Helix AI Engineer, Agentic Systems

About the role

Figure is pioneering a new era in robotics by designing autonomous humanoid robots capable of human-level intelligence within home and commercial environments. This specific role is centered on the development of advanced agentic architectures that serve as the cognitive core of these machines. You will own the design and implementation of the systems that allow robots to perceive their surroundings, retain experiences, and perform intricate sequences of actions without constant human direction. The position requires a deep partnership with cross-functional teams to ensure that the software capabilities align with the physical and operational goals of the humanoid platform. You will be responsible for bridging the gap between high-level artificial intelligence research and real-world deployment on embodied hardware. Success in this role means creating agents that can operate reliably for extended periods in unstructured settings. Your work will directly influence the autonomy and adaptability of the next generation of assistive and service robots. This is an opportunity to define the software backbone of a machine that interacts with the physical world.

Key facts

What you'll do

• Architect and deploy multimodal agents capable of autonomous operation over long durations without degradation in performance.
• Transform raw sensory data, including visual pixels and robotic proprioception, into coherent and actionable outputs for the robot.
• Construct episodic memory systems that enable long-horizon reasoning, allowing the agent to recall past events and states when making decisions.
• Develop sophisticated planning and tool usage mechanisms that facilitate complex multi-step task execution in dynamic environments.
• Implement stable perception-to-action loops that incorporate robust failure detection and recovery strategies to ensure continuous operation.
• Design and maintain comprehensive benchmarks and evaluation harnesses to quantitatively track agent reliability, safety, and reasoning performance over time.
• Conduct in-depth data studies across the entire training lifecycle, analyzing outcomes during both pretraining and post-training phases to optimize agent behavior.
• Apply reinforcement learning and reward modeling techniques to fine-tune agent actions, ensuring alignment with human intent and task objectives.
• Build and maintain scalable infrastructure that supports distributed experimentation, rapid prototyping, and large-scale model training workflows.
• Partner closely with internal engineering and research teams to seamlessly integrate agent models into the broader humanoid autonomy stack.
• Evaluate emerging methodologies for sensory processing to ensure the robot can interpret complex scenes and abstract concepts accurately.
• Optimize the efficiency of agent computations to meet real-time operational constraints on physical hardware platforms.
• Document architectural decisions and system behaviors to facilitate knowledge transfer and long-term maintainability of the codebase.
• Lead the definition of agentic success metrics, translating high-level product goals into measurable technical requirements.

Requirements

• Demonstrate a proven history of building autonomous agents that can execute multi-step tasks continuously in real-world or simulated environments.
• Possess direct experience developing agents that reason directly from raw environment observations, including pixel inputs from cameras and other sensors.
• Show practical knowledge of implementing systems for agent memory, planning, and sophisticated tool use to extend cognitive capabilities.
• Have a background in fine-tuning or training foundation models, with a specific focus on multimodal data that combines text, images, and other modalities.
• Exhibit proficiency in Python and deep learning frameworks, specifically PyTorch, for building and training complex neural network architectures.
• Display the ability to design, analyze, and iterate on machine learning systems with a high degree of experimental rigor and scientific methodology.
• Maintain strong software engineering skills necessary for building maintainable, reliable, and production-grade systems that can scale effectively.
• Demonstrate the capability to manage complex technical problems independently, guiding a project from initial conception through to final deployment and optimization.
• Commit to adhering to strict safety and reliability standards when developing components that interact with the physical world and human users.
• Bring a meticulous approach to debugging and testing agent behaviors to ensure robustness and predictability in diverse scenarios.

Nice to have

• Hands-on experience with embodied AI, robot policy training, or direct work within robotics simulation and control frameworks.
• A background in building or contributing to vision-language or vision-language-action foundation models that integrate visual and textual understanding.
• Expertise in long-horizon planning architectures and advanced agentic AI systems that handle complex temporal dependencies.
• Experience managing large-scale distributed training systems, including infrastructure on cloud platforms and high-performance computing clusters.
• A record of high-quality publications in top-tier conferences and journals focused on robotics, embodied AI, or machine learning.

Skills & tools

• Python
• PyTorch
• Multimodal foundation models
• Reinforcement learning
• Reward modeling
• Distributed training systems

Practical notes

Compensation is determined based on individual skills, experience, and job-related knowledge. Additional benefits and compensation components may be provided, which will be detailed if an offer is extended. The role is based in San Jose, Canada, and requires eligibility to work in that country. Full-time engagement is expected, with standard working hours applying. There are no specified travel requirements, visa sponsorships, or publication obligations listed for this position.