Jo: Compiler-Enforced Capability Boundaries for AI Code Safety

Jo introduces ultra-precise permission control where capabilities are granted at compile-time with granular specificity—not just 'database access' but 'read-only queries on user-owned rows.' Its two-world architecture separates AI-generated code from trusted code, letting a compiler enforce that LLM-written functions access nothing beyond their declared interface, making it practical to safely integrate AI-assisted code without security compromises.