AI-Native Manifesto: Six Gaps to Assurance Closure in Large-Scale Agile Development

A recent paper on arXiv (2608.07317) discusses the journey towards assurance closure in AI-driven large-scale agile software development, as outlined in the AI-Native Manifesto. This manifesto envisions a scenario where human oversight governs intent, risk, and exceptions, while AI agents handle a significant portion of the engineering tasks. The authors contend that reaching this goal necessitates more than improved code generation; it demands assurance closure, which involves defining truths, gathering credible evidence, maintaining validity amid changes, and managing uncertainty to limit agent authority. They highlight existing tools from formal methods, testing, and digital twins but point out six remaining gaps that hinder machine-operable assurance reasoning. To tackle these, they suggest a high-level architecture featuring six capabilities based on a common semantic assurance layer and pose four research questions for future exploration.