Analyst questions Bun’s Rust rewrite and Anthropic AI claims
A scrutiny of the Bun runtime’s transition to Rust suggests the project remains incomplete and costs may far exceed the reported $165,000, raising broader questions about corporate valuation hype surrounding artificial intelligence.
A blog post published on 27 July 2026 by an anonymous author on lockwood.dev has cast doubt on the narrative surrounding the Bun runtime’s rewrite in Rust. The analysis challenges claims made by Bun creator Jarred Sumner on 8 July 2026, which stated the rewrite was completed in 11 days between 3 and 14 May 2026 at a cost of $165,000 for Anthropic API calls. The author argues the project remains incomplete, noting that six weeks after merging to main, no release tag existed.
The post highlights that open pull requests generated by "Robobun"—a bot linked to Anthropic’s Claude—had risen from 1,277 on 9 July to 2,475 by 27 July. The author estimates ongoing costs could reach $10,000 daily, potentially totalling $800,000, and suggests Anthropic employees are directly contributing code, raising concerns about corporate valuation hype.
Sumner’s original announcement claimed the rewrite was achieved for $15,000 a day, a figure the author notes likely excludes the significant costs of the organisation’s Buildkite CI/CD cluster. The blogger observed that the Buildkite pipeline has been "constantly whirring" since the purported completion date, with merging open pull requests estimated to require 86 days of continuous pipeline running at current rates.
Further investigation into the codebase revealed that some pull requests were authored by Anthropic employees, suggesting the $165,000 figure may not account for all labour and resource inputs. The author pointed out that while the initial spike in Claude usage aligned with the rewrite period, Anthropic employee involvement in the Rust codebase has continued to ramp up, with commits continuing through weekends.
The analysis draws parallels to other projects, noting that Anthropic’s C compiler and Cursor’s FastRender web browser have seen no commits for months. The blogger concluded that while they are not opposed to AI, the current hype cycle surrounding its capabilities is excessive, urging caution when evaluating claims that support large corporate valuations.
