AI System Requirements

The bottleneck in AI development has shifted from coding to the stage of defining Intent. Thanks to tools like Claude Code and Cursor, converting natural language into code has become extremely fast, but a problem has emerged where humans themselves cannot clearly define what they want.

Humans are poor at creating abstract rules, but skilled at critiquing specific examples. Because of this trait, when building agent-based systems, significant friction arises in the process of defining infinite, recursive behavioral systems as finite, linear commands.

Simply injecting data is prone to failure. Even if vast amounts of data are learned through RAG, it fails to capture the organizational Context and tacit knowledge that isn't contained in the data. Data only captures text, not the underlying rationale behind judgments.

Also, AI agents often only understand a user's words literally, missing the actual intent. According to Speech Act Theory, language is not just a means of conveying information but also a means of performing specific actions. The ability to distinguish whether a user's statement is a simple statement of fact or an instruction demanding a specific action is essential.

Claude Code's interview feature can resolve technical ambiguity, but it cannot resolve business or conceptual ambiguity. Ultimately, the core of building AI systems is not simple code writing, but the process of precisely extracting and specifying intent that lies in the human subconscious.