Microsoft Just Showed Us How To Use New AI Agents...

Is, let's say you wanted to build a field service agent that you want to interface into copilot. All you gotta do is give it a system prompt, tell it that, hey, I want you to be a field service agent. Point it to a SharePoint site where there's a bunch of documents related to field service, add to it even additional data sources, in this case their dynamics as the system of record for field service. And you have an output which is essentially a field service agent, which now you can talk to and have a conversation with, just like you would with any other regular copilot conversation, right? So that simplicity, it's kind of like back in the day we just created an Excel spreadsheet. It's no more mystical than that. Just like how you could create an Excel spreadsheet that was a forecast. You can now create AI agents using a low code, no code tool like Copilot Studio put it into Copilot. You can even think of these as the new form of applications. And that.

So that was Satya Nadella at the AI Tour in London showing you all just how easy it is to create AI agents. Now, I think most people don't understand the entire AI stack that Satya Nadella talks about. He talks about Copilot and how this entire thing is going to be integrated with AI. And I think this talk is one of the very best ones that we get to see how the future of work is. We know that Microsoft is integrated into tons of computers, even if you're not on the Mac. And this is going to be a really integral piece of information because this is largely how the future of work is going to be on our personal computers. So I'm going to break down this keynote speech giving you guys the details, but this is one of the most insightful things that didn't get enough recognition. So firstly, Satya Nadella actually speaks about how these agents are going to have reasoning and planning, some of them are going to have memory, and how this brings together the agentic experience for not just work, but for a variety of different experiences that we're going to be.

Having in the future to make sense of it. And then lastly, you can feed it more context, more memory. So you put all these three things together, you're building out a very rich AI or agentic world in which you are going to have these AIs or agents, right? There will be some AIs and agents that are personal agents. There will be things that will work in the context of a team, in the context of an organization. Or business process or even cross organization. So this rich tapestry of AI agents that augment everything else that we built, right? So that's the other part. The entire digital infrastructure and tools that we today have get augmented in this agentic world with all these AI agents that we build using the scaling laws as the underlying force. Now, of course, so then this is.

Where we actually get Satya Nadella talking about how this is going to be the UI for AI, how copilot is basically going to be the interface for how AI interacts and integrates with your computer. I think that's the best way to put it because it lets people understand how the future of agentic AI will be when it manages to merge with your computer and other essential applications.

Pilot to us, you should think of copilot as the UI for AI. That's the simplest way I think about. We then have the Copilot and AI stack. So to be able to, for you to build your own AIs and AI agents and copilots, we have a full stack. And then lastly this new set of devices, which are these Copilot devices. And so I want to talk about each of these platforms starting with Copilot.

Now, as I said, if you start.

With this idea that this rich agentic world ultimately does need to meet us and we need to meet it, that means you need a UI interface, right? Just like the PC or the phone was the user interface or the apps on a phone or a PC with the interface to essentially digital technology. These Copilots or. And Copilot is the UI for all of this AI, right? Even in a world where there are a lot of agents that are working autonomously, they do need to raise exceptions, get permissions from us. And the question is, how does that happen? It happens through this new organizing layer for how in particular work gets done. In fact, work, work artifact and workflow is going to change. A great example of this is just a couple. A month ago we launched something called Pages. Just like say back in the day, in the 90s, we launched Excel or Word, which were basically editors, to create new artifacts. Pages is the first, I would say, user experience to create new AI first artifacts, right? I can search the web or my work for retrieving information and then I can put it into Pages and it's a document that I can then share across the organization. And I can work with AI and humans. In fact, I sort of say the metaphor I use is, I think with AI and work with my colleagues at work, right? That's the new Workflow, what is the previous workflow? I thought on my own I created artifacts and I shared it across the organization and collaborated. But now I not only have a cognitive amplifier effectively with AI where I do my work and then I create artifacts and I collaborate with my colleagues in order to get things done. And so that's really the beginning of this copilot era, where it's just not about a chat interface, but it shows how chat is just one modality of being able to retrieve information, but it does lead to more sophisticated workflows and collaboration. Now you extend. So the other thing is that this is not just about any particular artifact editor or workflow we created, but you can extend Copilot with any agent you build. In fact, Copilot Studio is a low code, no code way for you to be able to build agents. And these agents are really grounded in a rich set of data sources, starting with. In fact, the most important database in most organizations is the database that contains all, all your office information, right? Who works for whom? Who are my colleagues on this project? What documents are there related to a particular team or a project? What is the relationships between all of these documents and people and projects? All of that, and all the emails you've had, teams, conversations you've had, that's all in fact, in a first class database called the graph or the substrate that is now exposed through a graph in M3.

Now, in addition, for those of you who think that building AI agents might be difficult in the future, Satya Nadella shows us exactly how quickly it's going to go from prototype to working agent that you can use within your workspace that actually works really effectively and can do tasks on your behalf, is, let's.

Say you wanted to build a field service agent that you want to interface into copilot. All you gotta do is give it a system prompt, tell it that, hey, I want you to be a field service agent. Point it to a SharePoint site where there's a bunch of documents related to field service. Add to it even additional data sources, in this case their dynamics as the system of record for field service. And you have an output which is essentially a field service agent, which now you can talk to and have a conversation with, just like you would with any other regular copilot conversation, right? So that simplicity, it's kind of like back in the day we just created an Excel space spreadsheet. It's no more mystical than that. Just like how you could create an Excel spreadsheet that was a forecast, you can now create AI agents using a low code, no code tool like Copilot Studio put it into Copilot. You can even think of these as the new form of applications and that.

This is where we actually get the OpenAI demo of the agent that you saw in an earlier video that I posted. So this one is a lot more detailed because it is a hands on demo and this has a real working example. So I'm going to let them take it away for the six minutes that they have. But I think this is probably one of the most important pieces of content you'll watch because you truly will see how the future is changing right in front of you. Please don't miss this kind of content because I think the future is going to come quicker than you can imagine. And not paying attention to exactly how the world is changing is a grave mistake. But nonetheless, this is of course, these Microsoft Copilot agents.

Let's take a look. It all starts with an incoming email from a prospective client, much like you see on the screen right here. Now, previously they had had people on the back end essentially receiving these emails, parsing through them and figuring out what to do next, who should it be routed to, what expertise did they have in the firm. But this is where the autonomous agent comes in. Now an email comes in and the agent springs into action. What you see here is that it will begin to parse out the email, moving through the ambiguity of human language to, for instance, find out what the engagement's about, to check the engagement history, to also map it to their industry standard terms, and then finally to try and find the right person to take the next step within the firm. With all of this information in hand, the agent then goes about writing an email that takes all of this information and summarizes it for the receiving partner. And what you see on the screen is exactly that. In comes a whole bunch of human written email. The agent processes it, summarizes it, and sends it to the right partner in the firm to take that very next step. Now, it's worth pausing for just a moment here to reflect on what you're seeing. It happens so fast you might miss it. But essentially this agent has been given a loose set of instructions, kind of like you would to a human. And it deals with all of the messiness of human communication, figuring out what the right next touch point is for the customer. Now this is magic, but it's only half of the magic, because now we're going to go behind the scenes to see how easy it is to actually create an agent. Just like this. For this we will move over into Copilot Studio. Here you see that we have programmed up with McKinsey the agent, but not using a sophisticated programming language, instead using natural language, the same way that you would tell a colleague to get ready to do this task. You also see that what makes this agent autonomous is that we can set what's called a trigger. In this case, the trigger is set to watch an email address and to react immediately when an email comes in. But in fact, you can set it to look for events across a whole wide range of systems sitting there working for you 24 7, waiting for an event to come that gets it going. You also, just like a regular human colleague, add knowledge. Here we see a word documents, a SharePoint site and a database about engagements. But of course you can add additional knowledge sources. That includes line of business systems like SAP or ServiceNow or even databases. And finally, to finish up what you give this agent to do its work, you give it a set of actions. And we saw those in the flow. These are actions that include things like pulling out the relevant information or summarizing what a human has written. All of this together makes the agent powerful because it can deal again with all of that ambiguity that a human throws at. Now, what we saw was one email coming in about one new client engagement. But the exciting thing here is that this scales. How does it scale? Well, to see that, we'll go over to the activity pane where we can look at the long list of engagements that it's working on. Zooming in up top, for instance, we can see that it's worked on over 1300 engagements and there are 33 in progress. If we want more details, we can go into the analytics tab. What this means is that this agent is always working on behalf of the firm and that's very exciting for us. Now from here we also see that although the agent's amazing, it does sometimes need some human help. So we're going to jump into a case the second from the top here where we will see that it gets a little bit stuck as you look. It's gone through those steps that we saw previously, but it's stuck here at the one where it's looking for the partner. And if we zoom in, we can see why. Here, for instance, we see that it's picked the right partner, but that partner has now left the firm. It has an instruction that says if that's true, it needs to escalate to a human manager to give it someone else to go to now to see what that looks like we're going to switch over to copilot and see that interface with that human manager here at the bottom right you will see that a notification pops up in copilot then the manager gets all the information he or she needs and can provide the right person to route the email to back at the ranch going back to our agent we can see that it takes that information and fills out what it needs to do. Now we're excited about this because of the business value it can drive. McKinsey in its trials has shown that it can reduce lead time by 10, 90% reducing administrative overhead by 30% and as you look at this list what we envision is an orchestration layer just a bunch of agents that can be out there helping individuals, teams and entire functions to streamline and automate their processes no matter what industry they're in they're so easy to make. Anyone can do it. You design and set these copilots out to work in copilot studio you interact with.