How AI Agents Fit Into Your Research Workflow: Ideas, Experiments, and Paper Writing

In a research internship interview, you are asked how AI agents have changed your research. Describe your current day-to-day research workflow across three activities: coming up with and refining ideas, designing and running experiments, and writing papers. For each activity, explain where agents help you, where you choose not to rely on them, and how you check their output. ```hint Anchor it in one recent project Pick one recent project and describe what an agent actually did at each stage and what you checked afterward, instead of listing tools. ``` ### Constraints and Clarifications - Answer from your own practice. If you use agents only a little, or not at all for some activities, say so and explain why. - Do not claim to know how the prospective team uses automation in its research. Ask instead. - "Agent" can mean a tool that carries out several steps on its own, such as editing code, running scripts, or searching the literature. It can also mean a chat assistant. Make clear which kind you are describing. ### Clarifying Questions - Does the interviewer mean agents that act on their own, such as running code and launching jobs, or AI assistance of any kind? - Are there rules in this setting about sending unpublished data, code, or drafts to external AI services? - Is the interviewer mostly interested in productivity, or in how research quality and correctness are protected? ### What a Strong Answer Covers - A concrete breakdown across ideation, experimentation, and writing, naming the specific tasks delegated at each stage. - Verification habits: reviewing and testing agent-written code, sanity-checking experiment outputs, and checking every citation and factual claim. - Judgment about where agents are unreliable, such as novelty claims, invented references, and quiet bugs in evaluation code, with the researcher keeping responsibility for results. - Awareness of confidentiality, reproducibility, and authorship and disclosure norms for AI-assisted writing. - A view of how the workflow would change with more automation, and which decisions should stay with the researcher. ### Follow-up Questions 1. Describe a time an agent produced a plausible but wrong result. How did you catch it? 2. If an agent could run an entire experiment sweep overnight, how would you change the way you design experiments? 3. How do you decide whether an idea you developed with an agent's help is actually novel?