TLDR Tech / DevOps
Cloud Native for Agentic AI: Lessons from Building a Multi-Agent Security Platform on Kubernetes
Orange Innovation developed a real-time security operations platform leveraging CNCF projects such as Falco, Cilium, and Kafka, integrated with AI agents coordinated via the A2A protocol. Each agent is deployed as a distinct Kubernetes workload with its own identity and resource limits, enabling threat detection and response in regulated production environments. The system employs a classical machine learning model to pre-filter events before they reach LLM-driven agents, with deterministic policy controls through OPA and Kyverno to ensure human analysts review high-risk decisions via Mattermost, rather than solely relying on AI prompt engineering for safety.
Read the full story