OpenAI
Mid
Train frontier models to reliably operate computers, from data pipelines through major model runs
A research role on OpenAI's Agent Post-Training team focused specifically on computer use: designing experiments, building evals/environments/graders, and driving RL and data-pipeline improvements that shape agentic behavior shipped into Codex, ChatGPT, and the API. The role expects hands-on experience with LLMs, RL/RLHF/RLAIF, evals, and production ML systems, plus comfort operating in open-ended, noisy-signal research territory across research, product, infra, and safety boundaries.
Step into this interview
Free · a live voice mock calibrated to this exact role
What this interview tests
- Post-training/RL pipeline design for agentic computer-use behavior
- Building evals, environments, and graders that expose model failures
- Translating ambiguous behavioral problems into hypotheses and experiments
- Distinguishing benchmark movement from real product/user impact
- Cross-functional collaboration across research, product, infra, and safety
- Training-run velocity, reproducibility, and production readiness
Common question themes
Walk me through a post-training or RL experiment you ran on an agentic/tool-using model
How would you build an eval or grader to catch a specific class of computer-use failure
Describe debugging a hard, messy model failure and turning it into a concrete fix
How do you decide whether a model improvement is ready to go into a major training run
How do you balance research exploration against the need for reliable, production-ready agent behavior
Related interviews
OpenAI
Senior
Forward Deployed Engineer
OpenAI
Senior
Data Scientist, Identity
OpenAI
Senior
AI Deployment Engineer
Cohere
Senior
Senior Software Engineer, Agent Infrastructure
Figma
Senior
Software Engineer, AI Platforms
Mid