10 Years of Engineering. 11 Days of AI. Everything Changed.
A decade of engineering built a thriving ecosystem. Eleven days of AI rewrote the assumptions behind it.The Fastest Rewrite in Modern Software History.I almost didn’t write about this one, because by the time I sat down to actually process it, my group chat with other developers had already spent…
A decade of engineering built a thriving ecosystem. Eleven days of AI rewrote the assumptions behind it.The Fastest Rewrite in Modern Software History.I almost didn’t write about this one, because by the time I sat down to actually process it, my group chat with other developers had already spent three days arguing about it and I wasn’t sure there was anything left to add.Someone in that chat kept insisting it was purely a technical story: memory bugs, compile-time safety, the usual stuff. Someone else kept insisting it was purely political: a big company flexing on a small foundation that couldn’t fight back.I went back and forth between both positions for longer than I’d like to admit before I actually sat down and read the primary sources instead of the hot takes about them.Then I reread the timeline one more time, slowly, and realized the part everyone was skipping past was the quietest one: a $60,000-a-year donation that simply stopped showing up, with no email, no explanation, and no goodbye.That detail is the whole story, honestly. Everything else the rewrite, the AI agents, the public feud is the noise around one relationship quietly ending.Here’s what actually happened, as best as the public record shows it, and why I think it says more about where open source is heading than any AI benchmark I’ve read this year.The Language Nobody Trusted With Production CodeZig has spent most of its life being described the same way: promising, elegant, not ready for anything that matters. It’s a small, low-level systems language, positioned as a more modern alternative to C, with manual memory management and a compiler that prizes simplicity over abstraction. Serious engineers respected it. Very few were willing to bet a real product on it.Then came Bun.Jarred Sumner started Bun in April 2021 as a faster alternative to the JavaScript tooling most developers were stuck with a single toolkit combining a runtime, a bundler, a package manager, and a test runner, built on Zig instead of the usual suspects.He built the first version essentially alone, in a small Oakland apartment, before today’s AI coding tools existed. Bun leaned on Apple’s JavaScriptCore engine for speed and paired it with Zig’s low-level control, and it worked well enough to end up running on millions of machines.For years, Bun was the answer whenever someone doubted whether Zig could survive contact with a real, widely used product.I’ll admit I was one of the skeptics for a while. I remember installing Bun on a side project sometime in 2023, mostly out of curiosity, expecting it to choke on something within a week. It didn’t. It just worked, quietly, faster than the Node setup it replaced, and I genuinely forgot I was using anything unusual until a coworker asked what runtime we were on.I remember feeling a little smug about it at the time, like I’d found something before it was cool the kind of feeling that ages badly whenever the thing you found early turns into a headline a few years later for reasons you didn’t see coming.Bun proved what Zig’s critics never believed.That’s roughly the arc a lot of the Bun community went through, from what I can tell reading the reactions over the past few weeks. Genuine pride in something that shouldn’t have worked as well as it did, followed by a slow unease once the story behind it started coming apart in public.The Acquisition That Changed the IncentivesIn December 2025, Anthropic acquired Bun. On paper, this looked like a straightforward win for everyone: a fast, modern JavaScript runtime backed by a company with serious engineering resources, and continued support for the Zig ecosystem that had made it possible.Behind the scenes, according to accounts from both sides that surfaced publicly in the months after, the relationship between Bun and the Zig Software Foundation had already been fraying for a while. Sumner had set up a $60,000 per year donation to the foundation and credited Zig publicly for Bun’s performance, which by most accounts was a genuinely generous, unprompted gesture.But the Zig team, led by creator Andrew Kelley, had reportedly grown increasingly uneasy watching how Bun’s codebase evolved, describing what they saw internally as accumulating technical debt and rushed engineering decisions, long before any AI tooling entered the picture.So when the acquisition happened, the reaction inside the Zig Software Foundation wasn’t dread. It was reportedly closer to relief. And then the donation that had been arriving every year simply stopped, with no note attached. Scheduled meetings between the two teams went unattended, with neither side bothering to cancel them.Nobody officially ended anything. It just went quiet.Eleven Days, One Million Lines, One Language GoneHere’s the part that actually broke through into mainstream tech coverage: in mid-2026, Sumner’s team rewrote Bun’s entire core from Zig into Rust not over months, but in eleven days.The original Bun codebase ran to roughly 535,000 lines of Zig. Rather than migrating it by hand, Sumner orchestrated more than 50 parallel Claude Code workflows, generating over a million lines of Rust in the process.According to his own published account of the migration, the process followed a detailed, several hundred-rule internal guide: first a literal, line-by-line translation from Zig into Rust, followed by later passes where the AI agents refined the raw translation into more idiomatic, human-readable Rust.The results, by the numbers Sumner’s team reported: the rewritten version passed nearly the entire existing test suite, the compiled binary shrank by roughly 20%, HTTP throughput improved by a few percentage points, and well over a hundred bugs that had existed in the original Zig implementation were fixed along the way. The last Zig-based release of Bun shipped as version 1.3.14. Every release since has been Rust.Laid out plainly, the scale of what happened looks like this:I want to sit with that number for a second, because I don’t think it’s fully landed for most people outside the systems-programming world. A migration of that size, in a codebase with genuine memory-safety complexity, used to be the kind of project a team would budget six months to a year for, with a real risk of quietly introducing new bugs the whole way through. Eleven days is not a faster version of that process. It’s a different category of process entirely.The Fallout: Two Very Different Stories About WhyOne decision. Two truths. One irreversible split.This is where it stopped being a technical story and became something closer to a public breakup, and I mean that almost literally: both sides published their version of events within days of each other.Andrew Kelley’s account, published on his own blog, didn’t hold back. His argument wasn’t really about Rust versus Zig, or even about AI-generated code specifically. It was about what he described as a long-building mismatch in engineering culture.Bun’s code had been accumulating “hacks on top of hacks,” in his words, well before Claude Code ever touched it, and the rewrite was less a verdict on Zig’s limitations than a reflection of how Bun’s team had already been operating. He was candid enough to walk back one specific claim about the project’s fuzzing setup within days of his own post, while keeping the broader criticism intact, which, to his credit, is a more honest correction than most public disputes ever get.The counter-argument, made by other commentators following the story closely, was more skeptical of the official rationale. The stated reason for the rewrite was Zig’s manual memory management colliding badly with JavaScriptCore’s garbage-collected engine; every allocation across that boundary was, in Sumner’s own description, a small bet the team kept losing a few times a week.Rust’s compiler turns that entire category of bug into a compile-time error instead of a runtime crash, which is a real, defensible engineering reason to migrate. But critics pointed out that other viable fixes to that same memory problem wouldn’t have required a full rewrite and that choosing the most dramatic option, timed right as Anthropic needed a flagship demonstration of what its own AI coding tools could do, was not exactly a coincidence.Both of those things can be true at once. That’s the part I keep coming back to. A real technical problem existed. A real business incentive to showcase AI-driven rewrites also existed. Nobody has to be lying for both motives to have shaped the same decision.Separately, and almost as a footnote that I think deserves more attention than it got, Kelley formalized a total ban on AI-assisted contributions across every Zig repository shortly before this all became public, calling AI-generated contributions a drain on reviewer time that produces negative value, not merely zero value. Because Bun’s own workflow now depends heavily on AI-assisted development, and can no longer contribute changes upstream under that policy, Bun forked the compiler it depends on.Two projects that were, for years, held up as proof that a small open source language could support serious production software are now on opposite sides of a hard line about what AI is even allowed to touch.What This Actually Means, Beyond the DramaI think it’s tempting to read this story as a one-off, a single ugly falling-out between one founder and one foundation. I don’t think that’s the useful takeaway.The uncomfortable pattern underneath this is that AI-assisted rewrites have quietly turned “switching languages” from a multi-year strategic bet into something closer to a routine infrastructure decision.Historically, the cost of a language migration was the single biggest reason smaller, community-funded projects like the Zig Software Foundation had real leverage over the companies that depended on them; nobody was going to casually walk away from years of accumulated investment in a language. That leverage assumed migration was expensive. It increasingly isn’t.That has a direct, unglamorous consequence for anyone who maintains open-source infrastructure that larger companies quietly depend on: the thing that used to guarantee you a seat at the table being expensive to replace is eroding, project by project, rewrite by rewrite.The Zig Software Foundation still reported a healthy budget in the high hundreds of thousands of dollars this year, and it isn’t in immediate danger. But the $60,000 that vanished without a conversation is a preview of a bargaining position a lot of maintainers are going to have to rethink, and probably sooner than they’d like.I keep thinking about a smaller open-source project I contribute to occasionally, nothing close to Zig’s scale, maintained by two people in their spare time who’ve talked openly about how precarious their funding already is.Neither of them has said anything about this story directly, but I’d be surprised if it hasn’t crossed their minds the same way it’s crossed mine: if a foundation with $670,000 a year and a widely respected creator at the helm can lose a six-figure relationship to silence, what happens to projects with a fraction of that leverage and none of that visibility?Where I’ve Landed on ThisI don’t think there’s a clean villain in this story, and I’m genuinely suspicious of anyone who tells it like there is. A founder built something real, gave credit and money to the project that made it possible, and later made a defensible technical call using tools that didn’t exist when he started.A foundation watched a dependent project’s code quality decline for years, made a principled call about what kind of contributions it would accept, and got blindsided by silence instead of a conversation it clearly felt it deserved. Both of those are recognizably human failures, not villainous ones the kind that happen when incentives shift faster than relationships can adjust to them.What actually unsettles me isn’t the rewrite itself. It’s how ordinary this is going to become. Eleven days for over a million lines of production code is not a ceiling. It’s closer to a floor, and it’s going to keep dropping. The real question this story leaves me with isn’t “was the rewrite good engineering.”It’s whether the informal trust that’s held open source together for two decades credit, small donations, a standing monthly call can survive a world where switching away from any dependency, including the people who built it, got this cheap this fast.I don’t have a tidy answer to that. I don’t think anyone does yet, including the people at the center of this story. But I’d rather sit with the discomfort of that than pretend the eleven days were the interesting part.The real story was never Zig versus Rust.It was about how AI is changing the cost of software, and the relationships behind it.That’s a shift every developer and open-source maintainer should be watching.What’s your read on this engineering decision, business decision, or both wearing the other one’s clothes? I’d like to hear where you land in the comments.If this was useful, a clap helps more people find it, and if you want a deeper technical breakdown of how the actual Zig-to-Rust migration pipeline worked, let me know, and I’ll put one together.This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!10 Years of Engineering. 11 Days of AI. Everything Changed. was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.Source: Generative AI Pub — Published — Category: Image AI