Jarred Sumner once spent three weeks hand-porting a Go transpiler to Zig. No AI, no shortcuts. That code became the seed for Bun, the JavaScript runtime now powering some of the most aggressive AI coding tools on the market. Today, a GitHub bot called robobun has more contributions to the project than Sumner himself.
Five months ago, Anthropic acquired Bun outright. The deal, announced December 2, 2025, tied the runtime directly to Claude Code, Anthropic’s AI coding product that hit $1 billion in annualized revenue within six months of launch. Sumner’s reasoning was pragmatic: by late 2024, AI coding tools had crossed from demo to production, and a disproportionate number shipped on Bun.
Claude Code distributes as a single-file Bun executable to millions of users. That architectural choice—fast startup, native addon support, zero-dependency distribution—made Bun the default runtime for several AI-first developer tools including FactoryAI and OpenCode. When those tools succeed, Bun must not break. Anthropic now has every incentive to keep the runtime stable and fast.
Sumner’s own path to the acquisition was personal. He took four-hour walks around San Francisco with Anthropic engineers, discussing where coding tools were heading. He repeated those walks with competitors. He chose Anthropic. “This feels approximately a few months ahead of where things are going. Certainly not years,” he wrote at the time.
The numbers explain the urgency. Bun’s monthly downloads climbed 25% in October 2025, crossing 7.2 million. The project carried over four years of runway but generated zero revenue. Traditional paths—cloud hosting, paid tiers—felt mismatched for a world where AI agents write, test, and deploy most new code.
Bun’s original engineering bet was simple: write everything from scratch in Zig, avoid incremental fixes until the full picture emerges, and favor breadth-first exploration. That approach produced a runtime that transpiles JSX three times faster than esbuild and 197 times faster than Babel. By July 2022, Bun v0.1 combined bundler, transpiler, runtime, test runner, and package manager into a single binary. It hit 20,000 GitHub stars in a week.
The real inflection point came when AI coding tools discovered Bun’s single-file executables. Developers could bundle entire JavaScript projects into binaries that run anywhere, even without Bun or Node installed. Startup stayed quick. Native modules worked. Distribution simplified. The same traits that solved Sumner’s original 45-second iteration pain now solved distribution pain for AI-powered CLIs.
Post-acquisition, Bun remains open source under MIT license. The same team continues development publicly on GitHub. Node.js compatibility stays a priority. The roadmap now aligns more closely with Claude Code and the Claude Agent SDK, but retains independence similar to browser engines and their JavaScript runtimes.
Robobun’s lead in contribution count adds another layer. The bot handles force pushes, labeling, bug fixes, and test writing. It responds to review comments. In one setup, it tests fixes against earlier Bun versions before merging. Sumner has praised the productivity gains while acknowledging the shift in metrics. Traditional contribution graphs once measured human effort. They now capture a mix of human direction and machine execution.
Questions remain. How will contribution credit evolve when bots outpace founders? What does code ownership mean when agents generate the majority of new lines? Will runtime performance still dominate when humans review less of the output? Sumner has wagered that fast, predictable tooling becomes even more valuable in that world.
He is hiring. The team ships updates at a quick clip. Bun v1.3.13 arrived with parallel test improvements, lower memory usage for installs, and better source map handling. Each release tightens the loop between human intent and machine output. The original frustration—45 seconds to check if a change worked—feels quaint. Today the constraint is how quickly an agent can propose, validate, and deploy across thousands of lines.
Sumner once coded in a cramped Oakland apartment, tweeting progress between commits. Now he walks San Francisco streets with AI product teams and watches bots merge more PRs than he does. The project he started to solve his own iteration pain has become infrastructure for tools that multiply developer output by orders of magnitude. And Anthropic paid to own the stack underneath it all.
The numbers keep moving. Downloads rise. Revenue at Claude Code compounds. Robobun’s commit count grows. Bun itself ships faster than before. The question is no longer whether AI will change software engineering. It already has. The question is who controls the runtime that agents rely on when most code never passes through human hands first. For now, that runtime is Bun. And its creator no longer holds the top spot on its own contribution graph.
Source: Webpronews