Main

AI Architecture Reference

Visual overview of enterprise AI platforms and how OAIO pillars map to their capabilities. Use these references to communicate how your organization's agent strategy aligns with cloud provider infrastructure.

Why This Matters

AI agents are not just prompts—they require infrastructure for identity, governance, memory, and observability. Major cloud providers have built platform layers specifically for enterprise agent deployment. OAIO helps you prepare your organization to leverage these capabilities effectively, regardless of which platform you choose.

Platform Architectures

AWS

Amazon Bedrock AgentCore

Enterprise-grade infrastructure for deploying AI agents at scale

Your Agent Framework
LangGraphLangChainCrewAIStrandsCustom

AgentCore Runtime

Serverless agent hosting with session isolation

P4Experience

AgentCore Memory

Persistent context across interactions

P2Data

AgentCore Gateway

MCP-compatible tool orchestration

P4Experience

AgentCore Identity

Secure access to AWS & third-party services

P3Trust

AgentCore Policy

Real-time governance & action boundaries

P3Trust

AgentCore Observability

Execution tracing & debugging

P5FinOps

AgentCore Evaluations

Continuous quality scoring in production

P1Value
P5FinOps
Foundation Models (Bedrock + External)
NovaClaudeLlamaMistralCohereStability
Microsoft

Microsoft Copilot + Agent 365

AI agents integrated with Microsoft 365 ecosystem

Microsoft 365 Apps (Copilot-Enabled)
WordExcelPowerPointOutlookTeamsSharePoint

Microsoft Graph

Unified access to organizational data with RBAC

P2Data

Grounding Service

Context enrichment from user's data

P2Data
P4Experience

Copilot Studio

Low-code agent builder with multi-agent orchestration

P4Experience

Agent 365

Control plane for agent governance & management

P3Trust

Microsoft Purview

Compliance, data governance & sensitivity labels

P3Trust

Conditional Access

Identity-based security policies & MFA

P3Trust

Copilot Analytics

Usage metrics, adoption tracking & ROI

P1Value
P5FinOps
Microsoft 365 Service Boundary
All data stays within tenant boundary with existing RBAC enforced
OpenAIAnthropic

Bespoke Agent Stack

Custom-built infrastructure using native frontier model APIs

Your Infrastructure (You Build & Manage)
LangGraphLangChainCrewAIAutoGenCustom

OpenAI Assistants API

Thread-based conversations with tool use & code interpreter

P4Experience

Claude MCP

Model Context Protocol for tool & resource access

P4Experience

Vector Databases

Pinecone, Weaviate, Chroma for semantic retrieval

P2Data

Custom Auth Layer

OAuth/OIDC, API keys, role-based access

P3Trust

Guardrails & Filters

Custom prompt injection, PII, toxicity detection

P3Trust

Observability Stack

LangSmith, Helicone, custom logging pipelines

P5FinOps

Evaluation Frameworks

Custom eval suites, A/B testing, quality scoring

P1Value
P5FinOps
Direct API Access to Frontier Models
GPT-4oo1ClaudeGeminiLlamaMixtral
Higher Ownership Responsibility
You build and maintain all governance, security, and observability components
Google Cloud

Google Cloud Vertex AI

Enterprise AI platform with Gemini and multi-model support

Google Workspace Integration
GmailDocsSheetsDriveMeetChat

Vertex AI Agent Builder

No-code/low-code agent development platform

P4Experience

Grounding with Search

Real-time web & enterprise data grounding

P2Data
P4Experience

Vertex AI Search

Enterprise search across structured & unstructured data

P2Data

IAM & Security

Google Cloud identity, VPC-SC, data residency

P3Trust

Model Garden

Gemini, Claude, Llama + 150+ models via unified API

P4Experience

Evaluation Service

Automated quality metrics & human feedback

P1Value
P5FinOps

Cloud Monitoring

Metrics, logging, tracing for AI workloads

P5FinOps
Foundation Models (Model Garden)
Gemini 2.0ClaudeLlamaMistralImagenGemma

OAIO Pillar Mapping

The five pillars of OAIO are platform-agnostic principles. Here's how they map to specific capabilities in each platform.

OAIO Pillars to Platform Mapping

How each pillar maps to capabilities across all platform options

1Value & Adoption
AWS
AgentCore Evaluations for goal success rates
Microsoft
Copilot Analytics for adoption tracking
Bespoke
Custom eval suites, A/B testing frameworks
Google
Vertex AI Evaluation Service
2Data Readiness
AWS
AgentCore Memory + Gateway for data access
Microsoft
Microsoft Graph + Grounding Service
Bespoke
Vector DBs (Pinecone, Weaviate, Chroma)
Google
Vertex AI Search + Grounding
3AI Protection & Trust
AWS
AgentCore Identity + Policy for governance
Microsoft
Agent 365 + Purview + Conditional Access
Bespoke
Custom auth, guardrails, content filters
Google
IAM + VPC-SC + data residency controls
4Experience Design
AWS
AgentCore Runtime + multi-framework support
Microsoft
Copilot Studio + M365 app integration
Bespoke
OpenAI Assistants + Claude MCP + frameworks
Google
Agent Builder + Workspace integration
5FinOps & Economics
AWS
AgentCore Observability for cost tracking
Microsoft
Copilot Analytics + per-user licensing
Bespoke
LangSmith, Helicone, custom pipelines
Google
Cloud Monitoring + usage tracking

Platform Selection Considerations

AWS Bedrock AgentCore

  • Best for: Custom agent frameworks (LangGraph, CrewAI, Strands)
  • Strength: Infrastructure-level control and multi-model flexibility
  • Models: Nova, Claude, Llama, Mistral + external via Gateway
  • Ideal when: Building custom agents with specific tooling requirements

Microsoft Copilot + Agent 365

  • Best for: M365-centric organizations with existing investments
  • Strength: Native integration with Office apps and Microsoft Graph
  • Data access: Automatic grounding in organizational data
  • Ideal when: Extending Copilot with custom agents for business users

Bespoke Agent Stack

  • Best for: Maximum control and frontier model access
  • Strength: Direct API access to OpenAI, Anthropic, and others
  • Trade-off: You build and maintain all governance infrastructure
  • Ideal when: Specialized requirements that CSP platforms don't cover

Google Cloud Vertex AI

  • Best for: Workspace-centric orgs and multimodal use cases
  • Strength: Gemini models + 200+ models via Model Garden
  • Data access: Native Workspace integration and enterprise search
  • Ideal when: Leveraging Google's AI research and multimodal capabilities

Platform-Agnostic Readiness

OAIO prepares your organization to leverage any AI agent platform. The five pillars ensure you have the value cases, data foundations, governance models, and economic controls in place—regardless of which technology stack you deploy.

Remember: OAIO creates the blueprint—your implementation partner builds the agents.