Jarred Sumner

Sources#

Summary#

Creator of Bun. Wrote his first line of Zig on April 16, 2021, having bet on the language after reading its single-page reference on Hacker News, and built the initial version of Bun solo in one year, pre-LLM, in "a cramped Oakland apartment." Bun was acquired by Anthropic in December 2025; Sumner and the Bun team now work there.

His July 2026 post Rewriting Bun in Rust is the wiki's most methodologically detailed practitioner account of a large dynamic-workflow campaign — and the source that corrects Boris Cherny's secondhand stage version of the same project.

Positions worth carrying#

Language choice stopped being one-way. "Until very recently, programming language choice was a one-way decision for a project like Bun." The claim under the whole post is not that Rust is better than Zig but that the cost of switching collapsed. He is careful not to blame Zig: "other users of Zig don't have the bugs we had… I'll always be grateful."
Compiler errors beat style guides. The rewrite's motivation is a feedback-loop argument, not a performance one: style guides (TigerStyle, Google's 31,000-word C++ guide) depend on best-effort enforcement through code review, whereas safe Rust turns the same rules into compile failures.
Fix the process, not the code. His stated response to every workflow failure: "fixing the process that generates the code instead of hand-fixing the code." Each of the campaign's false starts was closed by editing a workflow prompt.
Review moves up an altitude. He did not read a million lines. He reviewed whether the adversarial reviewer agents were catching real Zig/Rust discrepancies, plus manual side-by-side sampling (Review as the Control Point).
The closing claim. "One engineer can do a lot more today than a year ago" — offered against a stated counterfactual of three engineers for a year, and a real alternative of doing nothing.

Evidence weighting#

Everything Sumner reports here is case-study with an explicit disclosure: an Anthropic employee, using a pre-release Claude Fable 5 and Claude Code, describing his own team's project. The account is unusually candid about failures (agents git reset-ing each other, stub-out gaming, disk exhaustion, 19 shipped regressions), which is what makes it useful — but it is a build log with no control arm, not a measurement.

The first outside audit of the same project (Tom Lockwood, 2026-07-27) goes out of its way to exempt him: "The Bun team never made that claim." Its target is the secondhand repetition of "the rewrite was done for $165k" — a claim Sumner's post does not make and in fact undercuts, since he states plainly that merging to main is not a versioned release. What the audit does establish against the post is an accounting boundary Sumner never draws: the ~$165k is tokens to green, and the CI, employee time, and post-merge tail sit outside it while the "three engineers for a year" counterfactual is fully loaded. See Dynamic Workflows: An Algebra for Agents.

Connections#

Bun — the runtime he created and ported
Dynamic Workflows: An Algebra for Agents — the orchestration primitive whose published methodology is his
Boris Cherny — told the same story secondhand a month later, with two details this account supersedes
Anthropic — employer since the December 2025 acquisition
Claude Fable 5 — the pre-release model the port ran on

Sources#

Rewriting Bun in Rust — Jarred Sumner, bun.com (2026-07-08, case-study)
How is the Bun Rewrite in Rust Going? — Tom Lockwood, lockwood.dev (2026-07-27, case-study, independent): the outside audit, which explicitly exempts Sumner from the claim it targets