Notion AI lead Sarah Sachs speaks with Sophia Dew about designing the environments and context that agents use to complete workAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction..
Sachs describes using local file systems, scripts and deterministic code for repeatable operations rather than asking language models to regenerate every step. She discusses Vercel workers, sandboxes and the role of observabilityAn AI agent sandbox is an isolated execution environment that limits which files, processes, networks, credentials, and external systems an agent can access..
She also explains Notion’s approach to sharing organizational contextContext engineering designs how relevant instructions, evidence, state and tool information are assembled for a model's task. across agents and models, and discusses testing and evaluation of agent behaviorEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria..
Watch on YouTube
