How Microsoft Can Help You Build Gen AI-Based Agents and Multi-Agent Apps

In the fast-evolving world of Generative AI (GenAI), the concepts of "agent apps" and "multi-agent apps" are transforming how organizations operate and innovate. While these terms may sound technical, they have significant implications for business operations, collaboration, and growth in the digital era. GenAI agents are not just boosting productivity; they’re also driving innovation across industries. In this article, we’ll explore why GenAI agents and multi-agent apps are essential and provide recommendations on how to leverage Microsoft’s capabilities to build them.

What are Agents and Multi-Agent Systems?

• Agents: An agent is an autonomous entity capable of perceiving its environment and taking actions to achieve specific goals. Think of it as a self-contained software program designed to make decisions, interact with its surroundings, and execute tasks without constant human intervention.
• Multi-Agent Systems (MAS): A multi-agent system comprises multiple agents working together to solve problems that are beyond the capabilities of a single agent. These agents collaborate, communicate, and sometimes even compete to achieve shared objectives. Multi-agent apps take the concept of GenAI agents a step further by enabling collaboration among multiple agents to solve complex tasks.

The Evolution of Generative AI Use Case Adoption

At Microsoft, we have been investing in GenAI based applications for years and have observed its evolution in waves:

Wave 1 Use Case Adoption

In the first wave of use cases we saw our customers adopt in the first half of 2022, we saw our customers evaluating use cases internal to their organizations. They did this primarily to learn the technology before they brought it into the customer channels and engagements. Organizations have adopted this approach with other technological waves like the Metaverse and Blockchain. It is a great way to minimize risk and avoid reputational risk. Deploying generative AI internally also reduces the need for regulatory oversight. The other dimension to this is that generative AI is fundamentally a productivity accelerator – so the center of gravity for use cases will be around augmenting employees to be more productive.

Wave 2 Use Case Adoption

Wave 2 use cases are characterized by bringing generative AI into the customer conversation. Here mainly Banks and Citizen Services Entities (mainly owned by the Government or States) have deployed public-facing chatbots or agents for several years now, but they have been very limited. Most bots can answer one or two questions before bouncing out the customer to either a contact center or resource on their website. Generative AI will enable these organizations to have a much more natural conversation with their customers or citizens

Wave 3 Use Case Adoption

Wave 3 use cases are much harder to predict. We believe that new classes of products and services could be developed using a combination of predictive AI and generative AI. As we innovate directly with customers to build out their roadmaps more of these use cases will manifest themselves. As new capabilities develop in Artificial General Intelligence (AGI), organizations could leverage AGI to automate processes across the bank in new ways, interact with customers in new ways, or apply AGI to dynamically manage the operations of their business units.

As an example, in the FSI industry we see a series of AGIs looking after core systems, payment flows could act in real time to identify fraud, perform ‘self-healing’ actions if systems are down or experiencing latency. AGIs could also help customers proactively manage their finances. For sure, Generative AI could enhance this further.

Microsoft: A Unified Platform for Building Generative AI Agents

Microsoft has developed a comprehensive platform for creating GenAI agents, offering three main entry points:

• Microsoft 365 (M365) Copilot
• Copilot Studio
• Azure AI Foundry

This structure enables different user profiles—users, makers, and developers—to create and deploy agents with varying levels of complexity.

An Integrated Ecosystem with Three Entry Points

1. Microsoft 365 Copilot: AI for End Users

If a user starts with Microsoft 365 Copilot, they can create simple agents using natural language without requiring technical knowledge. This functionality is already part of Copilot Studio, but the experience remains seamless and intuitive. For those who want more customization, agents can be extended with advanced functionalities through Copilot Studio.

2. Copilot Studio: Agent Creation and Extension

Makers and developers can start in Copilot Studio to build customized assistants or agents and publish them internally via Microsoft 365 Copilot or externally through the web and other channels. Additionally, they can leverage advanced capabilities from Azure AI Foundry to enhance these agents, creating more sophisticated solutions.

3. Azure AI Foundry: Flexibility for Developers

Developers seeking maximum flexibility can start with Azure AI Foundry, building agents without implementation limits. These agents can be published externally or integrated into internal solutions using Copilot Studio or the M365 Agents SDK.

A Framework for Different Creator Profiles

This platform adapts to different agent creator profiles:

• Users who intuitively create agents through Microsoft 365 Copilot.
• Makers who configure and extend capabilities using Copilot Studio.
• Developers who build advanced solutions with Azure AI Foundry.

Thanks to the seamless integration between these tools, the boundaries between these categories are blurred, allowing solutions to incorporate one, two, or all three components without friction.

Common Questions and Answers

With so many options available, it's natural for customers and partners to have a lot of questions. Today, I’d like to share my answers to the top three most common ones:

Can I rebuild Copilot for M365 using Copilot Studio?

No. While you can create your own copilot grounded in SharePoint, OneDrive, local files, and more using the Microsoft Graph, it does not offer the full breadth of features found in M365 app experiences, such as "build a PPT from this Word doc" or Semantic Index.

What is Microsoft’s approach to Responsible AI?

Microsoft follows a robust Responsible AI framework. Copilot Studio specifically utilizes Azure OpenAI Service within its generative AI features. Your data remains your data and is not used to train the underlying foundational model. Additionally, features like Generative Answers include built-in Responsible AI checks for query ingestion and output.

When should I use standalone Copilot Studio vs. Copilot Studio in Copilot for M365?

• Use Copilot Studio within M365 Copilot when you want to extend Copilot with plugins to third-party data, automate workflows, or gain more control over responses.
• Use standalone Copilot Studio when you need to build an independent copilot experience for employees (including those without M365 Copilot licenses) or for external-facing interactions via websites, SharePoint, social media, and more.
• Both options support the 1,200+ Power Platform connectors.

Conclusion

In conclusion, GenAI agents and multi-agent systems are revolutionizing industries by automating complex tasks, boosting productivity, and fostering innovation. As these technologies progress, they will play a crucial role in shaping the future of work and technology.

This evolution marks a fundamental shift in how organizations approach automation, customer engagement, and operational efficiency. Microsoft’s unified platform empowers users, makers, and developers to create intelligent agents customized to their needs, all while ensuring the responsible and effective use of AI. As AI technology advances, businesses that embrace these tools will be well-positioned to innovate and lead in the digital era.

Would you like to explore any specific aspects of GenAI agents or multi-agent systems in more detail?