Aug 31, 2026 · 19 min
Enterprises need a control plane before AI can scale
EP 61: "Shadow AI" — And No One's Watching | Rafi Khardalian, Actualyze AI
The episode argues that secure, reliable AI adoption depends less on access to models than on governance and organizational readiness.
- 1An orchestration layer can route models, providers, agents, and tools through one governed inference endpoint.
- 2Enterprise AI projects often miss their returns because cultural change and human expertise lag behind technical investment.
- 3Smaller and open-weight models may handle more enterprise workloads as privacy, cost, and deployment controls improve.
Don't miss
Khardalian explains how an inference endpoint can inspect traffic and mask PII before sensitive enterprise data reaches model providers.
The brief
Rafi Khardalian of Actualize AI argues that enterprises need an infrastructure-level control plane to coordinate models, providers, agents, and tools through a single inference endpoint.
The proposed layer turns model selection into a governance problem as well as a technical one, covering approval, procurement, data handling, reliability, and cost.
For regulated industries, the inference endpoint can inspect traffic and mask personally identifiable information before sensitive data reaches external model providers.
The conversation challenges the assumption that bigger AI investments guarantee returns, pointing instead to cultural transformation, realistic expectations, and continued human expertise.
Khardalian sees smaller and open-weight models gaining ground, while organizational readiness and safe deployment—not raw model availability—remain the central barriers.
Featuring
Rafi Khardalian
Cisco Systems, Inc.Listen to the full episode and explore every guest, topic, and moment on PodLume.