Manish Kapur of Sonar and Addy Osmani of Anthropic discuss how teams can maintain confidence as coding agentsAn AI coding agent is a tool-using AI system that can inspect, modify, and validate software within a repository. produce changes faster. They distinguish between software factories with active human involvement and more autonomous workflows.
Addy Osmani describes Anthropic’s use of an initial agent reviewCode review examines proposed source changes to assess their correctness, clarity and suitability before or around integration., browser and end-to-end checks, edit-triggered hooks and evidence for people to inspect. He emphasizes clear specifications, tests and shared team guidanceContext engineering designs how relevant instructions, evidence, state and tool information are assembled for a model's task..
Manish Kapur proposes multiple verification layers, including deterministic checksDeterministic software verification checks an output with a repeatable procedure that produces the same result for the same inputs and state., agent review and software supply-chain security. Both speakers discuss permissions, sandbox boundaries and the importance of retaining a human ownerHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work. for a change.
Watch on YouTube
