The proliferation of generative AI has also led to a proliferation of techniques and tools to use AI. Microsoft’s ecosystem itself while emerging as a versatile platform for building AI agents has a variety of agent options to navigate.
From low-code declarative agents focused primarily on instructions and grounding in business or public data to fully customized multimodal solutions, anyone from developers to business users have multiple pathways to create agents tailored to specific workflows. There are choices, but how can users unfamiliar with the ecosystem make the most appropriate choices (or only half-jokingly, “Why not use AI to choose?”).
This post examines four primary tools:
1) Agent Builder in Copilot Chat
2) Copilot Studio
3) Teams Toolkit
4) Azure AI Foundry
It contrasts their capabilities, target audiences, and paradigms.
But first, let’s explore two distinct types of agents that will impact choices ahead.
Declarative Agents vs. Custom Engine Agents: Foundational Differences
At the core of Microsoft’s AI agent strategy lies a distinction between declarative agents and custom engine agents.
Declarative agents leverage Microsoft Copilot’s prebuilt infrastructure, including its orchestrator, language models, and security framework. These agents are configured through natural language instructions, predefined actions, and enterprise knowledge sources like SharePoint or Graph connectors (plus optional web grounding, code interpreter, and image generator).
They excel in scenarios requiring rapid deployment within Microsoft Copilot Chat, such as onboarding assistants or IT helpdesk agents. By inheriting Copilot’s compliance and access controls, declarative agents minimize development overhead while ensuring enterprise-grade governance.
In contrast, custom engine agents operate on developer-selected and even developer-trained language models, enabling organizations to integrate proprietary language models, advanced orchestration logic, or multimodal capabilities. These agents support the most advanced proactive workflows (e.g., automated inventory restocking) and even cross-platform deployments.
While offering greater flexibility, custom engine agents carry higher operational complexity to maintain security and regulatory alignment.
Agent Development Tools: Capabilities and Tradeoffs
Agent Builder in Copilot Chat
Integrated directly into Microsoft 365 Copilot Chat, this tool allows business users to create declarative agents through conversational prompts. Users describe an agent’s purpose (e.g., “Market research advisor using corporate financial statements”) and configure knowledge sources via a guided interface.
Highlights:
• Natural Language Configuration: Agents are defined through free-form descriptions of their role, tone, and data sources.
• Built-In Grounding: Supports SharePoint files, public websites, and Graph connectors without requiring API integrations.
• Metered Consumption OR included in Microsoft 365 Copilot per user license: Agents accessing tenant data (e.g., internal SharePoint sites) follow pay-as-you-go billing if desired, or can use Microsoft’s per user licensing as well.
• Zero-Code Accessibility: Enables business analysts or subject-matter experts to deploy agents in minutes.
• Native Microsoft 365 Integration: self-created agents appear alongside others in Copilot Chat, requiring only basic publishing and sharing steps (still subject to organizational governance and administration).
• Limited Customization: Currently lacks support for custom APIs, multi-step workflows, multimodal agents, or external channel deployments.
• Dependency on Copilot Infrastructure: Cannot incorporate specialized models (e.g., medical LLMs)
Target Audience:
Individuals or line-of-business teams needing task-specific assistants without IT dependency. Example includes a targeted FAQ agent built on knowledge in a team’s SharePoint site. (SharePoint itself has agents, which is outside the scope of this post but which are a special case of declarative agents)
Copilot Studio
Positioned as a low-code/pro-code hybrid platform, Copilot Studio expands declarative agent capabilities with advanced features like Power Automate integrations and multi-channel publishing.
Highlights:
• Visual Orchestration: Drag-and-drop interface for designing conversational flows and connecting to Graph and Power Platform connectors (including outside the Microsoft ecosystem such as Salesforce or ServiceNow).
• Environments .
• Power Platform Foundation: Agents built in Copilot Studio take advantage of Power Platform environments and solutions.
• Balanced Flexibility: Supports both declarative configurations and custom engine agents.
• Centralized Governance: Admins manage agent access via Microsoft Admin Center and Power Platform Admin Center, with audit logs and policies for governance in Purview.
• Licensing: Requires Copilot Studio subscriptions for development unless using a “lightweight” Microsoft 365 Copilot license and only publishing to that channel.
Target Audience:
“Makers” in IT or operations roles building department-wide or enterprise-wide assistants.
Teams Toolkit for Visual Studio Code
A developer-centric toolkit for building declarative or custom engine agents as apps.
Highlights:
• Code-First Development: Create agents using SDKs and include adaptive cards, message extensions, and proactive triggers.
• Model Agnosticism: Integrate any LLM.
• Full Control Over AI Stack: Developers dictate orchestration logic, model choices, actions.
• Enterprise Scalability
• Learning Curve
Target Audience:
Software engineers creating complex assistants for scenarios like clinical trial management or supply chain optimization.
Azure AI Foundry
Microsoft’s cloud AI platform provides foundational services for building the most advanced agents from scratch.
Highlights:
• Multimodal Capabilities: Combine vision, speech, and language models in agents.
• Hybrid Orchestration: Deploy agents across multiple environments.
• Custom Fine-Tuning: Train domain-specific models using proprietary data.
• Architectural Freedom: Supports any agent design pattern.
• Global Infrastructure: Custom data residency and other needs in or across Azure’s regions for the most compliant deployments.
• Development Resources: Requires dedicated engineers for development and DevOps teams for ongoing maintenance.
• Consumption pricing: Metered billing
Target Audience:
Enterprises requiring bespoke agents for deep, scalable scenarios, custom models, or advanced real-time use cases such as fraud detection or audio analysis and subsequent real-time action.
Aligning Tool Choice with Organizational Needs
The Microsoft ecosystem accommodates diverse agent development needs, but tool selection hinges on three factors: technical expertise, workflow complexity, and solution requirements.
Business units prioritizing speed-to-value will favor Agent Builder or Copilot Studio, while engineering teams tackling unique challenges (e.g., real-time translation agents) require Teams Toolkit or Azure.
Critically, declarative agents reduce time to market but do not permit as much innovation or model selection as a tradeoff, whereas custom engine agents offer the most bespoke options and customization potential. As Microsoft continues blending these paradigms (which is abstracted from the users but evident to admins who find themselves jumping between the M365 Admin Center, Power Platform Admin Center, and Purview), organizations can architect agent portfolios that balance development and audience agility with their strategic AI investments.