Laurie Voss describes a growing mismatch between rapid AI code generation and the much slower process of establishing whether a change should ship. He contrasts code-output growth with delivery gains and argues that asking people to review ever larger diffs is neither scalable nor sustainable.
Laurie Voss distinguishes passing tests from being safe and maintainable enough to mergeAgent completion verification checks evidence that an agent achieved the requested outcome rather than merely producing intermediate work.. He discusses benchmark gaps, automated reviewers' false-positive filteringEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. and feedback from human acceptanceAn AI feedback loop occurs when AI outputs or their consequences become inputs that influence later model or agent behavior., while warning that these mechanisms still depend on what the review system can observe and evaluate.
Laurie Voss examines attempts to remove direct human review and argues that the work often moves into human-designed tests, harnesses and oversight instead. He highlights security, prompt-injectionPrompt injection occurs when untrusted content attempts to redirect an AI system's behaviour beyond the authority of that content. and untested-context risks, then recommends encoding organizational standards and domain knowledge into a reliable review system and monitoring behavior after deployment.
Watch on YouTube




