From Artificial Intelligence to Institutional Intelligence

Generative artificial intelligence and agentic systems can redistribute reading, research, synthesis, criticism, drafting, and coordination within courts. This article asks how that redistribution changes judicial cognition and which institutional safeguards are needed to preserve legitimate adjudication. Through an interdisciplinary narrative review and doctrinal analysis, it develops a framework of institutional intelligence anchored in Brazilian National Council of Justice Resolution No. 615/2025, as amended by Resolution No. 674/2026. Five categories organize the argument: institutional cognitive space describes the allocation of cognitive work; artificial adversariality identifies structured internal challenge among agents; the institutional rein integrates controls over delegated autonomy; meaningful human initiative emphasizes the continuing human authority to frame problems and determine purposes; and institutional AI curation describes the competence required to select, validate, contextualize, and reject artificial contributions. Recent research on human–AI collaboration and multi-agent failure qualifies the framework: neither human review nor artificial disagreement guarantees reliability, and short-term performance gains do not establish long-term professional competence. The article distinguishes technical auditability from legal reason-giving and internal agent debate from the parties’ right to be heard. It proposes workflow-level evaluation and an illustrative protocol connecting permissions, source verification, escalation, and responsibility. Its contribution is a judicial governance synthesis, not an empirically validated architecture. Legitimate delegation depends on preserving institutional competence and contestability throughout the production of decisions, rather than relying exclusively on human approval of a final text.