Product Manager, Physical AI Agents

Job description

Job responsibilities

• Own the roadmap. Define the strategic priorities and feature roadmap for our physical AI agent product, from on-robot policies to the developer SDK to the cloud-side evaluation and deployment surfaces.
• Translate frontier capabilities into product. Work closely with our research team to turn advances in VLAs, imitation learning, and reinforcement learning, world models, etc.
• Deeply understand the people building on us. Spend real time with developers, researchers, and enterprise users. Watch agents fail in the wild.
• Define what "good" means. Build the evaluation frameworks, metrics, and review cadences that make agent quality measurable — success rate, recovery behavior, safety envelopes, time-to-first-skill, intervention rate.
• Shape the open ecosystem. Decide what we open source, what we publish, and how we cultivate a community of builders. Help define how our platform boosts physical AI development.
• Drive cross-functional execution. Align research, engineering, hardware, design, and GTM around a coherent product narrative. Make tradeoffs in writing. Defend them. Revisit them when the data says you were wrong.

Job requirements

• 5+ years of product management experience on technically deep products — AI/ML, agents, developer platforms, robotics, autonomy, or similar.
• Strong AI agent sense. You have an opinionated, current view of where agents are going. You can look at a model release, a research paper, or a competitor launch and tell us what it means for our roadmap.
• Deep technical fluency in the modern AI stack: VLAs, imitation learning, RL, evaluation methodology, data pipelines, harness engineering, RAG, tool use, etc.
• Comfort at the hardware/software boundary. You understand sensors, actuators, latency, failure modes, and the gap between sim and real.
• Excellent product judgment under ambiguity. You make crisp decisions with incomplete information, communicate them in concise written documents, and update fast when you're wrong.
• Genuine love for builders. You care about the experience of the developer, researcher, or operator using our platform — and it shows in the product.
• High agency and bias for shipping. You ship v1s quickly and iterate to quality. You don't wait for permission. You don't manage a backlog; you define a category.