Build AI Agents with Zero Code Using Azure Logic Apps

Key Takeaways

[music] Hello and welcome to this episode of Sip and Sync with Azure. In this one, we're talking about AI agents and how to build them in a noode fashion. And I've got here from our product team, Rohita. Welcome to the show. >> Thank you. Thank you for having me. It's great to be here. >> Awesome. Well, tell us a little bit about what do you do at Microsoft? Yeah, I'm an engineering manager with Azure logic apps and and which has been an integration platform for for over a decade now we have been in this business >> and with agentic AI now like so we are we evolving the platform to be more agentic and make things super simple for users >> and so you're working on something you just you just oversimplified it but let me tell you the product is extremely amazing you're working on something that is allowing people to build drag and drop AI agents without writing any code and I think that is extremely powerful. >> How are you doing that? >> Exactly. Exactly. So we we are a platform a local code no good platform with in the integration business and we have built over 1,400 connectors uh from past services in Azure to S services like Microsoft Office service now third party services and legacy system like IBM mainframes and and that means we take care of handling different authentications uh and then taking care of different protocol details different interaction patterns. As a developer, you just pick the connector and we take care of the plumbing. >> And the connectors that you're talking about, these are the real connectors where an enterprise application comes together. Right. So when when we're thinking about AI agents, one of the biggest complexities of building an AI agent is actually not the AI part, right? >> No. No. The the biggest complexity I have seen customers running into is is running these agents in the system that they already running. >> Exactly. And what that mean is handling different authentication schemes and and and taking care of uh data governance and and compliance and then navigating different network security boundaries. So that's very hard problem to solve. >> Hard problem to solve and and you're doing that in a visual way without using much of code with all these connectors. Why don't we see it? >> Exactly. So let me show you what I got here. This is an an agented capbot that I have built for a picture. It says uh coffee retailer. We call it fourth coffee. Yeah. >> And they have built it for their account managers and account managers use it to like get information about the customers that they are managing. So why don't I show it here? Like for people who are familiar with logic app, this is the familiar logic app designer. And all you do is drag and drop here. And and for people who need to write code, you can write code too. Um, and this is an agent as you can see here. Like if I click on the agent, you can see this is using GPT41 model. And if I scroll down a bit, you can see the the the prompt, the system instructions. And this has been given bunch of tools. And these are all drag and drop from a library of connectors that we have. >> And the first one is a AI search connector. And it gives this agent access to a knowledge base that we have built collecting all the data about the customers from support cases to orders and and so on. And then I have already also given it access to a service now account the incident management system for coffee >> and uh access to Microsoft teams. Um so these use different authentication schemes. >> So I've got I've got a question there. So when we talk about AI agents there are I would say like three components to it. There's the tools component which is what you just talked about the integration with um with your embeddings the search and then the integration here with with your service now and and all these other um sources like teams and but there there's the second component which is um which is your memory as well and the data. How are we handling some of that in this? >> So, as a developer, you have multiple option to choose. Y >> again, it's configuration. You don't have to write code like we have short-term memory option and long-term memory options. As a user, you pick what you want. >> Okay, great. All right. So, keep going. Let's see one of the tools. >> So, this is a conversational agent. You get a built-in chat client. Uh so, if I click on here, so let's see this in action. Um, you get this link here to a chat client which is already built in and and you and your users are ready to get started as soon as the agent is built. >> And if I click in here, you get to and this chat client by the way is is embeddible into your own website as an iframe >> and it signs you in based on the oath configuration you have set on your app. And in this case, I logged in. is configured to for the Microsoft tenant. So it lets me log in and now you're ready to chat with your agent. So in this case I'm going to be the the account manager and one of my customers Contoso Cafe. >> Yeah. >> Uh has an order that's getting delayed and they have open a support ticket. So they have escalated to me to see if I can help. So I'm going to simply ask the chat and see if you can get me the details about it. Uh, and what's going to happen now is it's it's going to talk to the AI search knowledge base and see if it can find any information about what's going on. So, okay, as you can see, it was able to locate the support ticket and then looks like this is an order that's exceeding the credit limit. So, the risk department is is putting a hold. Uh so and and I'm going to go in and ask uh see if I can create an incident on on this on the risk team to see if they can help out. Okay. So there's a typo. Let me fix that. Maybe it'll fit it fix it on its own. >> Yeah. >> All right. So now what's going to happen is it's going to invoke the service now connector. Remember we have that connect in there and so I have the incident number. So I'm going to actually also send a notification to the wrist team that I open this ticket so they can take a look at this. So >> and that's through teams connection >> that's through teams connects. See like we on that. So now, so I'm going to say, can you send a notification to the rest here? So I'm expecting it to uh now as you can see it pops me up to authenticate because I have configured teams. Yeah. with the saying hey like you need to invoke the teams connector in my permission context. So I need to sign in here and let the agent talk to teams on my behalf. So so I got that out. So now I'm expecting it to call teams and so already call teams and then I can actually click here and then see in teams There you go. There's notification >> notification and did with my own account. >> That's awesome. Can we go back to the screen again where we saw the architecture or what's going on in >> Yeah. So now another thing you can do is if you go back as a developer and an operator of this agent, I want to know like if something goes wrong. >> Exactly. That's where I was going because things go wrong. >> Exactly. So now if I go to here I have the full list of the conversation this team is having with my users and you can go here and and take a look at all the conversation being exchanged and and for example in this particular one I can see all the message exchanges that happen. Not only that, you can see the tools that got invoked in the for example here. Let's take a look at the one that it did with the um so service now here. If I click on this link, okay, let's see if we can get to this one here. >> Incident. Yeah, this one. >> Yeah. So, there it is. >> Yeah, >> you've got >> Oh, there it is. So create record and then and then I can if I click on here I can see actual input output if there was an error tool failure you can see all of that >> in the monitoring we built into your agent itself >> full transparency. So where I see value of this is for enterprises that have maybe smaller teams or teams that are just getting started with AI agents. You can start off with no code low code experience like this still get all the visibility and the shooting aspects of of building and then when they are ready to even like take this further and maybe become pro code make this application proc code eventually >> they can do that from there. >> Yeah. Yeah. lots of extensibility points built into logic apps. You can bring your own code. We let people write code in multiple different languages. C, JavaScript, and we recently started supporting Python. So when you hit a wall, >> that's not a dead end. You can fall back. And >> that's the challenge with no code, low code applications. It's like if you suddenly want to do something that's not supported let's say then most enterprises want that extensibility as it's good that that provides that >> but um anything else you want to cover >> uh so I think logic app with the setup connectivity bringing and and and so far the challenge most time consuming part for many of our developers has been putting this uh logic together >> we give the connectors drag and drop but still they try to implement the logic of of automating business processes, >> right? >> But with agentic AI, the needs goes away. So the all of that logic can now become a simple prompt. >> Yeah. >> And you throw in the right set of tools and things just work. So with the new capabilities we have built in like you get all the connectivity and now it we are also now we have taken out the need for doing all this complex logic leave it to the AI >> right exactly and so are you able to um create this flow without actually dragging and dropping or >> no it is all drag and drop and and and the other part is we also do support like any multi-agent orchestration there The logic app is an orchestration platform by itself. So there are lots of built-in available orchestration patterns for like doing fan out, fan in, branching. So all of that now you can combine with the agents. You can put multiple agents interacting with each other in in numerous ways. >> Yeah, you can create we just saw a single um agent application with bunch of tools but you can imagine a multi- aent application with it as well. Awesome. That was great. Thank you for walking me through that demo. I'm sure you all enjoyed it, too. And if you are on this journey of building AI agents and don't know where to begin, this sounds like the perfect spot to be in. We'll put some links in the description below for you to get started. And I'll see you next time. [music]