Who needs structure?

Now that we have HTR and large language models, do we still need to manually structure historical data — or can we extract everything we need directly from the text? This roundtable challenges participants to think about what comes after transcription at scale: whether to start top-down by identifying document groups first, or bottom-up by extracting entities from individual pages; whether a triage phase is necessary before AI can be effectively applied; and where human expertise still fits in an increasingly automated pipeline. The discussion brings together Dutch archives automating metadata generation from backlogs, projects building pipelines for structured genealogical data, and researchers navigating diverse multilingual historical collections — all asking the same fundamental question from very different angles.