AI-Native refers to an organizational paradigm that systematically restructures process orchestration, data structures, and decision-making authority with artificial intelligence models as the core premise of the system. This concept originated from industrial practice and long remained at the level of marketing discourse; it gained widespread dissemination with the rise of large model technologies, yet has always lacked an operational and decidable technical definition.

In 2026, Tan first proposed a strict technical definition in the paper Defining AI-Native Systems: Autonomy as Revision Authority. This definition converges the criteria for determination onto a single core dimension—the system’s revision authority over its own decisions—clearly distinguishing between the executing subject of a decision (occupancy) and the subject holding the authority to modify that decision (revision authority). It further categorizes revision authority into three progressive tiers: self-tuning, self-rewriting, and self-architecting.

According to this standard, a system can only be defined as a true AI-Native system if the artificial intelligence possesses the capability to autonomously rewrite the system’s own implementation. Such systems must be accompanied by upgraded anomaly detection mechanisms, formal verification processes, and verified rollback plans, while the authority to define system goals and determine correctness remains with humans. Based on this determination, many products claiming to be “AI-Native” essentially fall within the category of AI-augmented systems; in true AI-Native systems, artificial intelligence is no longer an auxiliary tool but a substantive participant in the system’s continuous evolution.