How is the Bun Rewrite in Rust going?
A critical examination reveals the "Bun rewritten in Rust by AI" narrative, touted as a speedy, cost-effective triumph, might be more hype than reality. The author uncovers significant hidden costs, ongoing human involvement, and a lack of public release, casting doubt on the true value proposition of large-scale AI-driven rewrites. This piece resonated on HN by tapping into widespread skepticism about AI's current capabilities for comprehensive software development and maintenance.
The Lowdown
The author investigates the highly publicized claim that Bun, a JavaScript runtime, was entirely rewritten in Rust by AI for a mere $165,000 in Anthropic API calls over 11 days. This investigation aims to challenge the breathless claims often made about AI's capabilities, particularly when they influence company valuations.
- The original Bun blog post from Jarred Summner claimed a rapid, AI-driven rewrite of Bun into Rust.
- Six weeks after the supposed merge to
main, no new public release tag for Bun exists, an unusually long gap for the project. - Analysis of the Bun GitHub repository reveals a massive and growing number of open pull requests from 'robobun' (proxy for Claude-generated code), suggesting significant ongoing work.
- The author observes a consistent, increasing involvement of Anthropic employees in the Rust codebase and substantial CI/CD activity, implying costs far exceeding the initial $165k estimate.
- The true cost, factoring in continued AI usage and Anthropic employee time, is speculated to be approaching $800k, and the project is clearly not 'done.'
- Other much-hyped AI-generated projects, such as Anthropic's C compiler and Cursor's FastRender, are noted to have stalled with no recent commits.
Ultimately, the article serves as a cautionary tale against uncritically accepting AI-centric marketing narratives, emphasizing the often-unseen complexities and true costs of software development, even when assisted by advanced AI.
The Gossip
Hype vs. Hard Realities
Many commenters echoed the author's skepticism, arguing that while AI can generate initial code quickly, it falls short on the less glamorous, but crucial, aspects of software development: bug fixing, optimization, and long-term maintenance. They emphasized that the 'real bill' for an AI-assisted rewrite comes much later, involving significant human effort to refine and sustain the codebase, often debunking the 'triumphalist pronouncements' of AI's immediate payoff.
Defining 'Done' and 'Released'
A central debate revolved around what constitutes a 'release.' Some argued that Bun's Rust version *is* released, as it's been deployed to millions of Claude Code users and reportedly works well without major issues. Conversely, others maintained that without an official public release tag for the open-source project, it cannot be considered truly 'done' or available for the broader community, especially given Bun's prior production usage.
Motive, Money, and Marketing
Commenters largely agreed that the stated $165k cost for the rewrite was a significant underestimate, omitting substantial CI/CD expenses and ongoing human intervention from Anthropic. There was strong speculation that the entire rewrite served primarily as a marketing stunt by Anthropic to showcase its AI capabilities and boost valuations, rather than being a genuinely beneficial and transparent open-source initiative, with some even suggesting Anthropic cares more about its internal use than the public project.
Idiomatic Implications
A sub-discussion emerged regarding the quality of AI-generated Rust code, specifically concerning 'idiomatic Rust.' Some suggested that a human team would produce more idiomatic code, implying AI-generated code might rely on less optimal patterns or 'unsafe' blocks, making it harder to maintain. However, others questioned the precise definition of 'unidiomatic' in this context, highlighting the subjectivity.