Paul Dix: AI wrote 1M lines of code, then spent months refining it
On his blog, Simon Willison quoted a passage from Paul Dix's essay "The end of programming." Dix says AI wrote one million lines of code and then refined that codebase over the following couple of months until it became reliable software now running on millions of developer machines. He calls the result "absolutely mind blowing." Dix anticipates a skeptical objection: that the feat is less impressive because there was an "oracle" to compare the output against, which made porting from one programming language to another comparatively simple. He rejects that framing as selling the achievement short. His broader claim is that if you can build a verification system and give the AI proper direction, it can produce highly complex, highly sophisticated software and keep refining it until it works. The post does not name the specific software, company, or the two languages involved in the port, nor does it quantify the refinement period beyond "a couple of months" or give a figure for how many developer machines run the result.
Key facts
- Paul Dix writes that AI wrote 1M lines of code and then refined it over the following couple of months into reliable software.
- That software is now running on millions of developer machines, per Dix's account.
- Dix anticipates the objection that an existing oracle made the underlying language port simple, and argues this undersells the achievement.
- His stated recipe: build a verification system and give proper direction, and AI can produce and keep refining highly complex, sophisticated software.
- The specific project, the two languages involved, and the nature of the oracle are not named in the source.
Why it matters
The claim moves past one-shot AI code generation toward a sustained, multi-month refinement process that reportedly produced software now deployed at scale. That distinction, an AI system that keeps working on a codebase after the initial draft rather than producing it once, is central to the broader argument over how far AI-driven engineering can go on complex, long-lived software.
Who it affects
Software engineers and engineering leaders weighing how much of a large codebase they can hand to AI, and anyone tracking the debate over Dix's framing that this points toward "the end of programming" as a distinct human skill.
How to use it
Dix frames the underlying recipe as: build a verification system that can check the AI's output, then give the AI proper direction. On that basis, he says, AI can produce highly complex, sophisticated software and continue refining it until it works, rather than stopping after a first draft.
How solid is it
The account comes directly from Paul Dix, quoted verbatim by Simon Willison from Dix's essay "The end of programming." It is a first-person reflection by Dix, not an independently audited case study, and no project name, language pair, or user count is disclosed to allow independent verification.
Risks and caveats
Dix's rebuttal to the oracle objection is asserted rather than demonstrated: he calls the framing an undersell but does not detail why the verification setup made the task harder than the objection suggests. With the specific software, languages, and verification system all unnamed, the claim stands as an anecdote about one project rather than a generalizable result.
“The fact that AI wrote 1M LOC and then refined it over the course of the next couple of months to produce a reliable piece of software that is currently running on millions of developer machines is absolutely mind blowing. And you can say, "well it's not that impressive because they had an oracle to compare against, so it was simple to go from one language to another," but I think that's selling this entire thing short. If you can build a verification system and give proper direction, AI can produce a highly complex, highly sophisticated piece of software and it can continue to refine it until it just works.”
— Paul Dix, "The end of programming," quoted by Simon Willison