The emergence of Large Language Model (LLM)-based AI agents marks a pivotal shift in how many scientists approach research. Agentic coding, automated data analysis pipelines, and LLM-assisted workflows are rapidly being adopted across the earth sciences. Yet, the community needs a dedicated forum to exchange experiences, critically assess implications, and chart a collective path forward.

This session invites earth scientists to share firsthand experiences with AI agents in research practice, addressing both practical and conceptual dimensions: from concrete productivity gains achieved through agentic coding and automated workflows, to deeper questions about how these tools are reshaping experimental design, model development, data analysis, and scientific reasoning itself across the disciplines. Crucially, we ask whether agentic AI is beginning to generate genuinely new scientific understanding - not merely accelerating existing workflows, but enabling discoveries or insights that would have been difficult or impossible to reach by conventional means.