Coding agents traditionally operate on files, which differs significantly from interactive data analysis workflows. Data work involves loading data, inspecting it, running code, and observing real-time responses in memory. Previously, AI agents could not access this live, in-memory state, only the underlying files. This limitation is now being overcome. Marimo Pair introduces a coding agent that integrates directly into a running notebook environment. This agent gains full access to all Python variables held in memory. The notebook effectively becomes a collaborative workspace between the user and the agent. Users can direct the agent, for example, by asking it to focus on specific data points, and the notebook's visualizations will update accordingly. This integration eliminates the need for extensive setup or schema descriptions. It allows the agent to "see" and interact with the live data state just as the user does.