Build agents with Microsoft tools

  • 12 minutes

As this video explained, Microsoft provides a range of platforms for building and deploying AI agents. From easy-to-use tools that any public servant can use right away, to advanced development environments for technical teams building complex, mission-critical solutions, the right platform depends on use case, level of customization required, and technical resources available.

Three platforms—Microsoft 365 Copilot, Copilot Studio, and Azure AI Foundry—form a continuum from no-code to pro-code. Most public sector organizations will use all three, with different teams and use cases drawing on different capabilities.

Microsoft 365 Copilot agents

Microsoft 365 Copilot includes built-in capabilities for creating agents that help individuals and small teams work more effectively. Using Agent Builder—available directly in Microsoft 365 Copilot Chat and Microsoft Teams—public servants can create agents without writing any code.

These agents are grounded in Microsoft 365 data that the user already has access to through Microsoft Graph, including documents, emails, meetings, and SharePoint content. Because they respect existing Microsoft 365 permissions, they integrate with organizational data access controls without requiring additional configuration.

Microsoft 365 Copilot agents are well suited for:

  • Summarizing and searching across large SharePoint libraries or document collections
  • Answering questions based on team-specific knowledge, such as procedures, policies, or reference materials stored in SharePoint
  • Helping individuals manage recurring tasks like preparing briefings, synthesizing meeting notes, or drafting standard correspondence
  • Supporting small-group workflows where a shared agent reduces duplicated effort across a team

Note

Microsoft 365 Copilot agents are designed for personal or small-group use. For organization-wide deployments, citizen-facing services, or complex multi-step workflows, Copilot Studio or Azure AI Foundry provide the additional capabilities and control needed.

Microsoft Copilot Studio

Copilot Studio is Microsoft’s low-code platform for building more sophisticated agents. It provides a visual development environment that doesn’t require professional developers, while still supporting advanced customization through connectors, API integrations, and custom logic when needed.

With Copilot Studio, public sector teams can build agents that:

  • Connect to specific, curated knowledge sources such as documents, websites, databases, and APIs and ensuring agents provide authoritative, policy-accurate responses
  • Follow defined conversation flows and handle topics consistently across high volumes of interactions
  • Respond to triggers such as automated events like form submissions or database changes without requiring manual initiation
  • Collaborate with other agents in multi-agent configurations for more complex workflows
  • Deploy across multiple channels simultaneously, including websites, Microsoft Teams, and phone systems

Azure AI Foundry

Azure AI Foundry is Microsoft’s developer-first platform for building advanced, scalable AI agents and applications. It provides access to a model catalog of more than 1,900 models, spanning foundation models, reasoning models, small language models, multimodal models, and domain-specific models. Azure AI Foundry gives technical teams full control over agent architecture, orchestration, and deployment.

Azure AI Foundry is the right choice when:

  • Agents need to integrate with legacy systems, external APIs, or custom data pipelines that aren’t available through standard connectors
  • Workflows are complex, long running, and require durable state across hours, days, or weeks
  • The use case requires multi-agent coordination across departments, systems, or jurisdictions
  • Advanced capabilities like vision-based document analysis, multilingual processing, or custom model fine-tuning are needed
  • Organizations need full auditability, custom security controls, and detailed observability of agent behavior

Choose the right platform

The table summarizes key characteristics of each platform to help public sector organizations match the right tool to the right use case.

Platform Best for Technical requirement Deployment scope
Microsoft 365 Copilot Personal and small-group productivity agents None (no-code) Individual or small team
Copilot Studio Departmental and citizen-facing agents Low-code Department or organization-wide
Azure AI Foundry Complex, enterprise, and mission-critical agents Pro-code Enterprise or multi-system

Tip

Organizations don’t need to start with the most complex platform. Many successful agent deployments begin with a Copilot Studio agent for a well-defined use case, then scale to Azure AI Foundry as needs grow. Starting simple, demonstrating value, and expanding is a proven approach.

Key agent capabilities

Beyond the platforms themselves, a set of cross-cutting capabilities determines how sophisticated and effective an agent can be. Understanding these helps public sector teams design agents that match the demands of their workflows.

Knowledge

Agents need access to information to be useful. “Knowledge” refers to data sources agents draw on when generating responses and taking action. The three platforms handle knowledge differently:

  • Microsoft 365 Copilot uses implicit knowledge, drawing on Microsoft 365 data the user already has access to through Microsoft Graph (documents, emails, meetings, SharePoint content). Agents can’t be configured to use only specific sources.
  • Copilot Studio uses explicit, managed knowledge, and teams specify which documents, websites, or data sources an agent uses. This gives departments control over what the agent knows and where its answers come from, which is essential for ensuring responses are policy-accurate and authoritative.
  • Azure AI Foundry provides full control over knowledge architecture. Technical teams can build custom retrieval pipelines (using techniques like retrieval-augmented generation, or RAG) to connect agents to any data source with precise control over what information is retrieved and how it’s used.

Note

For most public sector use cases, explicit and managed knowledge is essential. Agents that interact with citizens or inform policy decisions need to draw from approved, authoritative sources—not from general internet searches or unverified data.

Durable agents

Many government processes don’t complete in a single session. Permit applications move through multiple reviewers over days. Benefits claims require supporting documents submitted at different times. Compliance investigations span weeks.

Durable agents maintain context, state, and memory across long-running or multi-step workflows—even when processes span days or require multiple human interactions. This means:

  • The agent doesn’t reset or “forget” where a case left off when a process is paused
  • Staff don’t need to manually restart work or re-enter context when resuming a workflow
  • Workflows continue automatically when the next step is ready, without requiring human prompting at each stage

This capability is especially powerful for complex casework, permitting, benefits processing, and compliance tasks—some of the highest-volume, most resource-intensive work in government environments.

Triggers

Triggers automatically start an agent’s workflow when a specific event occurs such as a form submission, a database record change, an email arriving, a calendar event, or a scheduled time. Rather than waiting for a human to initiate a process, the agent responds to the event and begins working immediately.

For public sector organizations, triggers enable:

  • Automatic initiation of intake processes when a citizen submits an online form or service request
  • Fraud detection alerts to be generated when transaction data falls outside established parameters, without waiting for a scheduled audit
  • Compliance checks to be initiated automatically when new documents are added to a regulated repository
  • Routine reporting workflows to run on a defined schedule without staff intervention

Triggers are available in Copilot Studio and can also be implemented in Azure AI Foundry for more advanced workflow automation.

Multi-agent orchestration

Complex public sector processes rarely fit within a single team or system. A benefits application might involve eligibility verification, document validation, means testing, approval routing, and citizen notification, each handled by different departments with different data and systems.

Multi-agent orchestration allows multiple specialized agents to collaborate, with each handling a distinct part of a larger workflow. A coordinator agent manages the overall process and routes tasks to the appropriate specialist agents, which complete their portions and return results. This pattern is well suited for:

  • Processes that cross departmental or system boundaries and require coordination across teams
  • Workflows requiring validation at multiple stages by different teams with different expertise
  • Mission-critical applications where specialized accuracy in each domain is essential to the overall outcome

Model context protocol (MCP)

Model context protocol (MCP) is an open standard that provides a secure, standardized way for agents to connect to external systems, APIs, and data sources. Rather than building custom integrations for each external system, organizations configure MCP to give agents structured, permission-controlled access to the data and tools they need.

For public sector organizations, MCP offers a compliant method to integrate AI agents with:

  • Legacy case management and record systems that predate modern API standards
  • External databases and government data registries
  • Third-party service providers and cross-agency data systems
  • Sensitive citizen data stores, with clear permission boundaries and full auditability of what data the agent accessed and when

Because MCP is an open standard adopted across major AI platforms, it also protects organizations from vendor lock-in. Agents built to use MCP can work with multiple AI systems and adapt as technology evolves.

Channels

Channels are the surfaces through which agents interact with users, more commonly known as the front-end interfaces people use to access agent capabilities. An agent’s core logic can be deployed across multiple channels simultaneously without rebuilding it for each one.

Available channels include Microsoft Teams, public-facing websites, SMS messaging, voice and phone systems, and custom applications. Multi-channel deployment is particularly valuable in helping public sector agencies provide accessible service delivery to all:

  • Citizens can access services through their preferred channel (web, phone, or in-person kiosk) without being forced into a single technology
  • Frontline staff can access agent assistance directly in the tools they already use, like Microsoft Teams, without switching applications
  • Organizations can reach residents who don’t have reliable internet access through SMS or voice channels, reducing the digital divide in service delivery

Resources