About the job
At Magic, we are on a mission to develop safe AGI that propels humanity's progress in addressing the world's most significant challenges. We believe that automating research and code generation is the most promising pathway to achieving safe AGI, enabling us to enhance models and address alignment issues more reliably than humans can achieve alone. Our innovative approach integrates frontier-scale pre-training, domain-specific reinforcement learning, ultra-long context, and inference-time computing to realize our vision.
About the Role
As a Software Engineer at Magic, you will engage in developing core systems and product surfaces that directly influence model capabilities and enhance user experience.
This position can align with areas such as Pre-training Data, RL Research & Environments, or Product Development, depending on your background and expertise. Regardless of placement, you will be expected to take full ownership of your work: identifying problems, crafting solutions, deploying to production, and iterating based on real-world results.
Working with Magic's long-context models presents unique technical challenges, including large-scale data acquisition, long-horizon post-training loops, and developing product workflows that make complex model behaviors understandable and manageable. You will work closely with these constraints, creating systems that are both technically sound and production-ready.
This role has the potential to evolve into a deeper specialization in data systems, post-training capability enhancement, or product engineering leadership based on your strengths and interests.
What You'll Work On
Depending on your team assignment, your tasks may include:
Developing and scaling large distributed data pipelines for pre-training
Designing filtering, mixture, and dataset versioning systems
Creating post-training datasets, evaluation frameworks, and reward pipelines
Conducting ablations that translate capability goals into quantifiable improvements
Building comprehensive product interfaces that integrate seamlessly with the model
Designing APIs, backend services, and frontend workflows for AI-first experiences
Enhancing the reliability, observability, and performance of production systems
What We’re Looking For
Solid foundation in software engineering principles
High ownership and comfort in navigating ambiguous problem domains
Proven experience in building scalable production systems
Ability to reason through complex technical challenges

