Vasuman Moza argues that understanding how a business actually operates is a major constraint on enterprise AI adoption. He describes forward deployed engineers mapping undocumented processes, exceptions and human responsibilities before building agents around existing systems rather than replacing them wholesale.
JD Pruett presents an engagement assistant that synthesises notes and documents, followed by a workflow assistant that checks context and missing edge cases during implementation. He distinguishes these tools from a proposed autonomous assistant that could later handle small client-requested changes without routine engineer intervention.
JD Pruett describes representing operational dependencies in a graph and separating two model-training problems: writing useful analysis from context and retrieving the correct context. His account combines post-training for concise process descriptions with reinforcement-learning environments for specialised graph-traversal tools.
Watch on YouTube
