# Implement Zero-Trust Governance for Autonomous AI Agents Using Kong and Mandrel

This presentation explains how the open-source Mandrel project works with Kong to govern autonomous and agent-to-agent AI systems through enforceable contracts instead of ad hoc prompt controls. Paul Williams of Barefoot Coders and Randy Burn of Kong argue that single Uber-style agents with broad tool and data access create a new monolith, while an agent mesh of specialized agents needs platform-level guardrails because governance, not developer adoption, is the main barrier to production use. Mandrel extends the standard agent card with machine-readable policies for risk profiles, human approval thresholds, compounded identity claims, hop and delegation limits, network allow lists, token budgets, and orchestration rules, then compiles those declarations into Kong gateway, mesh, and policy configurations rather than requiring a new runtime. The live demo shows Kong enforcing loop detection, hop exhaustion, and delegation boundaries in a fail-closed model, positioning Mandrel and Kong as the controls that enable faster, safer agent autonomy.