<p>This article examines the conditions under which artificial intelligence can be integrated into legal processes without displacing the normative structure on which legal decision-making depends. Its central claim is that the main difficulty raised by AI in law is not primarily technical but theoretical. It concerns the location of legal normativity and the process through which legal norms are produced from legal texts in relation to social reality. The article argues that Friedrich Müller’s Structuring Theory of Law provides an especially fruitful framework for addressing this difficulty because it distinguishes between the normtext, the normative programme, the normative sphere, and the concretisation process through which a legal norm is produced. On that basis, the article proceeds in four steps. First, it revisits the distinction between vagueness and open texture in legal language, arguing that large language models can sometimes mitigate lexical instability but cannot resolve the open texture that arises from the relation between legal language and social reality. Second, it clarifies the principal concepts of the Structuring Theory for readers unfamiliar with that tradition. Third, it develops a typology of legal automation consisting of automation of access, automation of subsumption workflows, and automation of authoritative decisions. Fourth, it analyses the way AI systems increasingly shape the social reality that legal actors subsequently encounter as the normative sphere of legal norms. The conclusion is not that law must reject automation, but that the compatibility of automation with law depends on the stage of concretisation at which it intervenes and on whether accountability, contestability, revisability, and institutional attribution are preserved.</p>