Agentic framework 2.4: for users who interact with agents, focus on transparency: their own responsibilities (for example double-checking outputs, optional personal approval thresholds), declaring the agent at the point of interaction, its authorised range of actions and decisions, how data are collected and used (with explicit consent where needed) and human contact points for escalation; for staff who build agents into their own workflows, add education on suitable and restricted uses, how to instruct agents, what actions they can take, typical failures such as hallucination or loops, refreshers and feedback loops, and protect tradecraft and business continuity by identifying core skills and giving enough training and work exposure that staff can still perform critical processes when agents fail.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.