Transcription
Perhaps you've seen the Copilot Wave 2 presentation with Satya Nadella and Jared Spataro. They announced a ton of new features for end users and also introduced this new concept of copilot agents. These are specialized chat agents for Microsoft 365. We are going to put a link for that video in the description box.
What you might not know is that just as an end user can create an agent in Copilot or a low code developer can create one in Copilot Studio, there are code-first approaches to building agents. Traditional developers can build and extend these agents with tools that are familiar to them.
In this video, we'll cover some of the major concepts and introduce videos, hands-on labs, and other materials to help you come up to speed. If you've been following this topic at Microsoft Build last spring or on community calls over the summer, you may have gotten an idea of what's coming. However, be advised that the names have changed.
Copilot extensions are now called agents, and custom engine Co-pilots and declarative Co-pilots are now called agents as well. Plug insurance are still plug insurance, but are now part of a broader category of copilot connectors.
But let's take a step back and explain Copilot agents from the beginning, from a code-first perspective and using the new terminology. This is with chat. It appears in Microsoft Teams, Outlook, Excel, PowerPoint, office.com, and more. It's a chat service similar to what you're used to in Teams. However, it has a number of additional features such as response streaming, feedback, and reference marks to make the interaction richer.
If you open the right flyout, you'll see a number of options and recent chats. This is where agents appear. Agents are tailored AI chat experiences that work in this chat user interface. Agents can come from several places.
The first option is that end users can build them from right inside the biz chat user interface. These are what we call declarative agents because they don't contain any imperative code, just information that Microsoft 365 Copilot can use to bring the agent to life. Users can enter instructions and select specific knowledge such as SharePoint sites or specific graph connectors to be used in the agent.
In addition, low code developers can build agents using Copilot Studio. This is an extensive topic on its own, so I'll put a link in the show notes where you can get more details on that. Finally, and the topic of this video series, traditional developers can build agents with code in Visual Studio Code, Visual Studio, or really any development environment that supports web services, JSON files, and zip packages.
These are packaged the same as Teams or Microsoft 365 applications and can be distributed in the Enterprise App Catalog or via the Microsoft 365 App Store. While there are so many ways to build agents, users don't need to concern themselves with those details. All agents appear in the same place in Copilot.
This is called the immersive experience because the user is in a branded chat directly with the agent. The other option, and this is a new one, is the in-context experience where users can @ an agent while using biz chat, even in a group context. You might have seen this slide before at Build last spring or even at Ignite last November.
The idea is that you can adopt Copilot, that is, use it as is, or you can extend Microsoft 365 Copilot, or you can build your own. These are still the main choices, but it's more refined now, and as I mentioned, some of the names have changed.
Microsoft 365 Copilot comes already loaded with orchestration, language models, and grounding in Microsoft 365 content. You can build on top of all that, thus extending Microsoft 365 Copilot by writing declarative agents that basically tell Copilot what to do in natural language and what information to include in a tailored chat experience.
You can also write plug insurance which call external REST services so that you can integrate with your own applications and run custom code as needed inside of web services. If you choose the build path, you'll be writing your own chatbot. The only part of Microsoft 365 Copilot you'll be using is the Biz Chat user interface.
This gives you complete control to use any large language model, orchestration, and other tools that you wish. For professional developers, an SDK called Teams AI Library allows you to fully interact with users via Biz Chat, so you can include identical reference links, feedback, and other UI details.
Now that you've seen both of the major options for building Copilot agents, we'd like to invite you to dive more deeply. We've prepared videos specifically on custom engine agents and declarative agents with tons of demos and code reviews. Links to these videos are in the show notes, so please check them out.
In addition, we're pleased to announce a virtual workshop called Copilot Developer Camp, which will guide you through hands-on labs building both types of agents. You can follow a path E where E is for extend and build declarative agents, including one that uses an API plug-in to access a REST API that's secured by entry ID.
And you can also follow path B where B is for build and build a custom engine agent using Azure AI Studio and Teams AI Library. You'll learn how to train the model with your own data and expose it as a copilot agent.
Browse to https://colonaka.ms/copilotDevCamp to check out all of these labs. Please join us so we will get you off to the right start. Thanks for watching and subscribe to our channel for more copilot development videos. Bye.