Agentic framework 2.1.1: weigh risk against benefit before building or deploying an agent (some tasks suit deterministic workflows better); impact factors are the domain's error tolerance and number of critical processes supported, access to sensitive data (worse with persistent memory), access to external systems, scope of actions (read versus write, few versus many tools) and reversibility; likelihood factors are autonomy, task complexity, exposure to external systems, reliance on third-party agents with limited visibility, and system complexity including multi-agent interaction; use threat modelling (memory poisoning, tool misuse, privilege compromise) and taint tracing of untrusted data, updated continuously.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.