Update: Bun in Rust
The race to integrate Artificial Intelligence into everything we do has generated some impressive narratives. One of the most discussed recently was the supposed rewrite of Bun – a fast JavaScript runtime and toolkit – into Rust, with the help of Anthropic's AI tools. The promise was unprecedented speed and efficiency, a true game-changer. But, as with everything involving big promises and inflated market valuations, it's crucial to maintain a healthy dose of skepticism.
The Dazzling Promise and Hidden Costs
On July 8, 2026, Jarred Summner, creator of Bun, announced the "Bun Rewrite in Rust." At the time, the news was presented as proof of the power of AI, specifically Anthropic's, to accelerate the work of open-source maintainers. Jarred stated that the rewrite was completed in just 11 days (between May 3 and 14, 2026), at a cost of US$165,000 in Anthropic API calls, and then integrated into the main branch. This amounts to US$15,000 per day, a value far beyond the resources of many open source projects.
It's hard not to draw a parallel between Anthropic's acquisition of Bun and the choice to rewrite it using Anthropic's own tool. The narrative that AI can do developers' work, faster and cheaper, is seductive and supports gigantic market valuations. However, this initial US$165,000 bill seemed, from the outset, "cautiously" incomplete. It did not include, for example, the operational costs of the organization's CI/CD (Continuous Integration/Continuous Delivery) cluster, which has apparently been running non-stop since the "completion" of the rewrite. And this is where the "red flags" begin to appear.
Diving into the Code: What the Data Reveals
I decided to investigate the claims thoroughly. On July 27, 2026, six weeks after the supposed merge of the rewrite to main, there was still no new release tag (version marker). Bun's last release tag was on May 12, 2026, meaning 11 weeks have passed without a new version – an unusually long period for the project.
Deeper analysis revealed an even more complex scenario. On July 9, the number of Pull Requests (PRs) opened by robobun (a proxy for PRs generated by Claude, Anthropic's AI) was 1277. By July 27, this number had skyrocketed to 2475 open PRs. To contextualize, a PR is a request to merge code changes, and each one needs to go through reviews and CI/CD checks. The time to merge a PR into main with Buildkite checks (the CI/CD system) takes about 40 minutes, potentially up to an hour and a half. If we wanted to merge all PRs opened by Claude at this rate, it would require 86 days of continuous pipeline execution.
It's evident that the cost of the rewrite goes far beyond the US$165,000 in API calls. Data analysis shows a peak in Claude's usage at the beginning of the rewrite, but now there's increasing involvement from Anthropic employees and robobun in the Rust code. If we assume a daily cost of US$10,000 since then, the rewrite would already be approaching US$800,000. This is without counting infrastructure costs and the time of the engineers involved. The machine keeps running, Anthropic employees are directly involved, and the rewrite is far from "completed" at the initially stated cost.
The AI Hype and the Lesson for Builders
The Bun situation is a microcosm of the current AI hype cycle. If in 2015 I needed to convince management that machine learning models outperformed humans in quantitative tasks, today the situation has reversed: belief in AI is so strong that it's seen as the solution to everything. And many company valuations depend on the narrative that AI will "eat the world."
However, the reality is more nuanced. Other ambitious AI-driven projects, such as Anthropic's C compiler and Cursor's FastRender web browser, haven't seen commits for months, indicating that initial enthusiasm may not have translated into long-term sustainability.
For us, builders and technology decision-makers, the lesson is clear: we need to be "canny" – astute and cautious – when evaluating claims, especially those that underpin large market valuations. AI is a powerful tool, but it's not a silver bullet that eliminates complexities or costs overnight. The ability to discern between marketing and technical reality, to analyze data, and to question narratives, is what allows us to build robust and sustainable solutions. The question that truly matters for any "business thing" is: "Was it worth the money? And are the companies worth the valuation they hold?" The answer, often, is more complex than the hype suggests.
Sources
📬 Enjoyed this? Subscribe to the Vaccari's Code newsletter for the next wave of software & AI trends, straight to your inbox: Subscribe here