Richard Feldman describes a common handoff problem in agent-assisted development: a branch and transcript may show the code but omit abandoned approaches and the reasoning behind decisions. He argues that teams need access to the context in which the work evolvedContext engineering designs how relevant instructions, evidence, state and tool information are assembled for a model's task..
He demonstrates a collaborative workflow that brings agent conversations, code review and teammate discussion togetherAI-assisted code review uses AI to inspect software changes for defects, risks, standards, missing tests, and opportunities for improvement while human reviewers retain responsibility.. He shows how a colleague can join a thread, ask the agent for context and inspect shared changes while retaining the discussion for later readers.
In questions, Feldman discusses observed collaboration scale, local and cloud execution, review sign-off and the limits of the current product. The demonstration reflects the speaker’s own tool and claims.
Watch on YouTube
