When the creator of a programming language publicly calls one of the world’s most valuable AI companies a smoke-blower, the developer world listens. Zig creator Andrew Kelley’s blistering response to Anthropic’s claim that AI rewrote the Bun runtime from Zig to Rust has done exactly that — and the 1,231 upvotes and 625 comments on Hacker News suggest the community is paying attention.
🔍 THE BOTTOM LINE
Anthropic needs the world to believe AI can replace software engineers. They’ve raised $132 billion and are approaching an IPO valued over $1 trillion. But when the creator of the language they’re claiming AI abandoned calls the narrative dishonest, it exposes the gap between marketing spectacle and engineering reality — a gap that matters for anyone making staffing, architecture, or investment decisions based on AI hype.
What Actually Happened
Bun, a TypeScript runtime originally written in Zig (one of the largest Zig codebases in existence), was acquired by Anthropic. Its founder Jarred Sumner then oversaw a massive agentic rewrite from Zig to Rust, claiming near 100% AI contributions. The rewrite was merged to mainline — and two months later, the explanation finally arrived, carried by headlines like The Register’s “Anthropic’s Bun Rust rewrite merged at speed of AI.”
The marketing was clear: AI is so powerful it can rewrite a major infrastructure project. The problem, as Ray Myers detailed, is that the same AI “was not powerful enough to catch a use-after-free.”
Andrew Kelley’s Response
Kelley’s response was blunt to an unusual degree. He characterized the Bun team’s experience as “poor communication, unrealistic expectations, low empathy, no experience. Just a total shit show” — and argued the code quality problems stemmed from engineering decisions, including overusing AI agents to write and review everything, not from Zig’s limitations.
The core dispute is about narrative control. Anthropic’s version: Zig wasn’t up to the task, AI saved the day. Kelley’s version: the Bun code was a mess because of bad engineering decisions, and the Rust rewrite was a marketing opportunity to showcase Anthropic’s Fable model.
Myers offers a third, more cynical read: management eagerly approved the Rust option because it was great marketing, Anthropic already uses Rust, and Zig is openly against using Anthropic’s products. The rewrite was a business decision dressed up as a technical necessity.
Why This Matters Beyond Bun
This isn’t just a programming language flame war. Anthropic is actively campaigning on the narrative that coding is going away. Their CEO Dario Amodei has said so at CFR events. That narrative has real consequences — people make architecture, product, and staffing decisions based on it, often driven by fear of being “left behind.”
The Zig pushback is significant because it comes from someone with no skin in Anthropic’s game. Zig’s code of conduct explicitly bans AI contributions. When the creator of a language that refuses to use AI looks at an AI-claimed rewrite and says the engineering doesn’t support the hype, that’s a credibility challenge from outside the bubble.
As Myers put it: “Anthropic is an unreliable narrator.” They need the Bun rewrite to be an AI triumph. The engineering details — the missing pros-and-cons analysis, the two-month delay in explanation, the padding of improvement lists with unrelated changes — suggest the story was reverse-engineered for marketing, not derived from honest technical assessment.
The Deeper Problem With AI Coding Claims
The Bun situation reveals a structural problem with AI coding benchmarks and claims. When a company that sells AI models also acquires the projects those models are applied to, the incentive to report honest results disappears. Bun claims near 100% AI contributions to the rewrite — but the rewrite was to unsafe Rust (allowing literal file-by-file migration), and the explanation was published months after the decision was made.
This pattern echoes concerns we’ve covered before about AI coding agent security and the gap between AI coding cost claims and reality. When the entity measuring AI capability is also the entity selling it, the measurements deserve skepticism.
NZ Angle
New Zealand’s tech sector, like Australia’s, is being told that AI will replace software engineers. The Albanese government is simultaneously grappling with AI’s impact on the workforce while Kiwi companies face the same pressure to adopt AI coding tools. The Zig pushback is a reminder that the “AI can write all your code” narrative is being driven by companies with trillion-dollar valuations riding on that belief being true — and that engineers on the ground, including those in NZ’s growing Rust community, should apply the same skepticism Kelley did.
❓ FAQ
Is the Bun rewrite actually bad? Not necessarily. The file-by-file port to unsafe Rust is a legitimate migration strategy, and the technical details in Bun’s blog post contain reusable insights. The criticism is about the narrative — claiming AI did something it didn’t, and attributing the decision to technical necessity when business incentives were clearly at play.
Why does Zig ban AI contributions? Zig’s code of conduct prohibits AI-generated contributions, believing that human understanding of every line of code is essential for systems programming where memory safety and performance are critical. This is a philosophical stance, not a technical limitation.
Does this mean AI can’t write code? No. It means one specific claim — that AI rewrote a major infrastructure project “at the speed of AI” — is being challenged by someone with direct knowledge of the codebase and the ecosystem. AI can write code; the question is whether the claims about what it achieved are honest.
Should companies be worried about using AI coding tools? The concern isn’t the tools themselves but the marketing around them. Making architecture or staffing decisions based on hype rather than verified results is the real risk. The Bun case shows how easily a business decision can be repackaged as an AI triumph.
🔍 THE BOTTOM LINE
The Zig creator’s takedown of Anthropic’s Bun rewrite narrative is the most credible challenge yet to the “AI replaces software engineers” story. It comes from outside the AI bubble, names specific engineering failures the marketing glossed over, and highlights a structural conflict: when AI companies own the projects their AI is applied to, honest assessment goes out the window. For anyone making decisions based on AI coding capabilities, the lesson is simple — trust the engineering, not the press release.
