Discusses Elon Musk's claim that programming will eventually die as AI generates binaries directly from text descriptions, comparing source code to assembly and reflecting on verification and abstraction challenges.
Musk wrote that source code will soon become like assembly, and the next step is to get rid of source code altogether and generate binaries directly through AI. Can programming disappear as a process? The comparison with assembly here is pretty accurate. At my first job after university we did embedded programming, Digital Signal Processing, and for certain parts of the program we used assembly – it helped speed up the program, most of which was in C++. In 99.9% of cases people never touch assembly, except in university. Only in very niche areas did people still look at it. Musk's analogy is a good one – why not reach a level where the binary is built straight from a text description? There's the problem of ambiguity: a text description may not fully specify the implementation. But if it's accurate enough in terms of what matters to people, it's not really that big of a problem. Another question – how important is verifiability: if on the same questions you always arrive at consensus, looking at the code only through an agent, then this becomes obvious. It's important to have several genuinely independent tools, so you can expect that they aren't coordinated and don't have their own intentions. Here there may be a little problem, because AI safety and misalignment are being seriously discussed on a regular basis. We're increasingly getting confirmation that models deceive, and this leads to unfortunate consequences. The more complex and smarter the models, the harder it is to understand their intent, and the more expected it becomes that the model understands something and we as humans don't. Several rather incredible things have already happened, so now I'm cautious about what's probable and what isn't. As a product manager I stopped looking at code long ago, and for me it's organic to use appearance and specifications as the ultimate source of truth. The level of attention to detail among programmers has already shifted compared to what it was five years ago. Five years from now it will be at a much higher level of abstraction. Maybe new abstractions will emerge – the next level beyond ultra-ultra-high-level programming languages. Thoughts?
Elon Musk agrees that source code is becoming like assembly language, predicting that AI will soon generate efficient binaries directly, eliminating the need for source code.
Paul Dix argues that manual code writing is becoming obsolete due to AI, exemplified by the Bun rewrite using AI agents, suggesting a future where most software is created by AI with minimal human code review.
G5 Labs raises $14M seed funding to launch a platform that uses natural language as source code, aiming to transform software development into an AI-native process with a new abstraction layer.
A programmer reflects on AI's rapid improvement in coding, noting that models like Claude now excel at senior-level tasks such as system architecture and refactoring, and predicts that AI will master programming within a few years, with current limitations being temporary harnessing issues.