Jobs › Argonne National Laboratory
Staff Scientist – Agents for Science
Posted Apr 27, 2026 · We last checked this listing on Oct 6, 2026
Apply at Argonne National Laboratory
Likely interview questions for this role
Written from this job description, not a generic list. Each one notes what the interviewer is really checking.
About your past
Tell me about an agentic or LLM-based system you built. What did you decide to let the model do on its own and what did you keep under human control?
What they are checking: Whether they make deliberate choices about autonomy and oversight when an agent can act on real systems.
Describe a time your research idea did not work. What did you do with that result?
What they are checking: Whether they learn from failed experiments and share negative results honestly, which matters in frontier research.
How do you decide between publishing a new method and building software that others will use?
What they are checking: Whether they value both research advances and deployable systems, as a staff scientist at a user facility must.
What would you do
An agent will have access to instruments and shared compute. How do you keep it from wasting resources or doing something unsafe?
What they are checking: Whether they build in guardrails, budgets and approval steps for agents acting in the physical and shared world.
How do you work with a domain scientist who is skeptical that AI agents can help their research?
What they are checking: Whether they can earn trust through real results and listen to the scientist's actual needs.
About the work
How would you design an AI agent that can plan and run a series of simulations on a supercomputer to test a scientific hypothesis?
What they are checking: Whether they can connect agent reasoning to real tools, schedulers and long-running jobs rather than toy tasks.
What are the biggest failure modes you have seen in agentic systems, and how do you guard against them?
What they are checking: Whether they have honest, hands-on knowledge of where agents break, such as looping, bad tool calls or made-up results.
How would you evaluate whether a science agent is actually helping scientists rather than just producing plausible output?
What they are checking: Whether they think rigorously about evaluation, which is the hardest open problem in agents for science.
How would you let several agents collaborate on one scientific problem, and when is a single agent better?
What they are checking: Whether they understand multi-agent designs well enough to know when the extra complexity pays off.
What would it take to scale an agent workflow from a laptop demo to thousands of concurrent tasks on HPC systems?
What they are checking: Whether they understand inference throughput, orchestration and state management at leadership-computing scale.
How should an agent handle knowledge sources like papers and databases so its claims can be traced back?
What they are checking: Whether they care about provenance and citation, since science cannot use answers nobody can verify.
Practice this interview out loud.
Offer builds a real interview for this exact role at Argonne National Laboratory from your resume and this job description, asks the questions one at a time, and tells you what landed. The first one is free.
Practice this out loudThe full job description
As published by Argonne National Laboratory.