--- title: Architectures for Agent Systems: A Survey of Isolation,... description: A comprehensive survey of enterprise LLM agent system architectures covering three core pillars: sandboxed execution environments, tool integration standards,... canonical: https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Architectures for Agent Systems: A Survey of Isolation, Integration, and Governance | daily.dev og:description: A comprehensive survey of enterprise LLM agent system architectures covering three core pillars: sandboxed execution environments, tool integration standards,... og:url: https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c og:image: https://api.daily.dev/og/posts/DHdWuuU4c.png og:image:alt: Architectures for Agent Systems: A Survey of Isolation, Integration, and Governance og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Architectures for Agent Systems: A Survey of Isolation, Integration, and Governance **[eunomia eBPF Blog](https://daily.dev/sources/eunomia)** · 46 min read · 0 upvotes · 0 comments ## Summary A comprehensive survey of enterprise LLM agent system architectures covering three core pillars: sandboxed execution environments, tool integration standards, and governance frameworks. Key developments include the Kubernetes Agent Sandbox (gVisor/Kata Containers-based microVMs with sub-second startup via pre-warmed pools), Model Context Protocol (MCP) as the emerging universal standard for agent-tool integration now under Linux Foundation's AAIF governance with 10,000+ published servers, and security challenges including prompt injection, tool supply chain vulnerabilities, and the MCP-SafetyBench finding that even top models have 30–48% of tasks compromised under attack. Open challenges include rollback of external side effects, unified cross-layer security models, and human-agent interaction paradigms. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What startup latency improvement does the Kubernetes Agent Sandbox achieve with pre-warmed sandbox pools? The Kubernetes Agent Sandbox achieves sub-second startup latency using pre-warmed sandbox pools, representing roughly a 90% improvement over cold-starting sandboxes. Google demonstrated this at KubeCon NA 2025, where the GKE Agent Sandbox pairs with a Pod Snapshots feature that checkpoints and restores running sandbox pods — including GPU workloads — cutting startup from minutes to seconds while avoiding idle resource waste. _Teams shipping agent infrastructure track sandbox performance benchmarks and Kubernetes primitives like these on daily.dev._ ### How many MCP servers have been published and which AI platforms support MCP? Over 10,000 MCP servers have been published covering dev tools to Fortune 500 internal integrations. Support is built into Claude, ChatGPT, GitHub Copilot, Google Gemini, VS Code, and Cursor, among others. MCP was open-sourced by Anthropic in late 2024 and became a founding contribution to the Linux Foundation's Agentic AI Foundation (AAIF) in December 2025, with AWS, Google, Microsoft, Anthropic, and OpenAI as platinum members. _Developers choosing between agent tool integration approaches follow MCP ecosystem developments on daily.dev._ ### What does MCP-SafetyBench reveal about LLM agent vulnerability to attacks? MCP-SafetyBench tests agents on multi-step tasks across five domains — web browsing, financial analysis, code repo management, navigation, and web search — while injecting 20 attack types. No current model is immune: even top-tier models had 30–48% of tasks compromised. A negative correlation was found between task performance and security — more capable models tend to be more exploitable because they more eagerly follow any instruction, including malicious ones. _Security engineers hardening agent deployments stay ahead of new attack taxonomies and benchmark results on daily.dev._ ## Similar posts on daily.dev - [7 Tools for Multi-Agent Infrastructure](https://daily.dev/posts/7-tools-for-multi-agent-infrastructure-nh43zuwkr) · Nordic APIs · 0 upvotes · 0 comments - [Agentic AI Security: What CISOs Say About Governing AI Agents](https://daily.dev/posts/agentic-ai-security-what-cisos-say-about-governing-ai-agents-peda204z1) · Docker · 0 upvotes · 0 comments - [The Architectural Shift: AI Agents Become Execution Engines While Backends Retreat to Governance](https://daily.dev/posts/the-architectural-shift-ai-agents-become-execution-engines-while-backends-retreat-to-governance-g6airkseg) · InfoQ · 2 upvotes · 0 comments - [The Current State of Agentic AI](https://daily.dev/posts/the-current-state-of-agentic-ai-yp4wym5dv) · Machine Learning Mastery · 3 upvotes · 0 comments --- Tags: [#kubernetes](https://daily.dev/tags/kubernetes), [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#ai-security](https://daily.dev/tags/ai-security) [View this post on daily.dev](https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Architectures for Agent Systems: A Survey of Isolation, Integration, and Governance","url":"https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c"},"datePublished":"2026-08-11T08:54:10.998Z","dateModified":"2026-08-11T09:12:50.546Z","description":"A comprehensive survey of enterprise LLM agent system architectures covering three core pillars: sandboxed execution environments, tool integration standards,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b59e04eefe0ea85df0b7e50151b0b12e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b59e04eefe0ea85df0b7e50151b0b12e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"eunomia eBPF Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"eunomia eBPF Blog","logo":"https://media.daily.dev/image/upload/s--z-nTl7TK--/f_auto,q_auto/v1786438304/logos/eunomia?_a=BAMAMicg0","url":"https://daily.dev/sources/eunomia"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"kubernetes,ai-agents,mcp,ai-security","timeRequired":"PT46M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"eunomia eBPF Blog","item":"https://daily.dev/sources/eunomia"},{"@type":"ListItem","position":3,"name":"Architectures for Agent Systems: A Survey of Isolation, Integration, and Governance"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/architectures-for-agent-systems-a-survey-of-isolation-integration-and-governance-dhdwuuu4c#faq","mainEntity":[{"@type":"Question","name":"What startup latency improvement does the Kubernetes Agent Sandbox achieve with pre-warmed sandbox pools?","acceptedAnswer":{"@type":"Answer","text":"The Kubernetes Agent Sandbox achieves sub-second startup latency using pre-warmed sandbox pools, representing roughly a 90% improvement over cold-starting sandboxes. Google demonstrated this at KubeCon NA 2025, where the GKE Agent Sandbox pairs with a Pod Snapshots feature that checkpoints and restores running sandbox pods — including GPU workloads — cutting startup from minutes to seconds while avoiding idle resource waste. Teams shipping agent infrastructure track sandbox performance benchmarks and Kubernetes primitives like these on daily.dev."}},{"@type":"Question","name":"How many MCP servers have been published and which AI platforms support MCP?","acceptedAnswer":{"@type":"Answer","text":"Over 10,000 MCP servers have been published covering dev tools to Fortune 500 internal integrations. Support is built into Claude, ChatGPT, GitHub Copilot, Google Gemini, VS Code, and Cursor, among others. MCP was open-sourced by Anthropic in late 2024 and became a founding contribution to the Linux Foundation's Agentic AI Foundation (AAIF) in December 2025, with AWS, Google, Microsoft, Anthropic, and OpenAI as platinum members. Developers choosing between agent tool integration approaches follow MCP ecosystem developments on daily.dev."}},{"@type":"Question","name":"What does MCP-SafetyBench reveal about LLM agent vulnerability to attacks?","acceptedAnswer":{"@type":"Answer","text":"MCP-SafetyBench tests agents on multi-step tasks across five domains — web browsing, financial analysis, code repo management, navigation, and web search — while injecting 20 attack types. No current model is immune: even top-tier models had 30–48% of tasks compromised. A negative correlation was found between task performance and security — more capable models tend to be more exploitable because they more eagerly follow any instruction, including malicious ones. Security engineers hardening agent deployments stay ahead of new attack taxonomies and benchmark results on daily.dev."}}]} ```