Burnin builds hard environments that break AI agents. We design continuous-time evaluation and training arenas for multi-agent coordination, coding agents, computer use In these envs agents face ticking clocks, cascading failures, and rules nobody wrote down. Small team from Meta FAIR, Pinecone, and AWS. Seed round closing this summer.

We're looking for a founding research scientist to own the research agenda: invariant learning (extracting implicit codebase rules for real-time grading), continuous-time evaluation design (why do LLM agents collapse under temporal pressure?), and new environment design (what is the field failing to measure?).

You have deep experience in at least two of: multi-agent systems, RL, program analysis, or evaluation methodology. PhD preferred, equivalent output without one is fine. You write production code, not just experiment scripts. You think about where models break, not what they can do.

Apply here: https://wellfound.com/l/2Carrr

Founding Researcher