The New Software Lifecycle

This article discusses a Google whitepaper on how AI is changing the software lifecycle, introducing concepts like 'vibe coding' and the shift from writing code to judging it. It emphasizes that an AI agent consists of a model plus a harness, where the harness (instructions, tools, sandboxes, orchestration) is often more critical than the model itself. The author highlights that most agent failures are configuration issues, not model flaws, and that context engineering—deciding what goes into static versus dynamic context—directly impacts costs and performance. The piece includes six reusable figures and practical insights for developers working with AI agents.