Agents can read a PDF in a demo. Feeding thousands of diverse documents into real-world decision-making is a different challenge, and it’s where automation fails silently. The first part of this talk covers what reliable document processing actually requires: extraction that can handle the documents that stump generic tools. Then we put this thesis to the test, in collaboration with Iberia’s tech team: how one of Europe’s largest airlines automated the processing of documents that stump traditional systems—including documents over 50 pages long with tables spanning every page—and what it took to trust that automation in operations. Because an agent that can’t prove what it has read isn’t truly autonomous. It’s just risk wrapped in an API.