Compare AI agent solutions
Choosing the right Microsoft AI agent solution starts with understanding your organization’s goals, technical environment, and user needs. This unit reviews the key decision criteria that developers and organizations can use to evaluate whether to adopt Microsoft’s Copilot experiences, extend them with custom logic and data, or build entirely custom agents and generative AI applications.
Comparison
The following table provides a comparison of the three core agent paths:
| Criteria | Adopt Copilot | Extend Copilot | Design and build a custom solution |
|---|---|---|---|
| Development effort | Low | Medium | Medium–High (varies based on data and complexity) |
| Data sources | Microsoft 365 tenant data | Microsoft 365 + third-party/internal data | Multiple data stores, services, and apps inside and outside Microsoft 365 |
| Target users | Internal users (Microsoft 365 tenant) | Internal users (Microsoft 365 tenant) | Internal users or external customers |
| Customization level (e.g., temperature, max_tokens, metaprompts) | Low | Medium | High |
| Workflow complexity | Simple, document-based tasks | Moderate, multistep workflows with business logic | Complex, multisystem orchestration and logic |
| AI capabilities required | Summarization, drafting, and Q&A | Custom actions, plugins, and data retrieval | Advanced AI (e.g., RAG, memory, autonomous agents) |
| Development resources | Business users and business teams | Low-code makers and IT pros | Makers, pro-code developers, and solution architects |
| Data, privacy, and security | Covered by Microsoft Purview and security | Inherits Microsoft 365 controls, but custom connectors may require additional review | You are responsible for implementing and validating compliance, privacy, and security controls |
| Cost and licensing | Microsoft 365 Copilot license or Copilot Chat | May require additional licenses for connectors, plugins, or Copilot Studio | May require Azure, Copilot Studio, or other licenses; additional infrastructure and consumption costs |
Choose your path
Review the following information to evaluate your decision criteria and choose the appropriate path for your scenario. Custom agents are ideal for developers building sophisticated, enterprise-grade AI assistants, while declarative agents are great for quickly extending Copilot with low-code tools for common business scenarios.
| Solution | Description | Key Highlights | Best For |
|---|---|---|---|
| Adopt | Deliver AI-powered productivity with Microsoft 365 Copilot, Copilot Chat, and out-of-the-box agents. | Quickest path to value with minimal configuration | Teams seeking fast productivity gains with minimal setup |
| Extend | Customize Microsoft 365 Copilot with declarative agents and Copilot connectors. | Balances customization with ease of integration | Organizations with unique data or workflows that need tailored solutions integrated with Microsoft 365 Copilot |
| Build | Create fully custom agents or generative AI applications using Microsoft 365 Agents SDK, Copilot Studio, Azure AI Foundry, and related tools and services. | Greater customization with additional responsibility and cost | Developers seeking fully customized, advanced, scalable, or commercial AI solutions |
Adopt
Choose this path if:
- Your users primarily need help with productivity tasks like summarizing, drafting, or organizing
- You want to get started quickly with minimal configuration
- Your organization already uses Microsoft 365 apps like Word, Excel, Outlook, and Teams
- You want to leverage Microsoft’s built-in security, compliance, and privacy controls
- You want predictable licensing and cost structure
Examples:
- A global sales team at a professional services firm uses Microsoft 365 Copilot in Outlook and Word to draft client proposals and summarize meeting notes, reducing time spent on documentation by over 30%.
- A project manager at a manufacturing company uses Copilot in Teams to automatically generate action items and follow-ups from weekly team meetings, improving task tracking and accountability.
- A marketing team at a global consumer goods company uses Copilot Chat to streamline campaign planning and content creation.
Extend
Choose this path if:
- You want to tailor Copilot to your business processes or data
- You need to connect Copilot to internal systems or external APIs
- You want to build conversational agents that feel like native Copilot experiences
- You want your agent to work within Copilot's orchestration and language models and leverage existing security and compliance controls
- You want to manage costs by leveraging existing Microsoft 365 licenses, but are prepared for potential additional licensing for connectors or Copilot Studio
Examples:
- An HR team at a healthcare organization builds a declarative agent in Copilot Studio to answer employee questions about benefits and policies, using SharePoint and internal HR systems as data sources.
- A finance department at a global logistics company builds a declarative agent with a plugin that retrieves real-time budget and forecast data from their ERP system, enabling department leads to make faster, data-driven decisions.
- A legal team at a global consulting firm uses a Copilot connector to index thousands of internal policy documents stored in a third-party document management system.
Build
Choose this path if:
- You need full control over the agent’s logic, orchestration, or user experience
- You’re building a customer-facing or commercial AI application
- You want to use custom models or advanced AI techniques like RAG or autonomous agents
- You want advanced Teams integrations (such as with meetings and channels)
- You want to make your agent available outside of Microsoft 365 and Copilot
- You want to support proactive messaging, enabling developers to define workflows and trigger agent behavior without user interaction
- You need multiple system integrations, such as managing logistics by integrating data from GPS, warehouse systems, and customer databases
- You have complex decision-making (for example, evaluating loan applications based on multiple factors like credit score, income, and employment history)
- You are prepared to manage additional costs for cloud infrastructure, licensing, and compliance
Examples:
- A product team at a travel tech startup creates a branded AI assistant embedded in their mobile app using Azure AI Foundry and OpenAI models. The assistant helps users plan trips, book flights, and manage itineraries using natural language.
- A customer service team at a retail company builds a custom engine agent, deployed as a Teams app, using the Teams SDK. The agent assists internal support staff by answering product-related questions, checking inventory levels, and initiating return workflows. It uses OpenAI models to interpret user intent and integrates with the company’s internal APIs to retrieve real-time data.
Reflection
Question:
After reviewing the decision criteria and solution paths, which approach—Adopt, Extend, or Build—best aligns with your organization's needs and goals? What factors influenced your choice?