Anthropic Research Engineer, Claude Agent Architecture
Barry Zhang
Profile
Barry Zhang is a research engineer at Anthropic, where he works on Claude’s agent architecture and the Skills framework — the two pieces that increasingly define how Claude behaves when it’s not just chatting but actually doing things in the world. He sits on the Applied AI team, which means his job is less about pushing the frontier of pretraining and more about figuring out what production-grade agents actually need to work for real customers. If you’ve read Anthropic’s engineering posts on agents and walked away thinking “huh, that’s the first sensible thing I’ve read on this topic” — there’s a decent chance Barry wrote it.
Before Anthropic, Barry was tech lead on Meta’s Monetization genAI team, where he reportedly claimed the company’s inaugural “AI Engineer” title. Earlier he worked on Meta’s knowledge graph team. He picked up both his bachelor’s in Industrial Engineering and his master’s in Computer Science from Northwestern University, finishing in 2021 — which makes him a relatively young voice for someone whose writing is now the de facto reference text on agent architecture inside a lot of engineering orgs.
His real influence comes from two pieces of writing. The first, “Building Effective Agents” (co-authored with Erik Schluntz in late 2024), pushed back hard against the prevailing LangChain-era complexity in agent frameworks and made the case for simple composable patterns: prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer. It became one of the most-cited engineering posts of the year — Simon Willison and most of the practical-AI crowd treat it as canon. The follow-up, “Equipping Agents for the Real World with Agent Skills” (October 2025), introduced Skills as the unit of agent capability: folders of instructions, scripts, and resources that agents discover and load on demand. Skills is now baked into Claude.
What makes Barry worth following for developers is that he’s allergic to the thing that ruins most AI engineering content — abstraction for its own sake. His repeated public refrain is “don’t build agents for everything, keep it simple, think like your agents,” and he’s willing to argue, on conference stages, that the model underneath is more universal than people think and that most teams should build skills rather than yet another bespoke agent. If you’re shipping anything agentic, his work is the shortest path from toy demo to something that doesn’t fall over in production.
Key Articles & Papers
Building Effective Agents Equipping Agents for the Real World with Agent SkillsVideos
YouTube
