我在 AI 方面并不是原始主义者。我在 ML（机器学习）和 AI 方面的工作是我最自豪的工作之一。然而，这一轮的炒作周期是我见过的最夸张的炒作。早在 2015 年，我就不得不向管理层说明，希望他们理解，一个在定量任务上比人类表现更好的黑箱线性代数模型，仅仅就是更好。现在似乎世界完全反过来了，管理层非常直接地相信 AI 很棒，以至于他们希望将它应用到一切上。而那些估值有时还取决于 AI 能否吞噬一切：
以我最小的鼠标声音，我想问一个关于商业事务最相关的问题：花的钱值吗，公司的估值合理吗？
附言：Anthropic 的 C 编译器：https://github.com/anthropics/claudes-c-compiler 和 Cursor 的 FastRender 网页浏览器：https://github.com/wilsonzlin/fastrender 几个月没有任何提交。
I think it’s important to be very Canny when someone makes a claim that supports a company’s large valuation.
The Bun rewrite seems well positioned as proof-positive that AI and specifically Anthropic’s AI can do the work of open-source maintainers, for some money, but faster.
On the 8th of July 2026 Jarred Summner of Bun fame posted about “Rewriting Bun in Rust”：https://bun.com/blog/bun-in-rust. At the time, I felt pretty Canny, having already read about Anthropic’s C compiler and Cursor’s FastRender web browser. It seems like there is a lot of valuation money riding on the coding capabilities of AI, and a lot of marketing about those capabilities.
It’s not a huge stretch to link the acquisition of Bun by Anthropic with the choice to rewrite Bun with an Anthropic tool. In his article Jarred claims that over 11 days (between 3rd and 14th May 2026) and at a cost of $165,000 for Anthropic API calls the rewrite was done and then merged to main . This represents a cost of $15,000 dollars a day, a cost well outside the means of many open source maintainers. This number doesn’t seem to include the leviathan whirring of the CI/CD in the org’s Buildkite cluster, which appears to have been constantly whirring ever since the “rewrite” was “completed”.
Why do I put those two words in scare quotes?
Well, on the 9th of July a peer of mine in a group chat said something unsurprising given our current environment - breathlessly proclaiming that the era of AI had emerged, and its herald was the “Rewriting Bun in Rust” blog post. So, I decided to look into this. Since the 9th I’ve been having a closer look at the claims and the code, and today I cloned the Bun repo:
As of today, the 27th of July 2026, a six week period after the rewrite was merged to main, there’s still no release tag. It’s now been 11 weeks since the last Bun release tag:
The last time there wasn’t a Bun release in a month was between the 26th of October 2022 and the 7th of December 2022 when there was a gap of six weeks and November was skipped, between v0.2.2 and v0.3.0:
On the 9th of July, the number of open pull requests from robobun：https://github.com/oven-sh/bun/pulls/robobun (a proxy for PRs made by claude code) was 1277. As of this moment, on the 27th of July that number is 2475 open PRs. I don’t claim to know for sure but as far as I can see the process of merging to main with Buildkite checks completed seems to take 40-ish minutes (sometimes it looks like it takes up to an hour and a half). At this rate if we want to merge all the open Claude PRs it’ll only take running the pipeline for 86 days, continuously.
Generously, some of these PRs don’t deal with Rust code. Further investigation was warranted. I initially clicked through some of the PRs and found some extremely：https://github.com/oven-sh/bun/pull/34660 reviewed：https://github.com/oven-sh/bun/pull/28932 ones：https://github.com/oven-sh/bun/pull/34467.
At this point I began to suspect that much of the cost of the rewrite was off the books. While $15k a day in tokens widened my eyes a little (and maybe I’m naive, maybe that’s low), I hadn’t seen the numbers for the CI/CD costs of Buildkite, and as you might have noticed from above：https://github.com/oven-sh/bun/pull/34660, some PRs were written by Anthropic employees. When I did analysis of the data it became clear to me that the project was ticking along with more Claude credits and more Anthropic employee involvement:
One of the assumptions I’ve made here is that Jarred Summner’s commits during the rewrite were using Claude.
It’s pretty clear that there was a big spike in Claude usage when the rewrite began. It seems like, now, Anthropic employee and Robobun involvement in Rust is ramping up:
It’s nice to see that the Anthropic employees involved seemed to have a good work-life balance, taking breaks on the weekend. But, it looks like that period is over. It seems like we’re ramping up to something, and there’s still no release tag.
What’s pretty obvious to me is that we can’t take it at face value that the rewrite is “done” or that it was “done” for $165k USD. The Bun team never made that claim, but I’ve seen and heard those breathless claims that the rewrite is proof of something , and I do still think we need to be Canny about claims that support a large valuation. Anthropic is dogfooding this, the machine is still ticking along, and Anthropic employees are directly involved. If we imagine that the rewrite is still costing $10k a day, we’re approaching $800k in money spent on this rewrite.
I’m not a primitivist when it comes to AI. Work I’ve done on ML and AI has been some of the work I’ve been most proud of. This current hype cycle is, however, the worst I’ve ever seen the hype get. Back in 2015 I had to plead with management, trying to get them to understand that a black-box bunch of linear algebra that was better than humans at a quantitative task was, just, better. Now it seems like the world has completely reversed, and management is so forthright in their belief that AI is good, that they want it rubbed on everything. And those valuations are sometimes contingent on AI eating everything :
In my tiniest mouse voice ever I want to ask probably the most relevant question one can ask about a Business Thing: Was it worth the money and are the companies worth the valuation?
P.S. Anthropic’s C compiler：https://github.com/anthropics/claudes-c-compiler and Cursor’s FastRender web browser：https://github.com/wilsonzlin/fastrender haven’t had any commits for months.
P.P.S. I’m looking for a job：https://www.linkedin.com/in/thomas-lockwood-3048b056/.
扫码直达文章详情