How to Pass Context in an Agentic AI Flow

Agentic AI systems require a fundamentally different approach to context than traditional OAuth flows. While traditional context covers who the user is, what resource they access, and what they're permitted to do in a single login session, agentic systems introduce complexity through autonomous, non-deterministic behavior, multi-agent orchestration, LLMs, and MCP servers. Context in agentic flows must account for situational context (environment state), resource context, user context, model context, and task history. Prompt engineering alone is insufficient — context engineering is needed to incorporate all these variables and ensure agents retrieve the right information and respect appropriate permissions.