Deep Dive: Microsoft Agent Factory — Work IQ, Fabric IQ & Foundry IQ

Topic: Microsoft Agent Factory 䞭的䞉倧智胜层Work IQ、Fabric IQ、Foundry IQ
Category: Cloud AI / Enterprise AI Platform
Level: 䞭级
Last Updated: 2026-03-10


📌 䞭文版


1. 抂述 (Overview)

圚 AI Agent智胜䜓时代埮蜯构建了䞀䞪宏倧的愿景让 AI 䞍仅仅是”聊倩助手”而是胜借理解䜠的工䜜、理解䜠的数据、理解䜠的䞚务的真正”数字同事”。这䞪愿景的栞心就是 Microsoft Agent Factory——䞀䞪让䌁䞚胜借倧规暡构建、郚眲和管理 AI Agent 的平台䜓系。

䜆 Agent 芁做到”真正有甚”䞍胜只靠倧语蚀暡型LLM本身的通甚知识。它们需芁组织内郚的䞊䞋文——䜠的邮件、䌚议、文档、䞚务数据、数据仓库等等。这就是埮蜯掚出的 䞉倧 IQ 智胜层 芁解决的问题

IQ 层 䞀句话诎明 数据来源
Work IQ 理解䜠怎么工䜜的 Microsoft 365邮件、䌚议、聊倩、文档
Fabric IQ 理解䜠䞚务数据的含义 Microsoft FabricOneLake、Power BI、数据仓库
Foundry IQ 连接䜠所有䌁䞚知识的 Azure Blob、SharePoint、OneLake、公共眑络等

甚䞀䞪比喻来理解

劂果把 AI Agent 比䜜䞀䞪新入职的员工那么 Work IQ 让它知道”公叞里倧家怎么协䜜、最近圚讚论什么”Fabric IQ 让它看懂”公叞的数据报衚和䞚务指标意味着什么”Foundry IQ 让它胜”翻阅公叞的知识库和各种文档资料”。

䞉者各有䟧重䜆可以组合䜿甚共同䞺 Agent 提䟛党面的组织䞊䞋文。


2. 栞心抂念 (Core Concepts)

2.1 什么是 Agent Factory

Agent Factory 是埮蜯对其 AI Agent 构建平台的统称。栞心理念是让构建 Agent 像工厂流氎线䞀样标准化、规暡化、可治理。

Agent Factory 涵盖了倚䞪平台和工具

  • Microsoft Copilot Studio䜎代码/无代码的 Agent 构建工具
  • Microsoft Foundry原 Azure AI Foundry面向匀发者的 Agent 匀发平台
  • Agent 365Agent 的统䞀管理控制面板泚册、安党、治理
  • 䞉倧 IQ 层䞺 Agent 提䟛智胜䞊䞋文Work IQ、Fabric IQ、Foundry IQ
┌─────────────────────────────────────────────────────┐
│              Microsoft Agent Factory                 │
│  ┌───────────────┐  ┌───────────────┐  ┌──────────┐ │
│  │ Copilot Studio│  │  Microsoft    │  │ Agent 365│ │
│  │ (䜎代码构建)  │  │  Foundry      │  │ (治理)   │ │
│  │               │  │ (Pro-code匀发)│  │          │ │
│  └───────┬───────┘  └───────┬───────┘  └────┬─────┘ │
│          │                  │               │        │
│          └──────────┬───────┘               │        │
│                     â–Œ                       │        │
│  ┌─────────────────────────────────────────────────┐ │
│  │           䞉倧 IQ 智胜层Agent 的倧脑         │ │
│  │  ┌───────────┐ ┌───────────┐ ┌────────────────┐ │ │
│  │  │  Work IQ  │ │ Fabric IQ │ │  Foundry IQ    │ │ │
│  │  │ 工䜜䞊䞋文│ │ 䞚务语义  │ │ 䌁䞚知识检玢   │ │ │
│  │  └───────────┘ └───────────┘ └────────────────┘ │ │
│  └─────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────┘

2.2 䞺什么需芁䞉䞪 IQ

䞀䞪关键问题是倧语蚀暡型本身并䞍知道䜠公叞的任䜕事情。它只有公共互联眑的训练知识。芁让 Agent 圚䌁䞚内真正有甚就需芁把䌁䞚的”内郚知识”安党地泚入到 Agent 的掚理过皋䞭。

䜆”䌁䞚知识”本身就分䞺䞉倧类

  1. 工䜜协䜜数据谁和谁匀了什么䌚、讚论了什么、邮件诎了什么→ Work IQ
  2. 结构化䞚务数据销售数据、库存数据、客户数据以及这些数据的”䞚务含义”→ Fabric IQ
  3. 广泛的非结构化知识技术文档、产品手册、知识库文章、倖郚眑页→ Foundry IQ

每䞀类数据的存傚䜍眮、栌匏、访问控制方匏郜完党䞍同所以需芁䞉䞪䞓闚的智胜层来倄理。


3. 工䜜原理 (How It Works)

3.1 Work IQ — “理解䜠怎么工䜜”

Work IQ 是 Microsoft 365 Copilot 背后的智胜匕擎。它是让 Copilot 胜借理解”䜠的工䜜䞊䞋文”的关键。

䞉层架构

Work IQ 由䞉䞪玧密集成的层组成

┌─────────────────────────────────────┐
│           Skills & Tools            │  ← 执行层具䜓技胜和工具
│  (调床䌚议、检玢文档、发邮件等)      │
├──────────────────────────────────────
│           Context Layer             │  ← 䞊䞋文层记忆䞎语义理解
│  (Memory、Semantic Index、          │
│   Business Understanding)           │
├──────────────────────────────────────
│           Data Layer                │  ← 数据层原始工䜜数据
│  (M365 Tenant Data、Copilot        │
│   Connectors、Dynamics 365/        │
│   Dataverse)                        │
└─────────────────────────────────────┘
数据层 (Data Layer)
  • Microsoft 365 租户数据以 Microsoft Graph 䞺栞心包含
    • SharePoint/OneDrive 䞭的文件Word、Excel、PPT 等
    • Outlook 邮件
    • Teams 䌚议和聊倩记圕
    • 甚户和组织结构信息
    • 协䜜暡匏元数据谁和谁经垞合䜜、沟通频率等
  • Copilot Connectors连接非埮蜯系统的数据劂 Salesforce、SAP 等预眮数癟䞪连接噚也支持自定义。

  • Dynamics 365 / Power Apps 数据Dataverse将 CRM、ERP 等䞚务系统数据纳入让 Copilot 胜同时掚理”生产力数据”和”䞚务数据”。䟋劂询问”䞊呚我和䟛应商的 Teams 通话䞭提到的零件问题䌚怎样圱响我的库存和销售”
䞊䞋文层 (Context Layer)
  • Memory记忆
    • 星匏记忆甚户䞻劚告诉 Copilot 的偏奜劂”我喜欢甚䞻劚语态”
    • 隐匏记忆Copilot 从聊倩历史䞭掚断出的持久掞察
    • 未来将融入曎倚掻劚暡匏来自 Teams、Outlook、Word 等的䜿甚暡匏
  • Semantic Index语义玢匕䞍仅做关键词匹配还做语义级别的检玢——理解”意思”而非仅仅”字面”。

  • Business Understanding䞚务理解通过 Ontology本䜓和 Glossary术语衚捕获䞚务工䜜流的过皋性知识让 Copilot 胜像”䞚务䞓家”䞀样理解䜠的任务。
技胜䞎工具层 (Skills & Tools Layer)
  • Skills技胜䞓闚的指什集告诉 Copilot “劂䜕做某件事”劂”劂䜕调床䌚议”、”劂䜕从倖郚源获取数据”。
  • Tools工具实际执行技胜的工具——MCP Server 工具、Agent Flows、API、Plugin 等。
安党䞎隐私

Work IQ 始终尊重已有的 Microsoft 365 权限暡型

  • 甚户权限只胜访问䜠有权限看的数据
  • 安党组分配
  • 敏感床标筟 (Sensitivity Labels)
  • DLP 策略
  • 笊合 GDPR 和 EU Data Boundary
Work IQ API

匀发者可以通过 Work IQ APIRESTful 接口将 Work IQ 的智胜胜力集成到自己的应甚和 Agent 䞭。也有 CLI 工具和 MCP Server 暡匏可甚。


3.2 Fabric IQ — “理解䜠的䞚务数据意味着什么”

Fabric IQ 是 Microsoft Fabric 䞭的语义智胜层。劂果诎 Work IQ 让 Agent 理解”人们怎么工䜜”Fabric IQ 则让 Agent 理解”数据诎的是什么”。

䞺什么需芁 Fabric IQ

想象这样䞀䞪场景䜠的数据仓库里有䞀匠衚叫 tbl_cust_ord_v2里面有字段 qty_shipped、dt_exp、flg_breach。对于 AI Agent 来诎这些列名几乎无法理解。

Fabric IQ 的栞心目标就是䞺数据赋予䞚务含义建立䞀套统䞀的”䞚务语蚀”让 Agent 胜理解

  • qty_shipped = “已发莧数量”
  • dt_exp = “预计亀付日期”
  • flg_breach = “是吊存圚合同违纊”
栞心组件
组件 诎明
Ontology本䜓 栞心定义䌁䞚词汇和语义层——实䜓类型劂 Customer、Order、Shipment、属性、关系和䞚务规则并绑定到真实数据
Graph囟 囟数据库胜力存傚节点和蟹支持路埄查扟和关系遍历劂”订单 → 发莧 → 枩床䌠感噚 → 冷铟违纊”
Data Agent数据 Agent 基于生成匏 AI 的对话匏问答系统连接 Ontology 理解䞚务抂念
Operations Agent运营 Agent 监控实时数据并掚荐䞚务操䜜
Power BI Semantic Model 䞺报衚和亀互匏分析䌘化的语义暡型床量倌、层次结构、关系
Ontology 是关键

Ontology 是 Fabric IQ 的栞心它做的事情类䌌于”䌁䞚数据字兞 + 关系囟谱”

┌──────────────┐    places     ┌──────────────┐
│   Customer   │──────────────▶│    Order     │
│              │               │              │
│ - name       │               │ - order_date │
│ - segment    │               │ - total_amt  │
└──────────────┘               └──────┬───────┘
                                      │ contains
                                      ▌
                               ┌──────────────┐    triggers    ┌──────────────┐
                               │   Shipment   │───────────────▶│  Cold Chain   │
                               │              │                │  Breach      │
                               │ - shipped_qty│                │              │
                               │ - exp_date   │                │ - severity   │
                               └──────────────┘                └──────────────┘

这䞪 Ontology 告诉 Agent

  • “Customer”实䜓有哪些属性数据圚 lakehouse 的哪匠衚
  • “Order”和”Customer”之闎是什么关系
  • “Cold Chain Breach”代衚什么䞥重性意味着什么

有了这些语义信息Agent 就可以回答类䌌这样的问题

“哪些 A 级客户的最近订单出现了冷铟违纊违纊䞥重皋床是高的”

Fabric IQ 的价倌
  • 数据统䞀跚 OneLake 䞭的倚䞪数据源lakehouse、eventhouse、语义暡型建立统䞀视囟
  • 语蚀䞀臎䞀䞪抂念只定义䞀次Power BI、Notebook、Agent 郜䜿甚盞同的语义
  • 治理和信任减少跚团队的定义䞍䞀臎和数据重倍
  • 跚域掚理通过 Graph 遍历关系铟劂”订单 → 发莧 → 䌠感噚 → 违纊”来解释结果
  • AI 就绪䞺 Copilot 和 Agent 提䟛结构化的 grounding让回答基于䌁䞚语蚀和䞚务规则

3.3 Foundry IQ — “连接䌁䞚所有知识”

Foundry IQ 是 Microsoft Foundry原 Azure AI Foundry平台䞭的托管知识层。它解决的是䞀䞪非垞实际的问题䌁䞚的知识散垃圚各倄——Azure Blob Storage、SharePoint 文档库、OneLake、甚至公共眑页——Agent 需芁䞀䞪统䞀的方匏来检玢这些知识。

栞心抂念
┌─────────────────────────────────────────┐
│           Knowledge Base                │  ← 顶层资源猖排检玢行䞺
│  ┌─────────────┐  ┌─────────────┐       │
│  │ Knowledge   │  │ Knowledge   │ ...   │  ← 知识源Azure Blob、
│  │ Source A     │  │ Source B    │       │     SharePoint、OneLake、Web
│  │(SharePoint) │  │(Azure Blob) │       │
│  └─────────────┘  └─────────────┘       │
│                                         │
│  ┌─────────────────────────────────────┐│
│  │      Agentic Retrieval Engine       ││  ← 智胜检玢匕擎
│  │  - 分解倍杂问题䞺子查询             ││
│  │  - 并行执行搜玢                     ││
│  │  - 语义重排序                       ││
│  │  - 聚合统䞀结果                     ││
│  └─────────────────────────────────────┘│
└─────────────────────────────────────────┘
Knowledge Base知识库

知识库是 Foundry IQ 的顶层资源它定义

  • 芁查询哪些知识源
  • 控制检玢行䞺的参数劂检玢掚理力床minimal / low / medium
Knowledge Sources知识源

支持的数据源包括

  • Azure Blob Storage存傚的各种文档
  • SharePoint䌁䞚文档库
  • OneLakeFabric 的统䞀数据湖
  • 公共眑络数据Web 抓取

Foundry IQ 䌚自劚倄理

  • 文档分块 (Chunking)
  • 向量嵌入生成 (Vector Embedding)
  • 元数据提取
  • 增量数据刷新
Agentic Retrieval智胜检玢

这是 Foundry IQ 最栞心的胜力——䞍是简单的关键词搜玢而是

  1. 问题分解将倍杂问题拆分䞺倚䞪子查询
  2. 并行搜玢同时圚倚䞪知识源䞭执行子查询
  3. 语义重排序甚 LLM 对结果进行语义重新排序
  4. 结果聚合将倚源结果统䞀返回
  5. 匕甚远溯返回的数据垊有匕甚标泚可远溯到源文档
权限感知

Foundry IQ 内眮了䞥栌的权限控制

  • 同步 Access Control Lists (ACLs)
  • 尊重 Microsoft Purview 敏感床标筟
  • 圚查询时区制执行权限甚户只胜看到有权限的内容
  • 䜿甚调甚者的 Microsoft Entra 身仜进行端到端权限验证

4. 䞉倧 IQ 对比 (Comparison)

绎床 Work IQ Fabric IQ Foundry IQ
定䜍 工䜜协䜜智胜层 䞚务数据语义层 䌁䞚知识检玢层
栞心问题 Agent 劂䜕理解”人们的工䜜方匏” Agent 劂䜕理解”数据的䞚务含义” Agent 劂䜕查扟”散垃各倄的知识”
所属平台 Microsoft 365 Copilot Microsoft Fabric Microsoft Foundry (Azure)
数据来源 邮件、䌚议、Teams 聊倩、文档、Dynamics 365 OneLakelakehouse、eventhouse、Power BI 语义暡型 Azure Blob、SharePoint、OneLake、公共眑络
栞心胜力 语义玢匕、记忆系统、技胜/工具执行 Ontology 本䜓建暡、Graph 囟遍历、Data Agent 知识库管理、Agentic Retrieval 智胜检玢
兞型问题 “䞊呚和 Alice 的䌚议里讚论了什么” “哪些 A 级客户䞊月的销售额䞋降了” “公叞关于退莧政策的文档是怎么诎的”
权限暡型 M365 权限、敏感床标筟、DLP Fabric 工䜜区权限 Entra ID + ACL 同步 + Purview 标筟
独立还是组合 可独立䜿甚也可䞎其他 IQ 组合 可独立䜿甚也可䞎其他 IQ 组合 可独立䜿甚也可䞎其他 IQ 组合

5. 关键配眮䞎参数 (Key Configurations)

Work IQ

配眮项 诎明 垞见场景
Custom Instructions 甚户级自定义指什劂语气偏奜、栌匏芁求 䞪性化 Copilot 行䞺
Saved Memories 持久化的甚户记忆 让 Copilot 记䜏长期偏奜
Copilot Connectors 非埮蜯数据源连接噚 连接 Salesforce、ServiceNow 等倖郚系统
Sensitivity Labels 敏感床标筟讟眮 控制哪些内容可被 Copilot 倄理
Work IQ API RESTful API 接入 匀发者将 Work IQ 集成到自定义应甚

Fabric IQ

配眮项 诎明 垞见场景
Ontology 定义 实䜓类型、属性、关系、纊束和规则 建立䌁䞚统䞀数据语义
数据源绑定 将 Ontology 实䜓绑定到 lakehouse/eventhouse/语义暡型 连接实际数据
Graph 配眮 节点、蟹、遍历规则 倍杂关系查询和圱响分析
Data Agent 知识源 连接 Ontology 䜜䞺 Agent 数据源 让 Agent 理解䞚务术语

Foundry IQ

配眮项 诎明 垞见场景
Knowledge Base 知识库定义和检玢参数 讟眮检玢范囎和掚理力床
Knowledge Sources 数据源连接Blob、SharePoint、OneLake、Web 连接䌁䞚知识存傚
Retrieval Reasoning Effort 检玢掚理力床minimal/low/medium 平衡检玢莚量和成本
Indexer Schedule 玢匕噚增量刷新计划 保持知识库数据新鲜
ACL 同步 权限控制列衚同步 确保权限感知的检玢

6. 实战经验 (Practical Tips)

最䜳实践

  • 从 Work IQ 匀始劂果䜠的组织已有 Microsoft 365 Copilot 讞可Work IQ 是即匀即甚的。先让员工䜓验 Copilot 的工䜜䞊䞋文胜力。
  • 甹 Fabric IQ 统䞀数据语义圚匕入 AI Agent 之前先花时闎定义奜 Ontology——这是”䞀次定义、到倄䜿甚”的投资。
  • Foundry IQ 适合知识密集型场景劂果䜠的 Agent 需芁查阅倧量文档劂客服知识库、产品手册䌘先配眮 Foundry IQ。
  • 䞉者组合䜿甚效果最䜳让 Agent 同时拥有”工䜜䞊䞋文 + 䞚务语义 + 知识检玢”䞉重胜力。

垞见误区

  • ❌ 讀䞺只芁有 LLM 就借了LLM 没有䜠公叞的任䜕知识IQ 层才是让 Agent 变”聪明”的关键。
  • ❌ 混淆䞉䞪 IQ 的甚途Work IQ ≠ Foundry IQ。前者是理解工䜜暡匏后者是检玢文档知识。
  • ❌ 応略权限配眮IQ 层䌚继承已有的权限暡型䜆劂果䜠的 M365/Fabric/Azure 权限本身就配眮䞍圓Agent 返回的结果也䌚有问题。
  • ❌ 跳过 Ontology 盎接甚 Data Agent没有奜的 Ontology 定义Data Agent 无法理解䜠的䞚务语蚀。

安党泚意

  • 䞉䞪 IQ 层郜䞍存傚甚户的源数据只做按需检玢
  • 所有检玢郜基于调甚者的身仜和权限
  • 支持 Microsoft Purview 敏感床标筟
  • 笊合 GDPR 和欧盟数据蟹界芁求
  • Agent 365 提䟛统䞀的 Agent 安党治理控制面板

7. 䞉者劂䜕协同工䜜 (How They Work Together)

䞀䞪完敎的䌁䞚 Agent 场景可胜同时需芁䞉䞪 IQ 层

甚户提问: "䞊呚和䟛应商 Contoso 的䌚议䞭提到的零件短猺问题
          䌚怎样圱响我们的 Q2 销售预测请参考公叞的䟛应铟应急手册。"

┌──────────────────────────────────────────────────────────┐
│                    AI Agent 的掚理过皋                      │
│                                                          │
│  Step 1: [Work IQ] 查扟䞊呚䞎 Contoso 的 Teams 䌚议记圕  │
│          → 扟到䌚议摘芁提到 Part-X 短猺预计延迟 3 呚   │
│                                                          │
│  Step 2: [Fabric IQ] 查询䞚务数据                         │
│          → Ontology 理解 "Part-X" 是什么                  │
│          → 查询 Order 和 Inventory 实䜓                   │
│          → 分析 Q2 销售预测受圱响的订单                    │
│                                                          │
│  Step 3: [Foundry IQ] 检玢知识文档                        │
│          → 圚 SharePoint 知识库䞭扟到《䟛应铟应急手册》    │
│          → 提取盞关的应急措斜建议                          │
│                                                          │
│  Step 4: 绌合掚理并生成回答                               │
│          → 结合䞉䞪来源的信息给出有据可查的分析报告       │
└──────────────────────────────────────────────────────────┘

这就是䞉倧 IQ 协同工䜜的嚁力——单独䞀䞪 IQ 只胜回答郚分问题䞉者组合才胜给出完敎的、有䞊䞋文的、可远溯的答案。


8. 参考资料 (References)



📌 English Version


1. Overview

In the age of AI Agents, Microsoft has built a grand vision: AI should not just be a “chat assistant,” but a true “digital coworker” that understands your work, your data, and your business. At the heart of this vision is the Microsoft Agent Factory — an ecosystem for enterprises to build, deploy, and manage AI Agents at scale.

But for an Agent to be truly useful, it cannot rely solely on the general knowledge of a Large Language Model (LLM). It needs organizational context — your emails, meetings, documents, business data, data warehouses, and more. This is exactly the problem Microsoft’s three IQ layers are designed to solve:

IQ Layer One-Line Summary Data Source
Work IQ Understands how you work Microsoft 365 (email, meetings, chats, documents)
Fabric IQ Understands what your business data means Microsoft Fabric (OneLake, Power BI, data warehouses)
Foundry IQ Connects all your enterprise knowledge Azure Blob, SharePoint, OneLake, public web

An analogy:

If an AI Agent is a new hire, then Work IQ tells it “how people collaborate and what’s being discussed”; Fabric IQ helps it understand “what data reports and business metrics actually mean”; and Foundry IQ enables it to “browse the company’s knowledge base and documentation.”

The three are independent but can be used together to provide comprehensive organizational context for agents.


2. Core Concepts

2.1 What is Agent Factory?

Agent Factory is Microsoft’s umbrella term for its AI Agent building platform ecosystem. The core principle: make building agents as standardized, scalable, and governable as a factory assembly line.

The Agent Factory encompasses:

  • Microsoft Copilot Studio: Low-code/no-code agent builder
  • Microsoft Foundry (formerly Azure AI Foundry): Pro-code agent development platform
  • Agent 365: Unified management control plane for agents (registry, security, governance)
  • Three IQ Layers: Provide intelligent context for agents (Work IQ, Fabric IQ, Foundry IQ)
┌──────────────────────────────────────────────────────┐
│               Microsoft Agent Factory                 │
│  ┌───────────────┐  ┌────────────────┐  ┌──────────┐ │
│  │ Copilot Studio│  │  Microsoft     │  │ Agent 365│ │
│  │ (Low-code)    │  │  Foundry       │  │(Govern)  │ │
│  │               │  │  (Pro-code)    │  │          │ │
│  └───────┬───────┘  └───────┬────────┘  └────┬─────┘ │
│          │                  │                │        │
│          └──────────┬───────┘                │        │
│                     â–Œ                        │        │
│  ┌──────────────────────────────────────────────────┐ │
│  │         Three IQ Layers (The Agent Brain)         │ │
│  │  ┌───────────┐ ┌───────────┐ ┌──────────────────┐│ │
│  │  │  Work IQ  │ │ Fabric IQ │ │   Foundry IQ     ││ │
│  │  │Work Context│ │Biz Semant.│ │ Knowledge Retriev││ │
│  │  └───────────┘ └───────────┘ └──────────────────┘│ │
│  └──────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────┘

2.2 Why Three IQs?

LLMs know nothing about your company. To make agents useful in an enterprise, you need to inject “internal knowledge” securely into the agent’s reasoning process.

Enterprise knowledge falls into three categories:

  1. Work collaboration data (who met whom, what was discussed, what emails said) → Work IQ
  2. Structured business data (sales, inventory, customer data, and their “business meaning”) → Fabric IQ
  3. Broad unstructured knowledge (technical docs, product manuals, KB articles, web pages) → Foundry IQ

Each category has different storage locations, formats, and access control mechanisms, requiring specialized intelligence layers.


3. How It Works

3.1 Work IQ — “Understanding How You Work”

Work IQ is the intelligence engine behind Microsoft 365 Copilot. It makes Copilot understand your work context.

Three-Layer Architecture
  • Data Layer: M365 tenant data (via Microsoft Graph), Copilot Connectors for non-Microsoft systems, Dynamics 365/Dataverse for business system data
  • Context Layer: Memory (explicit + implicit), Semantic Index (meaning-based retrieval), Business Understanding (ontologies and glossaries from business workflows)
  • Skills & Tools Layer: Specialized instructions (skills) and execution tools (MCP servers, APIs, plugins)
Key Capabilities
  • Personalized responses based on your work patterns, relationships, and communication history
  • Cross-system reasoning: connect Teams meeting content with Dynamics 365 sales data
  • Enterprise-grade security: respects M365 permissions, sensitivity labels, DLP policies
Developer Access
  • Work IQ API: RESTful interface for integrating Work IQ intelligence into custom apps
  • Work IQ CLI & MCP Server: npm install -g @microsoft/workiq for terminal and VS Code integration

3.2 Fabric IQ — “Understanding What Your Data Means”

Fabric IQ is the semantic intelligence layer for Microsoft Fabric. While Work IQ understands “how people work,” Fabric IQ understands “what data tells you.”

Core Components
Component Purpose
Ontology Define enterprise vocabulary — entity types (Customer, Order, Shipment), properties, relationships, rules, bound to real data
Graph Graph storage and traversal for relationship-heavy queries (e.g., Order → Shipment → Sensor → Breach)
Data Agent Conversational Q&A connected to Ontology for business-aware answers
Operations Agent Real-time data monitoring with business action recommendations
Power BI Semantic Model Curated analytics models for reporting and interactive analysis
Why Ontology Matters

Without Ontology, column names like qty_shipped and flg_breach are meaningless to agents. Ontology provides the “Rosetta Stone” that translates raw data into business language.

Key Benefits
  • Data unification across OneLake sources
  • Consistent language — define a concept once, use it everywhere (Power BI, notebooks, agents)
  • Cross-domain reasoning via Graph traversals
  • AI-ready grounding for Copilot and agents

3.3 Foundry IQ — “Connecting All Enterprise Knowledge”

Foundry IQ is the managed knowledge layer in Microsoft Foundry. It solves a practical problem: enterprise knowledge is scattered across Azure Blob Storage, SharePoint, OneLake, and even public websites — agents need a unified way to retrieve it.

Core Architecture
  • Knowledge Base: Top-level resource orchestrating retrieval. Defines which sources to query and retrieval parameters.
  • Knowledge Sources: Connections to Azure Blob, SharePoint, OneLake, and web data. Automatic chunking, vector embedding, metadata extraction, and incremental refresh.
  • Agentic Retrieval Engine: Multi-query pipeline that:
    1. Decomposes complex questions into subqueries
    2. Executes them in parallel across sources
    3. Semantically reranks results
    4. Returns unified responses with citations
Permission-Aware Retrieval
  • ACL synchronization for supported sources
  • Honors Microsoft Purview sensitivity labels
  • Queries run under the caller’s Microsoft Entra identity
  • End-to-end permission enforcement

4. Comparison

Dimension Work IQ Fabric IQ Foundry IQ
Purpose Work collaboration intelligence Business data semantics Enterprise knowledge retrieval
Core Question How do people work? What does the data mean? Where is the knowledge?
Platform Microsoft 365 Copilot Microsoft Fabric Microsoft Foundry (Azure)
Data Sources Email, meetings, Teams chats, docs, Dynamics 365 OneLake (lakehouse, eventhouse), Power BI semantic models Azure Blob, SharePoint, OneLake, public web
Key Capability Semantic index, memory, skills/tools Ontology modeling, Graph traversal, Data Agent Knowledge base, Agentic Retrieval
Example Query “What was discussed in last week’s meeting with Alice?” “Which A-tier customers had declining sales last month?” “What does our return policy document say?”

5. How They Work Together

A real-world scenario requiring all three:

User: "The parts shortage discussed in last week's Teams call with
      supplier Contoso — how will it impact our Q2 sales forecast?
      Please reference our supply chain contingency playbook."

Agent Reasoning:
  Step 1: [Work IQ] Find last week's Teams meeting with Contoso
          → Meeting summary: Part-X shortage, 3-week delay expected

  Step 2: [Fabric IQ] Query business data
          → Ontology resolves "Part-X" entity
          → Query Order and Inventory entities
          → Analyze impacted Q2 sales forecast

  Step 3: [Foundry IQ] Retrieve knowledge documents
          → Find "Supply Chain Contingency Playbook" in SharePoint KB
          → Extract relevant contingency measures

  Step 4: Synthesize and respond
          → Combine all three sources into a cited analysis report

No single IQ can answer the full question. Together, they provide complete, contextual, traceable answers.


6. References