Linux creator Linus Torvalds compares AI code assistants to compilers, dismissing claims that AI autonomously writes software

@twtayaan This is very true - AI is amazing at writing 100% of my code but for deeper work we still need great engineers

@MoreauGabarain @twtayaan That’s my work reduce energy demand and emissions

@JudyandHuahua @twtayaan Yes, the free versions can ‘forget’ and get side-tracked mid-chat. The pay-to-use versions are breathtakingly powerful. I’ve used paid ChatGPT … and paid Claude wins hands down.

AI is already breathtakingly powerful as a productivity aid and will only get better.

@Yairtech If AI is writing 100% of the code, then who's reviewing it?

@twtayaan Linus right as usual. Same cycle, new abstraction layer. LLM bug spam is the underappreciated signal.

@twtayaan People don't see the future. The future is someone will make an AI dedicated to bug finding and it'll just run al day finding bugs. That'll be a whole industry.

@IoannisKap1776 I think that's the key distinction. Not knowing everything is normal not understanding what your code is doing is where problems start.

@twtayaan "But nobody says "the compiler wrote my code." So why are we saying AI wrote it?"

AI generates code, so its different this time

Most people who are busy hyping AI barely understand how ai actually work.I have been exposed to machine learning 8 years ago and AI is nothing different from that.We have been using ai tools but they hype here is now overrated if AI is going to replace the Engineers as most are http://saying.How is AI going to be trained? Linus is right here

@twtayaan @AoifeMFoley Good analogy. More ai, more code, more bugs. More work to test and fix. More activity, more problems. Like with compilers.

@twtayaan @karomancer based and true. Tell em king

@twtayaan Understanding AI tools is key, and it's unrealistic to expect broad programming knowledge from one person.

@twtayaan I think he is right about the basic pillars, namely being able to understand or follow what the AI tool is doing. Because I consider it perhaps impossible for an ordinary programmer to possess such a broad scale of programming knowledge....🤔

@JohnDParody Maybe someday. But if you don't understand the system yourself, how would you know when the AI is wrong?

I’ve said this many times. Technology grifters are always overstating their capabilities just to inflate the amount of money they can raise and so they can sell stock at inflated prices. Bubbles always occur, they pop, then things get real. Many go under, but the spinsters still have their new jets!

@log_npierce Generating code is easy. Maintaining it for years is the hard part.

@twtayaan Compilers are deterministic. They aren't writing anything; they are translating. If you ask an AI to write an app that does X, it needs to decide from a million possible ways of doing X. That's writing code.

@twtayaan nobody sells compilers. soon if we push for local LLMs, nobody would sell AI models either.

@twtayaan in my enterprise work the other engineers and I do a pre-review - in my own repos i am getting a review of composer, grok 4.3 and claude

@codematix Well said. The more powerful the tool becomes, the more important it is to understand the system behind it. Otherwise you're just automating mistakes faster.