This is Part 3 of a 3-part series on extending AI to enterprise file data, showing how the knowledge pipeline is surfaced through enterprise AI agents and user experiences including Microsoft 365 Copilot.
Table of Contents
Why the user experience layer matters in enterprise AI
Microsoft Foundry: Orchestrating enterprise AI agents
The Copilot agent as the enterprise user experience
Why this pattern works for regulated and complex environments
Extending Copilot beyond SharePoint‑centric scenarios
From storage to knowledge to action
Put the architecture into practice
Introduction
In Part 1, From Enterprise File Storage to an AI-Ready Data Foundation using Azure NetApp Files and OneLake, we established a zero‑copy data foundation that makes enterprise file data AI‑addressable using Azure NetApp Files and Microsoft OneLake, without migration, duplication, or workflow disruption.
In Part 2, From File Data to AI‑Powered Knowledge Pipelines using Azure NetApp Files object REST API, we showed how that foundation is activated through a knowledge pipeline using Azure AI Search and Azure OpenAI, transforming unstructured files into a living, queryable knowledge system grounded in Retrieval‑Augmented Generation (RAG).
In this final part, we focus on the user experience: how this architecture is safely and effectively surfaced to end users through enterprise AI agents, including Copilot, where employees already work.
Co-authors:
- Thomas Willingham, Azure NetApp Files Product Manager
- Sean Luce, Azure NetApp Files Product Manager
Why the user experience layer matters in enterprise AI
Enterprise AI adoption rarely fails because models are insufficient. It fails when users are asked to change how they work, learn new tools, or trust opaque answers.
For knowledge‑centric scenarios, success depends on three things:
- Users interact with AI inside their existing workflows.
- Answers are grounded in authorized enterprise data.
- Responses are traceable, and auditable.
Microsoft 365 Copilot provides the ideal natural language experience for this layer. However, Copilot itself does not “understand files.” It relies on external knowledge sources and well‑defined grounding mechanisms to retrieve and synthesize information safely.
This is where the Copilot Agent completes the architecture, with Microsoft Foundry providing the orchestration layer behind it.
Microsoft Foundry: Orchestrating enterprise AI agents
Before enterprise knowledge is exposed to end users through enterprise AI agents such as Copilot, it must first be assembled, grounded, and governed as an AI capability. This is the role of Microsoft Foundry.
Microsoft Foundry is the control plane for building and governing enterprise AI agents. It provides the environment where enterprise AI agents are created and connected to authoritative knowledge sources. It orchestrates how Azure AI Search, Azure OpenAI models, and enterprise data sources, including Azure NetApp Files surfaced through OneLake, work together to deliver an accurate, grounded response.
Rather than embedding intelligence directly inside Copilot, Foundry acts as the control plane for enterprise AI. It defines:
- Which knowledge sources an agent can access,
- How retrieval and grounding are enforced,
- How models are constrained to retrieved enterprise content, ensuring responses remain grounded, explainable, and auditable.
This separation ensures that Copilot remains the user experience, while Foundry governs how enterprise AI agents reason over data, securely and predictably.
Together, this user experience layer enables secure, seamless access to enterprise data for AI, unlocking real business value through governed, traceable Copilot experiences.
The Copilot agent as the enterprise user experience
The Copilot agent is built and governed within Microsoft Foundry, but users interact with it through Copilot.
At a high level, the agent performs three critical functions:
- Receives natural language questions from users inside M365 experiences such as Teams or Copilot Chat,
- Queries the Azure AI Search index built over enterprise file data referenced through OneLake,
- Generates grounded responses using Azure OpenAI, restricted to retrieved content and enriched with source citations.
This separation of responsibilities is deliberate. Copilot does not read file storage directly, nor does it reason over raw documents. Instead, it queries a curated, indexed knowledge source designed specifically for enterprise RAG scenarios.
The result is a user experience that feels conversational but behaves predictably.
End‑to‑end user query flow
When a user asks a question in Copilot – such as:
“What is our incident escalation procedure for data breaches?”
The interaction follows a controlled, enterprise‑safe flow:
- User submits a question in Copilot,
- The Copilot Agent forwards the query to Azure AI Search,
- Relevant document chunks are retrieved using semantic vector search,
- Azure OpenAI synthesizes a response using only the retrieved content,
- Source filenames are included in the response for transparency.
At no point does the model infer information from outside the enterprise dataset or hallucinate answers based on its general training.
This is not a chatbot layered on top of file storage. It is a governed retrieval and reasoning pipeline surfaced through a familiar, conversational experience.
Why this pattern works for regulated and complex environments
This user experience model is especially powerful for industries where trust, traceability, and compliance are non‑negotiable.
Because answers are always grounded in specific source files:
- Legal teams can verify interpretations against contract language,
- Security teams can audit responses back to official runbooks,
- Compliance teams can validate that policies, not assumptions, are being cited.
Just as importantly, data never leaves its system of record:
- Files remain governed and secured in Azure NetApp Files,
- Access controls continue to be enforced at the storage layer,
- OneLake provides virtualization, not duplication,
- AI services operate on indexed representations, not copied datasets.
This preserves existing governance models while extending their reach into AI‑driven experiences.
Extending Copilot beyond SharePoint‑centric scenarios
Most Copilot examples focus on collaboration content stored in SharePoint or OneDrive. These are valid and important scenarios.
However, enterprise knowledge does not live exclusively in collaborative repositories.
Large organizations depend on:
- Network file systems containing decades of operational knowledge,
- Hybrid environments where SMB and NFS are the primary interfaces,
- High‑performance storage platforms that cannot be re‑platformed.
By combining Azure NetApp Files, OneLake, a RAG‑based knowledge pipeline, and Copilot, this architecture extends Copilot’s reach to the realities of the enterprise data estate, without forcing that estate to change.
From storage to knowledge to action
Taken together, the three layers form a cohesive architecture:
- Part 1: Data Foundation
Azure NetApp Files + OneLake make enterprise file data AI‑addressable without migration or duplication. - Part 2: Knowledge Pipeline
Azure AI Search + Azure OpenAI transform files into a living knowledge system using RAG. - Part 3: AI Orchestration & User Experience
Microsoft Foundry orchestrates enterprise AI agents, and Copilot provides a natural, trusted way for users to ask questions and receive grounded answers.
This is not a single-use solution. It is a repeatable architecture pattern for enabling secure, seamless access to enterprise data for AI, unlocking real business value safely, predictably, and at scale.
Key takeaway
The real promise of enterprise AI is not replacing systems of record but making them intelligible.
With this architecture, Azure NetApp Files becomes more than high performance storage.
It becomes the foundation of an AI‑ready data estate, delivering grounded insights inside Copilot, without compromising performance, security, or trust.
Together, these three parts form a complete architecture for extending AI to enterprise file data – bridging the gap between systems of record and AI-powered user experiences.
Put the architecture into practice
This three‑part series outlines a repeatable pattern for bringing AI to enterprise file data; without migration, duplication, workflow disruption, or governance compromise.
To go deeper:
- Review the reference architecture described in this series and map it to your environment,
- Identify candidate file‑based knowledge scenarios (security, compliance, operations),
- Engage your data, storage, and AI teams, with us at askANF@microsoft.com to apply this pattern using Azure NetApp Files, Microsoft OneLake, and Copilot.
Enterprise AI succeeds when systems of record become systems of intelligence.
Learn more
- From Enterprise File Storage to an AI-Ready Data Foundation using Azure NetApp Files and OneLake
- From File Data to AI‑Powered Knowledge Pipelines using Azure NetApp Files object REST API
- Configure object REST API in Azure NetApp Files | Microsoft Learn
- OneLake, the OneDrive for data - Microsoft Fabric | Microsoft Learn
- Microsoft Foundry | Microsoft Learn
- Build a Copilot agent for Microsoft 365 | Microsoft Learn
Microsoft