2019
156 Takip Edilen240 Takipçiler

@dreamsofcode_io Changed my approach to LLMs and agents a lot the last few weeks
Basically write all code that goes into master by hand and then lean on them for discussing/rubber ducking/bouncing ideas etc
Feel like they have been kneecapping the models aggressively
GIF
English

Last night was probably one of the worst experiences I’ve had with LLMs / Agents generating code (GPT-5.5 btw)
Every commit produced regressions, the agent was constantly taking shortcuts, and there was enough hallucinations to make one develop a new sense of reality.
I came incredibly close to rage quitting using Agents all together.
It’s times like those I wonder if we’re all being stupid letting these things loose just for some multiples of perceived productivity.
English

do you also see this workflow resulting in a better final product/outcome faster or just overall better?
that's my main issue with LLMs and coding right now having trying many different approaches. i always end up with the feeling that it would overall have been better to just write it by hand
English

finally found the right metaphor for this shift in how i use opencode.
i used to treat it like 3D printing, where you build the thing layer by layer and commit to each piece as you go
now it feels more like progressive rendering, you start with a blurry version of the whole thing, then keep making full passes over it, and each pass sharpens the entire shape
doing this with gpt 5.5 and voice prompting is the first time things feel like they're clicking

English

coding without LLMs has been a great switch
but, searching for information the old way have not. the ability to get a baseline idea of _anything_ in minutes with the option to question it is so valuable
will likely get a openai sub soon again purely for search, bouncing ideas, quick POCs etc
English

The switch from sqlc to bun (not the JS runtime) in andurel seems to have been beneficial from an ergonomics standpoint
Locality of behavior is a real thing. Being able to quickly add a new model without writing the sql+compiling+converting is a real benefit
Now I mainly focus on good patterns for this setup
Still not 100% sold but promising
English

Been super busy re-writing the crate-operator api that gets installed on deploy crate servers
It's getting to a much nicer place and I've a much deeper understanding of it
Goes to show that using too much AI in the name of urgency likely ends up hurting you long-term
Getting closer to opening up Deploy Crate for public sign-up
English







