Member of Technical Staff, Research (+ Equity) at Primitive Labs
Primitive Labs is a venture-backed startup founded by former Amazon AGI leaders, building behavioral intelligence for AI-native software. Backed by a16z Speedrun and advisors from OpenAI and Google DeepMind, they are making human behavior a first-class primitive of software development to create agents that model real user decision-making.
About the role
You will own research directions from hypothesis to production, building the simulation architectures and memory systems that allow AI to model human behavior with high fidelity. By blending computational cognitive science with agent architectures, you'll ensure software understands real people, defining a research program through published papers and technical contributions.
Location: San Francisco and New York, USA
Why this role is remarkable
- Join a high-caliber founding team from Amazon AGI and AWS, working alongside advisors from OpenAI, Google DeepMind, and NVIDIA.
- Shape a brand-new research program from the ground up, with the autonomy to publish papers, contribute to open-source, and speak at conferences.
- Work on a unique frontier of AI—behavioral intelligence—that moves beyond generic task completion to high-fidelity human modeling and simulation.
Responsibilities
- Design and execute rigorous experiments for agent behavior, translating findings into model, memory, and runtime improvements.
- Build complex simulation architectures of real-world user decision-making to develop behavioral-fidelity benchmarks for agent performance.
- Train and evaluate behavioral intelligence agents against human data, ensuring AI decisions remain human-aligned and contextually relevant.
Requirements
- Proven track record leading research in machine learning, agents, simulation, or cognitive science, with a focus on human behavior.
- Exceptional engineering skills to build the systems needed to test research ideas, moving fluidly between theoretical questions and production code.
- Deep experience working with large models and a passion for behavioral fidelity rather than just optimizing for simple task completion.