{
  "schemaVersion": 3,
  "id": "article:modern-ai-stack",
  "slug": "modern-ai-stack",
  "title": "The Modern AI Stack: Demystifying the Architecture of Artificial Intelligence",
  "canonicalPath": "/articles/modern-ai-stack/",
  "sourcePath": "content/articles/2026/modern-ai-stack/article.md",
  "agentBriefPath": "content/articles/2026/modern-ai-stack/agent.md",
  "thesis": "Artificial intelligence is not a single monolithic product but a modular stack of distinct engineering layers; understanding the layers helps builders and observers place tools, companies, and research advances in context.",
  "status": "published",
  "maturity": "seed",
  "publishedAt": "2026-07-07",
  "updatedAt": "2026-07-18",
  "audiences": [
    "general",
    "students",
    "builders"
  ],
  "topics": [
    "ai-stack",
    "agent-architecture",
    "mcp",
    "rag",
    "inference",
    "model-alignment",
    "tooling"
  ],
  "claims": [
    {
      "id": "claim-001",
      "claim": "Modern AI systems are best understood as a modular stack rather than a single monolithic model.",
      "confidence": "medium",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-hermes-agent",
          "snippet": "Hermes Agent is a self-improving AI agent runtime built by Nous Research that spans orchestration, memory, and end-user interfaces, demonstrating how a single product crosses multiple layers.",
          "supports": "direct",
          "assessedAt": "2026-07-07"
        },
        {
          "sourceId": "source-claude-desktop",
          "snippet": "Claude Desktop bundles model alignment, inference, simple orchestration, and specific tools into a single end-user application.",
          "supports": "direct",
          "assessedAt": "2026-07-07"
        }
      ],
      "counterevidence": [
        {
          "summary": "The stack framing is the author's proposed taxonomy; other analysts use different layer counts or group the same capabilities differently.",
          "assessedAt": "2026-07-07"
        }
      ],
      "verification": {
        "status": "verified",
        "reviewedAt": "2026-07-18",
        "reviewer": "kimi-code-cli",
        "note": "Verified 2026-07-18 (meta#61 backlog burn-down): evidence packets checked against cited sources; spot-checked live."
      }
    },
    {
      "id": "claim-002",
      "claim": "Retrieval-augmented generation is a common pattern for grounding models in proprietary or real-time data.",
      "confidence": "high",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-pinecone",
          "snippet": "Vector databases such as Pinecone store embeddings so models can retrieve relevant private or fresh documents at query time.",
          "supports": "direct",
          "assessedAt": "2026-07-07"
        },
        {
          "sourceId": "source-vllm",
          "snippet": "RAG has become a standard architectural pattern for connecting LLMs to external knowledge bases and enterprise data.",
          "supports": "indirect",
          "assessedAt": "2026-07-07"
        }
      ],
      "counterevidence": [
        {
          "summary": "RAG is not the only grounding approach; fine-tuning, knowledge graphs, and long-context windows are alternatives for some use cases.",
          "assessedAt": "2026-07-07"
        }
      ],
      "verification": {
        "status": "verified",
        "reviewedAt": "2026-07-18",
        "reviewer": "kimi-code-cli",
        "note": "Verified 2026-07-18 (meta#61 backlog burn-down): evidence packets checked against cited sources; spot-checked live."
      }
    },
    {
      "id": "claim-003",
      "claim": "Model Context Protocol (MCP) is emerging as a common adapter for agent-tool integration.",
      "confidence": "medium",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-anthropic-mcp",
          "snippet": "Anthropic released the Model Context Protocol in late 2024 as an open standard for connecting AI assistants to data sources and tools.",
          "supports": "direct",
          "assessedAt": "2026-07-07"
        }
      ],
      "counterevidence": [
        {
          "summary": "MCP is still early; other tool-integration approaches (direct API wrappers, custom SDKs, OpenAPI specs) remain common in production.",
          "assessedAt": "2026-07-07"
        }
      ],
      "verification": {
        "status": "verified",
        "reviewedAt": "2026-07-18",
        "reviewer": "kimi-code-cli",
        "note": "Verified 2026-07-18 (meta#61 backlog burn-down): evidence packets checked against cited sources; spot-checked live."
      }
    },
    {
      "id": "claim-004",
      "claim": "End-user applications such as Claude Desktop bundle multiple lower layers into a single GUI.",
      "confidence": "high",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-claude-desktop",
          "snippet": "Claude Desktop is a downloadable application that packages Anthropic's aligned model, inference, simple orchestration, and file-reading tools behind a graphical interface.",
          "supports": "direct",
          "assessedAt": "2026-07-07"
        }
      ],
      "counterevidence": [
        {
          "summary": "Some end-user applications expose model or tool choices (e.g., via settings or plugins), so the bundling is not always absolute.",
          "assessedAt": "2026-07-07"
        }
      ],
      "verification": {
        "status": "verified",
        "reviewedAt": "2026-07-18",
        "reviewer": "kimi-code-cli",
        "note": "Verified 2026-07-18 (meta#61 backlog burn-down): evidence packets checked against cited sources; spot-checked live."
      }
    },
    {
      "id": "claim-005",
      "claim": "Base-model scaling remains important, but much of the visible product engineering and investment has moved up the stack toward orchestration and tool protocols.",
      "confidence": "medium",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-anthropic-mcp",
          "snippet": "The release and rapid adoption of MCP signals industry attention shifting to standardized tool protocols.",
          "supports": "indirect",
          "assessedAt": "2026-07-07"
        },
        {
          "sourceId": "source-hermes-agent",
          "snippet": "Hermes Agent's growth as a self-improving agent runtime illustrates engineering focus moving toward orchestration and persistent memory rather than base-model training.",
          "supports": "indirect",
          "assessedAt": "2026-07-07"
        }
      ],
      "counterevidence": [
        {
          "summary": "Frontier labs continue to invest heavily in base-model research and scaling; the observation is about product-engineering focus, not total research investment.",
          "assessedAt": "2026-07-07"
        }
      ],
      "verification": {
        "status": "verified",
        "reviewedAt": "2026-07-18",
        "reviewer": "kimi-code-cli",
        "note": "Verified 2026-07-18 (meta#61 backlog burn-down): evidence packets checked against cited sources; spot-checked live."
      }
    }
  ],
  "sources": [
    {
      "id": "source-hermes-agent",
      "title": "Nous Research Hermes Agent on GitHub",
      "url": "https://github.com/nousresearch/hermes-agent",
      "type": "product",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-openrouter",
      "title": "OpenRouter",
      "url": "https://openrouter.ai/",
      "type": "product",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-anthropic-mcp",
      "title": "Anthropic: Model Context Protocol",
      "url": "https://www.anthropic.com/news/model-context-protocol",
      "type": "product",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-meta-llama-3",
      "title": "Meta: Llama 3",
      "url": "https://ai.meta.com/blog/meta-llama-3/",
      "type": "product",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-vllm",
      "title": "vLLM documentation",
      "url": "https://docs.vllm.ai/",
      "type": "documentation",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-langgraph",
      "title": "LangGraph",
      "url": "https://langchain-ai.github.io/langgraph/",
      "type": "documentation",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-crewai",
      "title": "CrewAI",
      "url": "https://crewai.com/",
      "type": "product",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-claude-desktop",
      "title": "Claude desktop app",
      "url": "https://claude.ai/download",
      "type": "product",
      "accessed": "2026-07-07"
    },
    {
      "id": "source-pinecone",
      "title": "Pinecone vector database",
      "url": "https://www.pinecone.io/",
      "type": "product",
      "accessed": "2026-07-07"
    }
  ],
  "related": [
    {
      "type": "article",
      "id": "article:agent-control-planes"
    },
    {
      "type": "article",
      "id": "article:ai-delegation-orchestration"
    }
  ],
  "agentInstructions": [
    "Use claim IDs as retrieval units and source IDs for evidence tracing.",
    "Treat the nine-layer model as a proposed taxonomy, not an industry consensus.",
    "When citing product examples, note that market positioning can change quickly."
  ],
  "provenance": {
    "createdAt": "2026-07-07",
    "createdBy": "human",
    "agents": [],
    "reviews": [
      {
        "reviewer": "sibling-agent",
        "reviewedAt": "2026-07-07",
        "status": "approved",
        "scope": [
          "thesis",
          "claims",
          "sources",
          "privacy",
          "anchor-examples"
        ],
        "notes": "Two-pass independent sibling-agent review under SDL governance; first review identified blockers (pi.dev misattribution, unsupported strong claims, missing taxonomy framing, missing citations) that were resolved.",
        "contentHash": "8f9b94c0a22be0dad275ff1dcb7c07f23043764896bdc23fe7293fd499c7974a"
      },
      {
        "reviewer": "human",
        "reviewedAt": "2026-07-07",
        "status": "approved",
        "scope": [
          "thesis",
          "claims",
          "tone",
          "privacy",
          "sources",
          "publication"
        ],
        "notes": "Human author approved publication after sibling-agent review passed and fixes were applied.",
        "contentHash": "8f9b94c0a22be0dad275ff1dcb7c07f23043764896bdc23fe7293fd499c7974a"
      }
    ],
    "policy": {
      "id": "policy:default",
      "version": "1.0.0"
    }
  },
  "contentHash": "8f9b94c0a22be0dad275ff1dcb7c07f23043764896bdc23fe7293fd499c7974a",
  "diagnostics": {
    "accepted": [
      {
        "rule": "low-source-diversity",
        "scope": "article",
        "rationale": "Documented in docs/diagnostics-triage-2026-07-15.md",
        "documentedAt": "2026-07-15"
      },
      {
        "rule": "duplicate-evidence-source",
        "scope": "article",
        "rationale": "Documented in docs/diagnostics-triage-2026-07-15.md",
        "documentedAt": "2026-07-15"
      }
    ]
  },
  "generatedAt": "2026-07-18T00:00:00.000Z",
  "articleUrl": "https://aura-knowledge.github.io/articles/modern-ai-stack/",
  "agentJsonPath": "/agents/articles/modern-ai-stack.json",
  "agentMarkdownPath": "/agents/articles/modern-ai-stack.md",
  "sourceRepoPath": "content/articles/2026/modern-ai-stack/article.md",
  "sourceGitHubUrl": "https://github.com/aura-knowledge/aura-knowledge.github.io/blob/main/content/articles/2026/modern-ai-stack/article.md",
  "tokenEstimate": 430,
  "sectionOutline": []
}