@neil_xbt - Karpathy's Software 3.0: English
@neil_xbt on X Neil (@neil_xbt) argues that Andrej Karpathy has framed a new era of software, where natural language is the programming interface, LLMs act as the computer, and prompts serve as code. The post is a hype-heavy thread built on Karpathy's Software 1.0/2.0/3.0 framing, with little detail beyond that framing itself.
- Software 1.0 was hand-written explicit code, Software 2.0 was trained neural network weights, and Software 3.0 is English prompts run on LLMs.
- The post claims the history of software was a workaround for computers not understanding humans, and that workaround is now optional.
- The author promises a 40-minute explanation but the post itself offers only the three-stage framing and no evidence or examples.
Karpathy's framing, as this post presents it, is a three-stage history of software. Software 1.0 is explicit code that humans write line by line. Software 2.0 is neural network weights learned from data rather than hand-written logic. Software 3.0, the new stage, is English: natural language serves as the programming interface, large language models act as the computer, and prompts take the place of code. The author's core claim is that all of software development existed as a workaround for the fact that machines could not understand people, and that this workaround is now optional.
That claim is the substance, and it's worth separating what's sourced from what's rhetoric. The post gives no quotes, dates, or links to Karpathy's original remarks, so the "framework" is secondhand and the word "declared" is doing a lot of work. The three-stage taxonomy itself is a real idea Karpathy has discussed publicly, and it's a useful lens: it explains why prompt engineering, agents, and LLM-based apps feel like a different kind of programming. But the post asserts that the workaround "just became optional" without showing any evidence, such as what fraction of production systems now run on prompts, or where prompts fail compared with code. Reliability, cost, testability, and determinism are the obvious objections, and the post doesn't address them.
If you want to read it, skim it as a clear statement of a popular framing, not as a settled prediction. The "40 minutes" promise is a hook, not a description of the content. The underlying question worth carrying forward is which parts of software actually get easier when the interface is natural language, and which parts stay stubbornly hard.