Why sandboxing your agent is not enough

2026-07-07 · Source: Cloud Native Computing Foundation · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Robotics & Autonomous Systems · Depth: Intermediate, short

Summary

The agent-sandbox and agent-substrate projects address distinct but related challenges in deploying AI agents securely and efficiently within Kubernetes environments. Agent-sandbox, a Kubernetes SIG Apps project, focuses on providing secure, manageable, and Kubernetes-native agent execution through strong identities, persistent storage, and isolation via a Sandbox CRD and controller. Conversely, agent-substrate, a standalone project, aims for higher scale, better resource efficiency, lower latency, and dynamic lifecycle management by enabling agents to run as on-demand serverless workloads. This model allows agents to execute in secure worker pods for short bursts, suspend when idle, and resume on any available worker, decoupling the worker pod lifecycle from the agent "actor." The author argues that while agent-sandbox is essential for security, agent-substrate is crucial for operational scalability, making large-scale agent fleets economically viable by allowing many agents to share worker resources, exemplified by six AIRE agents using a single worker pod.

Key takeaway

For AI Architects designing agentic AI systems on Kubernetes, relying solely on sandboxing solutions like agent-sandbox will lead to resource inefficiencies at scale. You should integrate dynamic, on-demand agent runtimes such as agent-substrate to achieve operational scalability and economic viability. This approach allows hundreds of agents to share worker resources, reducing idle consumption and enabling ephemeral, event-driven execution without dedicated pods.

Key insights

Sandboxing AI agents is necessary for security, but insufficient for scalable, efficient, and economically viable large-scale deployments.

Principles

• Agent security requires strong identities and isolation.
• Decouple agent lifecycle from Kubernetes pods.
• On-demand execution improves resource efficiency.

Method

Agent-substrate enables agents to execute in secure worker pods for short bursts, suspend when idle, and resume on any available worker, managed by a control plane.

In practice

• Use shared worker pools for multiple agents.
• Dynamically execute agents based on invocation.
• Scale worker pools horizontally for concurrency.

Topics

• AI Agents
• Kubernetes
• Agent-sandbox
• Agent-substrate
• Serverless Workloads
• Resource Efficiency
• Cloud Native Security

Code references

• kubernetes-sigs/agent-sandbox
• agent-substrate/substrate
• kubernetes/community

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, MLOps Engineer, AI Architect

Related on AIssential

• The Future of AI Agents are Sandboxes — AI agent sandboxes provide secure, isolated, and configurable environments, expanding agent capabilities while mitigating security risks.
• Build and Run Your Own AI Agent in the Cloud — AgentCore provides managed AWS services for deploying and operating framework-agnostic AI agents, complementing agent frameworks like Strands.
• The Invisible Crisis in AI Engineering: Autonomous Agents and Smart Routing Architectures — Autonomous AI agents face an "invisible crisis" of spiraling token costs, necessitating Smart Routing for economic viability.
• Build AI Coding Agents that Run ∞ in the Cloud — New tools empower AI coding agents with enhanced web access, workflow adaptability, and cloud-native execution.
• OpenAI acquires AI agent orchestration startup Ona — Cloud-based sandboxes enable persistent, secure, and resource-rich execution for long-running AI agents, overcoming local machine limitations.
• The Sequence Opinion #864: Every AI Agent Needs a Computer — AI agents require architectural access to a full computer environment for true operational capability, moving beyond token emission.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Cloud Native Computing Foundation.