Benjamin Clavié describes knowledge work as turning information into actionable judgments. Code is a special case: stable identifiers and explicit task boundaries make it easier to search, while legal, financial and research questions often require an agent to infer context and break down an open-ended request.
Benjamin Clavié uses retrieval benchmarksRetrieval is the process of selecting relevant stored information and returning it to an AI system for the current task. to argue that tool quality affects both answer accuracy and operating cost. He stresses optimized lexical baselines, multimodal access to documents and measuring how many searches an agent needs, rather than treating every accuracy change as meaningful.
Benjamin Clavié then describes an orchestrator that delegates focused research and receives short evidence memos before composing an answerAgent orchestration coordinates AI agents, tools, people, tasks, state, and control flow so a larger workflow reaches a verified outcome.. His reported results illustrate the potential of better search organization, while the broader lesson is to design retrieval tools and agent workflows together instead of relying on ever-larger context windows.
Watch on YouTube




