AI features in React apps are a negotiation between two opposing forces: autonomous agents and deterministic workflows. But instead of splitting the difference, what if we could get these opposing forces to work together safely and predictably?

Introducing Chartering: Rather than generating tool-calling code, the LLM authors the workflow itself as a charter, a declarative state machine. Separating implementation from orchestration gives us an iterative development lifecycle, complete with validation, testing, data privacy, and governance. Crucially, this is a runtime architecture: those guarantees reach the end users interacting with the live agentic features in your React app, not just developers at build-time. The same architecture runs in production today at Navion, an AI-native logistics platform coordinating thousands of fuel deliveries, where a dispatch assistant authors charters and a durable, governed runtime executes them in front of real customers.

The end result unlocks the strengths of both sides: an agent that's free to think and safe to ship.

Target audience: frontend and full-stack engineers adding agentic features to products, especially those already using or curious about state machines and durable execution. No ML background required; prior exposure to state machines helps but isn't assumed.

The architecture is presented as a vendor and product-neutral mental model, grounded in real production use at Navion Logistics (BlueFolders.ai) — a pattern to apply with an open source implementation, not a product to buy. Technical requirements: standard A/V (HDMI), confidence monitor preferred; no special setup.