Last week Bun 1.4 shipped. It’s the release with the big overhaul: the whole runtime was ported from Zig to Rust, more than a million new lines of code. The community mostly argued about the language switch. I find another part far more interesting – how this rewrite actually happened.
One developer drove it: Jarred Sumner. His tools were a pre-release version of Fable 5 and a near-unlimited token budget. He didn’t hand-translate line by line. He built a harness where several agents translated the Zig codebase into Rust in parallel. Over eleven days that produced 6,778 commits. At API list prices, the run would have cost roughly $165,000. The merge that revealed it all showed over a million inserted lines.
And the code works. After that first pass, agents kept testing and fixing for months before Bun 1.4 became an official release. Paul Dix takes exactly this pattern as the basis for a bold claim. Dix builds databases – he’s founder and CTO of InfluxData. In his essay ‘The end of programming’ he writes that coding as we know it – writing code by hand, having colleagues review it line by line – is headed for extinction.
His point isn’t that humans become useless. It’s that the sheer volume of working software produced by agents drowns out the hand-written part. We’ll have more software than ever. Most of its code just won’t be written or even read by people. As evidence, Dix points to a chart from GitHub’s report on the August 17 outage: the amount of code being pushed has climbed steeply since last year.
Dix doesn’t stop at other people’s examples. In an InfluxDB fork he built two sizeable features this way himself – an Iceberg integration and a system for replicating data between edge nodes. He supplied the architecture and requirements; Fable did the rest through sub-agents while he only supervised. 14 hours for one, 28 for the other. Then he had it run on real infrastructure and fix its own bugs.
What sticks with me: the line between ‘prototype’ and ‘real software’ is blurring. The prototype is the working software, and it doesn’t get better through code review but through use and testing. Dix expects the rest of us to get this way of working in a year or two – with affordable tokens instead of six-figure bills.
As always, I’m neither a cheerleader nor a doomer here. But the Bun overhaul is hard proof of what Fable-level intelligence plus enough tokens can already do. And honestly, that’s a different league from ‘AI helps with the prototype’.
Sources: Paul Dix: The end of programming · Simon Willison · Bun in Rust · Hacker News