Large language models fail at hard and medium difficulty problems where they can’t stitch together well-known templates. You should still use them.

InfoWorld is a source of news, analysis, and commentary on technology trends, IT strategies, and business innovation. With a focus on enterprise technology and digital transformation, InfoWorld offers insights and guidance for IT decision-makers, software developers, and technology professionals. From  articles on cloud computing and cybersecurity to product reviews and industry trends, InfoWorld helps readers navigate the complexities of modern IT environments and make informed decisions to drive business success.

InfoWorld

Large language models excel at generating repetitive, well-understood code like CRUD applications and web development, but struggle significantly with algorithmic problems and novel implementations. Through practical experience porting a patch system from Python to TypeScript, the author demonstrates that LLMs fail at medium to hard difficulty problems where they cannot rely on well-known templates. While LLMs are valuable tools for routine development tasks, they require human oversight and cannot autonomously handle complex algorithmic design or domains with limited training examples.

What you absolutely cannot vibe code right now

I find LLMs most useful generating the tedious, well known grunt work. It gives me back time to write the code I actually want to write. If LLM’s never got better than that, that would be fine with me

Exactly they can do basic stuff and help with more complex code, but they can’t do it by themselves 💯
And I’m not sure we need or really want them too. Code is too important to be left up to a random program.

Vibe coding, I guess lasted six months?

This article really resonates with how I feel. With the advent of LLM, it’s easy to fall into the illusion that you can create anything just by throwing in a prompt, but in reality, it’s more like hiring an assistant who is either more skilled than you or not very useful. I find vibe coding itself very enjoyable, but I want to continue enjoying this work until they grow to the point where they can completely take over all tasks independently.

As it stands now, vibe coding is simply not feasible to build real-world applications by non-technies. I’m an experienced developer and I embrace LLM’s, AI and all that. It speeds me up by 10x. But it gets annoying at times as the LLM’s make so many mistakes, be it in security, maintainability or simply not following instructions (context problem, even with proper Cursor rules setup).