Agentic systems
Agentic systems driven by large language models increasingly run real workflows, calling tools that read and change persistent data. Before such systems are deployed, they need guarantees about the system as a whole, not only about the model’s individual answers. We formalise deployments that combine an LLM, a tool-orchestration harness and relational operational data, and study how to verify them against the business requirements that govern how a workflow runs and how its data may evolve.


