its a weird phenomenon that’s spread all over twitter like crazy a bit while back. in fact, it’s kinda started at the same time as the big AI boom, when everyone was trying to set up their own openclaw and purchase every single mac mini on the market. it’s… scary, sometimes, to see how good it can be in comparison to their human partners.

though, this is still a niche. twitter, after all, has an algo which shows u related stuff so no wonder all i was seeing everywhere was AI bro after AI bro back then. its just like the entire crypto fiasco all over again - hype for not much, really. but… i did state a little while back (in fact, in /blogs/hello-world) that AI models do have a proper use in programming as well

wait, you’re defending AI??

yeah, kinda. it’s been in use for enough time already that there’s no longer any fear of it stealing any programmers job. no matter what prompt or how much more data it steals, it’ll never be able to give non-programmers the tools to build their own apps and programs - at least not in an ideal way. in order to utilize LLMs to speed up your process (or work on your behalf), you still need the technical knowledge behind it (e.g. “write an app that does x” is not as detailed as “create a web app using react 19 + redux + tailwind that does x”). not only that, but no matter the LLM, all of them seem to lose context or have the tendency to rewrite code or styles throughout the codebase, making it near impossible to maintain.

though, if you pair it with yourself - thats when it really hits. it speeds up the process of prototyping so much. in fact, as i’m writing this post, i have my little deepseek agent working in the background on a little react-native project that i have been postponing for months. the fact i can have a working prototype so easily is impressive. but, even then, the code it generates is riddled with deprecated functions (e.g. runOnJS instead of scheduleOnRN), misused react effects, etc. so my intervention is still needed at the end of the day. and this applies to ALL models, not just deepseek (imo, opus 4.7 does even worse than ds v4 pro ironically)

using LLMs is also really really useful to understand complex codebases (e.g. pear-desktop) fast. it’s also a good way to find vulnerabilities / performance or security issues that you may omit on your own during development (obviously, a serious QA team would find them, but im talking from the perspective of an oss / personal dev). (though, claude mythos does sound really intriguing when it comes to vulns. maybe ds v4.1 or v5 manages to learn some of that skill too lol)

is this blog AI too??

never ever will AI be able to write such bullshit like i do. this blog is personal, its my human work and only mine. i even disabled my autocompletions just in case, though i wouldnt accept them either way. i love writing my bullshit, because, its my bullshit!

so youre gonna slop ur way through life

yes and no. while i have integrated deepseek way more deeper into my life nowadays (i dont care about my data, privacy folks - please - the idf already have it, id be glad if the ccp have it as well), i still have 1 or 2 projects where i pop open zed with AI off and write whatever my minds desire (e.g. the shithole that is doggoscript). and of course, i still r/w code myself - a machine cant do it perfectly after all, not even with 14314 mcp servers attached.

and, unlike the many many twitter vibecoders there are, i do not abuse my models to shove useless PRs and issues into random repos all over github. my little buddy is strictly interacting with what im allowing it to, so the slop it generates should only affect me and my projects. i also do thoroughly check any code before i even THINK about creating a PR just because i feel like im a bother, so slop is definitely not making it out there from me.

conclusion

i’m not scared to admit i find it satisfying and impressive that these complex models have managed to generate such good results nowadays. i can only hope they improve more and more, since - as i’ve stated before - engineers, designers, and programmers are still required to guide them through correctly.

AI is a tool that should be used, as any other tool, for the good of the public. bad actors (and greedy corporations) will exist for as long as humanity, we just have to fight through it all.