Business builders. Your judgement is needed. Artificial intelligence can write the code. But who checks the process? Who questions the assumptions? Welcome to the Human Agency.

A spreadsheet. A process document. An idea for an app by Friday. Before a coding tool builds anything, it may receive the business rules, the relationships and the sensitive details behind your work. The first governance decision happens before that prompt.

Meet PALO Document Workbench. A desktop pilot for the step before coding. It helps you turn PDF and Excel documents into an inspectable model, and then into a blueprint you review. You bring the process knowledge. Artificial intelligence proposes. You decide.

Mission one. Read the evidence. Import the included synthetic example, then inspect the candidates the model extracts. Follow a proposed rule back to its source. Check what the document actually says. A matching quotation is useful evidence, but it does not prove that the interpretation is right.

Mission two. Map what matters. Explore entities, attributes and relationships. Read the data dictionary. Inspect the process steps. Look for an assumed identifier, an invented constraint, or a relationship that does not fit your business. The graph is a proposal to challenge. Your knowledge makes the difference.

Mission three. Review before export. Include a candidate, correct its description, or exclude it. PALO requires your individual decisions before it prepares the blueprint for export. Then inspect what will leave the workspace. Business logic can still be confidential. Share only what your organisation authorises.

Your Mac runs the local inference engine. Open the desktop app and use its local browser workspace. The guided setup downloads approximately four point seven gigabytes of model files, with your consent. No separate Python or terminal setup is needed. Company inference endpoints require explicit configuration and approval from I T.

Three PALO layers inform the workflow. The Framework provides lifecycle and evidence references. PALO A M bounds the extraction task. PALO A I principles inform application checks. This pilot does not integrate the separate PALO A I production runtime. It does not certify compliance or control uploads to other tools.

This is a controlled pilot, and we need your critical judgement. Can you get started? Can you spot a wrong assumption? Can you understand the export decision? Start with the synthetic example. Report problems without sending confidential documents. Follow your organisation's policies.

Join the Human Agency. PALO Workbench, for Apple Silicon Mac. Volunteer for the guided pilot at PALO framework dot org. Build with A I. Keep humans in command.