Research Scientist

About the Company

A days-old, pre-seed AI company building agents that take ownership of complex objectives and pursue them autonomously for days, months, and eventually indefinitely, a new class of long-running agents that keep making progress toward an outcome instead of stopping after a single task. The founding team includes leaders and founding members from top frontier AI labs and elite quantitative firms. The core thesis is that as agents work over longer horizons, the bottleneck shifts from capability to intent: understanding what a person or organization actually wants, preserving that understanding over time, and knowing when human judgment is truly needed.

About the Role

You'll train agents that use computers and tools to pursue complex outcomes over long horizons, owning ambitious research bets end to end, from hypothesis and data through training, evaluation, and deployment. The work centers on post-training LLM and multimodal agents with reinforcement learning and related methods, and on long-horizon agent capabilities: planning, memory, recovery, proactivity, self-improvement, and calibrated escalation. A core thread is developing methods for agents to understand and preserve user intent across long-running trajectories. This is a build-real-systems research seat, not a place for isolated prototypes. As an early member, you'll help shape the research agenda and technical direction.

What You'll Own

• Post-training LLM and multimodal agents with reinforcement learning and related methods
• Research on long-horizon capabilities: planning, memory, recovery, proactivity, self-improvement, and calibrated escalation
• Computer-use and tool-use environments, training-data pipelines, benchmarks, and evaluations
• Methods for agents to understand and preserve user intent across long trajectories
• Rigorous experiment design, identifying the highest-leverage problems, and turning results into deployed capabilities
• Helping shape the research agenda and technical direction

What We're Looking For

• 2 to 7 years of experience (frontier-lab research or engineering experience is valued, with room for exceptional outliers)
• Exceptional ML research and engineering ability
• Deep experience in one or more of: LLM post-training, RL, agents, reasoning, long-horizon systems, memory and context, computer use, evals, or human-agent interaction. Pure classical ML or narrow non-agent work is a weaker fit.
• Strong experimental judgment, and the habit of building complete systems (environments, pipelines, benchmarks, evals) rather than isolated prototypes
• The ability to move fluidly between research ideas, large-scale experimentation, and production systems
• A record of significant research ownership: an important model, agent system, benchmark, paper, open-source project, or major research bet
• High agency and ambition, with comfort pursuing uncertain directions where the right approach isn't yet known
• Bonus: a strong publication, open-source, or benchmark record in these areas, experience building environments, data pipelines, or eval harnesses, or a "spike" background such as olympiad, quantitative, or world-class competitive achievement

Logistics

San Francisco, in person, with an intense pace (9-9-6). Compensation is $250K to $500K base plus 1% to 5% equity. Visa sponsorship is available. The process includes a paid, in-person work trial of 2 to 3 days on a live research work stream.