Assessing the "Rust-ification" of Bun: Reality vs. Hype

By Tom Lockwood | July 27, 2026

When a company makes bold claims that directly inflate its market valuation, it is imperative to remain canny.

The narrative surrounding the Bun rewrite is being framed as a "smoking gun"—proof that Anthropic’s AI can replace open-source maintainers with superior speed, provided you have the capital. On July 8, 2026, Jarred Sumner announced the initiative to "Rewrite Bun in Rust."

Given my previous reading on Cursor’s FastRender browser and Anthropic’s own C compiler, I viewed this announcement with skepticism. There is an immense amount of venture capital and marketing momentum tied to the perceived coding prowess of LLMs. It is a short leap to conclude that Anthropic's acquisition of Bun naturally led to the decision to rewrite it using Anthropic's own tools.

The Financial Claims

Jarred asserts that the rewrite was completed merged into the main branch over an 11-day sprint (May 3rd to May 14th, 2026).

The reported cost: $\text{Total API Cost} = \$165,000$ $\text{Daily Burn} = \frac{\$165,000}{11 \text{ days}} \approx \$15,000/\text{day}$

For the average open-source maintainer, a \15,000$ daily spend is an impossible luxury. Furthermore, this figure conveniently ignores the massive overhead of the Buildkite CI/CD cluster, which has been running incessantly since the rewrite was supposedly "finished."

Investigating the Evidence

On July 9, a colleague breathlessly claimed that the "era of AI" had officially arrived, citing the Bun blog post as the herald. I decided to dig into the actual repository.

# Cloning the Bun repository to verify claims
$ git clone https://github.com/oven-sh/bun
Receiving objects: 100% (1200304/1200304), 1.23 GiB | 11.97 MiB/s, done.

The Release Gap

As of July 27, 2026—six weeks after the merge—there is still no official release tag. We are now 11 weeks past the last stable release.

The PR Bottleneck

The volume of open Pull Requests from robobun (the proxy for Claude Code) tells a different story:

• July 9: 1,277 open PRs
• July 27: 2,475 open PRs

If we assume a merge pipeline (including Buildkite checks) takes between 40 and 90 minutes, the math for clearing the backlog looks like this:

The "Off-the-Books" Labor

Upon reviewing the PRs, I noticed several were heavily reviewed or authored by Anthropic employees. This suggests the \165\text{k}$ token cost is a fraction of the actual investment.

Observations on the workflow:

• Massive spike in Claude API usage at the start.
• Increasing involvement of robobun and Anthropic staff.
• Human developers maintaining a healthy work-life balance (weekends off).

Figure 1: Conceptual trend of human intervention increasing as the "AI rewrite" progresses.

If the project continues to burn roughly \10\text{k}$a day in resources and labor, the total cost is likely approaching **$$800,000$**.

The Hype Cycle: 2015 vs. 2026

I have a deep respect for ML and AI; some of my proudest professional achievements are in this field. However, the current hype is unprecedented.

• In 2015: I had to beg management to believe that a "black box of linear algebra" could outperform humans at quantitative tasks.
• In 2026: Management is so convinced of AI's omnipotence that they want to apply it to everything, regardless of efficacy.

Many corporate valuations are now predicated on the idea of "AI eating the world."

"In my tiniest mouse voice ever, I want to ask the most relevant business question: Was it actually worth the money, and are these companies truly worth their valuations?"

It is telling that other "AI-driven" projects, such as Anthropic’s C compiler and Cursor’s FastRender browser, have seen zero commits for months. We must remain skeptical of the claim that the Bun rewrite is "done."