Teams at Cognition

Conduct AI safety and alignment research for autonomous software agents. Build release-gating evaluations, red-team models and agent harnesses, develop alignment methods, and improve permissions, oversight, escalation, and model behavior. The role combines empirical research engineering with production agent development, requiring collaboration across safety, post-training, and agent teams, plus contributions to external safety research.

Lead Cognition’s safety function for Devin and its underlying models. Set safety strategy, build evaluations and red-teaming for agentic risks, shape model training and agent-harness safeguards, establish deployment policies, and hire and lead safety researchers and engineers. The role requires hands-on frontier AI safety or alignment experience, agentic systems expertise, research credibility, technical fluency, and strong judgment under uncertainty.

Own late-stage LLM training decisions across data mixing, capability injection, synthetic data, annealing schedules, context extension, evaluation, and scaling. Develop methods that improve coding, mathematics, and long-horizon reasoning while translating research into production training systems. The role requires end-to-end understanding of the LLM pipeline, hands-on large-model training experience, strong optimization and machine learning fundamentals, and demonstrated original technical contributions.