Product Manager, Claude Code Model Performance

Anthropic•San Francisco, CA

6h•$305,000 - $460,000•Hybrid

About The Position

As a Product Manager on Claude Code's model performance team, you will drive model launches end-to-end, build evals that measure what matters, and partner directly with researchers and product engineers to translate model improvements into developer-facing outcomes. Claude Code is the most capable coding agent in the world but there’s much more we can do to extract the maximum performance from our models. We're looking for a PM who has personally built agentic evals, thinks in systems, uses Claude Code every day, and has refined model taste. You should be as comfortable influencing our research team as you are getting in the weeds of transcripts. You will be the connective tissue between frontier research and the millions of developers who depend on Claude Code to do their best work.

Requirements

Have personally built agentic evals (e.g. SWE-bench-style task suites)
Are a daily Claude Code user and can articulate what behaviors you’d want to change or add to the model
Have an engineering background and 2+ years in product management, or equivalent experience driving product direction as an engineer
Have a deep grasp of AI concepts and are comfortable going deep on model behavior, prompt engineering, and evaluation methodology
Are a systems thinker: when you find a problem, you build the infrastructure that prevents its whole class
Have launched products or capabilities in ambiguous, research-adjacent environments
Have a creative, hacker spirit and love solving puzzles

Responsibilities

Own model launch planning and execution for Claude Code: define readiness criteria, coordinate across research and product engineering, and ensure launches land cleanly with developers
Design and implement agentic evals that measure real-world coding performance
Drive the engineering team's eval roadmap
Partner with researchers working on coding capabilities to define target behaviors and influence model development with evidence from real usage
Talk with users and analyze transcripts to understand capability gaps and turn research progress into shipped improvements
Synthesize signal from internal users, external developers, and competitive benchmarks into clear priorities