Why Bun switched to Skillsync to hire systems engineers

In conversation with

Conner PhillippiCOO, Bun

Before Bun's acquisition by Anthropic in 2025, Conner led hiring for one of the hardest engineering searches in the industry: finding low-level systems engineers willing to work on-site in San Francisco on a runtime, bundler, transpiler, and package manager all built for maximum performance. Today, the team he helped build is working on critical infrastructure at Anthropic, including Claude Code.

Hiring great systems engineers is unusually hard. Hiring great systems engineers who want to build an open-source JS runtime written in Zig, 100% on-site in SF? Even harder.

These developers don't optimize their LinkedIn profiles. Their real work lives on GitHub: contributions to runtimes, compilers, browser engines, and performance tooling. If an engineer didn't build in public, Bun didn't think they'd be a good fit.

When Conner Phillippi joined as COO, hiring engineers was one of his top priorities. The company hadn't made an engineering hire in at least a quarter, and the problem was both process and targeting. He tried everything: publishing on job boards (YC, LinkedIn, X, Kleiner Perkins), doubling down on LinkedIn Recruiter, and even writing a custom Python script that pulled GitHub contributors by language and repo. Inbound was noisy. LinkedIn didn't yield results. The script was better but slow.

Then he found Skillsync on X. And within two minutes, half the team was on the first page of results.

Bun's end-to-end offer rate hovered around 1-1.5%. They were doing four to five interviews a week, but most candidates couldn't clear a notoriously difficult interview process. The first on-site was a Zig debugging exercise, and AI tools were explicitly encouraged. The funnel wasn't a volume problem. It was a targeting problem.

Skillsync flipped the approach. Instead of casting a wide net and filtering down, Conner searched for exactly what Bun needed: low-level systems engineers who had contributed to game engines, open-source runtimes, bundlers, and JavaScript tooling, based in SF.

The first three pages of Skillsync results surfaced Bun's entire existing team, plus a pipeline of candidates the company either already coveted or hadn't discovered yet. It replaced LinkedIn Recruiter, job boards, and the custom Python script, all with a single search.

Skillsync matched on the signals that actually mattered: contributions to specific repositories (JavaScriptCore, WebKit, V8, SpiderMonkey, Hermes), proficiency in C, C++, Zig, Rust, and Go, and verifiable public work.

This aligned with how Bun already evaluated talent. Conner describes spending 20 to 30 minutes in someone's GitHub profile, examining commit history, problem-solving approach, project choices, as remarkably revealing, even in an AI-assisted world.

Bun's interview process even encouraged AI tool usage: Claude Code, Cursor, anything. Because systems programming is hard regardless. And as Conner puts it: give a great engineer AI and they're 10x. Give a weak engineer AI and it exacerbates the gap.

The principle is straightforward: when you source from proof of work instead of profiles, the candidates who enter your pipeline are more likely to clear a high bar. The kind of engineers Bun needed are now working on some of the hardest infrastructure problems at Anthropic:

Beyond technical signals, Conner shared the qualities that defined Bun's best engineers, attributes that now inform hiring at Anthropic:

Velocity. "Take the most complex project and ask: what would it take to ship this tomorrow? People who figure out how to make things happen versus give a list of reasons why something can't."

Strong opinions, loosely held. "I would rather have conflict over non-conflict any day. It means you care. People that don't give an opinion just don't care."

Intellectual flexibility. The best engineers update their thinking when presented with better information. Stubbornness kills velocity.

Open-source passion. Game engines, browser engines, compilers, deeper JS tooling. Contributors to WebKit, V8, SpiderMonkey, and Hermes were especially strong signals.

This approach also works for non-technical recruiters. By having a clear conversation with the engineering lead about what attributes matter, like languages, target repositories, and types of contributions, anyone can build a strong mental model for what a great candidate looks like, even without writing code themselves.

Skillsync replaced LinkedIn Recruiter, job boards, and a custom GitHub scraper with a single search. Within minutes, Conner found candidates that matched the exact profile Bun needed, engineers who had been invisible through traditional channels.

For teams hiring deeply technical roles where LinkedIn is a poor signal, Skillsync surfaces the candidates that traditional recruiting tools miss entirely. It works because it searches for proof of work, not profiles.