Rubber Duck Thursdays: Building an AI agent app

Añadido: 2026-05-09

[00:05:10]Hi.

[00:05:12]Hey everyone. Hello. Hello.

[00:05:16]How's everyone doing this morning? Are we doing good? I am going to I was asking where everyone is watching from.

[00:05:25]We did this last time and I really liked it. Just knowing where where people are watching from would be great. My name is Marlene and uh if you're here last week, we chatted a little bit. Uh I think I see again some people from India. Hi uh Manudon from India. It's nice to see you.

[00:05:48]Uh, who else do we have on the screen?

[00:05:51]We have Edwin, who's uh from Nigeria, which is great. I've been to Nigeria as well a little bit a long time ago, but I I was in Nigeria, so it's good to see you, Edwin.

[00:06:05]Hi. Uh, let me know in the chat where you are from right now. I'm actually usually I am in London but today I am actually in Germany in Cologne for a conference.

[00:06:22]There is a conference called the Azure uh cloud summit and so I'm in Cologne Germany for that. Uh, so this is where I'm actually calling you from and this is why my background if you remember from last time it was a different background but now it's because I'm in Cologne in Germany. So, oh I do see some people. Hi. Ah, I do see Stefan is also in Germany. Hi Stefan. Good to see you.

[00:06:51]Which part of Germany are you are you in right now? I'm in Cologne today. Uh, so actually very sunny day in Cologne which I like. Hi from Bangladesh. Nice to see you from Bangladesh. Great to see you on the stream. Victor is calling in from Spain. Nice to see you. How's the weather in Spain? This is the thing is that I love warm weather and I I sometimes in London I I don't love the weather there. Barus is cold. Good to see you Roman from Barus.

[00:07:25]And we have from Malaysia, Chai Idi, it's good to see you from Malaysia. I've never been to Malaysia, so it's it's very cool to see you there. Uh, we do have someone from Brazil, which is very cool as well. Um, but yeah, like I mentioned, today we have Yeah, thank you. I am in uh in Germany today, but I still wanted us to do some co-working today. I do see someone from Bangladesh. Nice to see you. Um I do wanted us to do some some co-working and kind of look at some of the things that I'm actually going to be talking about at this conference. So, I mentioned that I am at Let me show you the conference I'm at just so that uh let me find it online. European Cloud Summit is everyone does does every has everyone tried Azure before? This is like Microsoft's other product outside of GitHub as well. It's like Azure cloud was pretty popular and this is let me open I'm in simple browser and if you do this you can open a new tab and you can look at websites in simple browser which I really like.

[00:08:47]So, this is the this is the uh event that I'm at and it's the European AI Cloud Summit. We're in Cologne.

[00:08:59]And if we go, these are all my co-workers that are going to be speaking at this event. We also have Build coming up. By the way, if you haven't registered for Build, let's actually do that. Let's open a new tab and let's look for Micros I think I have to paste a URL in here actually but Microsoft build is very cool and that's coming up and that should be very interesting but uh for this conference I'm going to be speaking about building agents with lang chain and I've been using co-pilot GitHub copilot to help me with my talk and also to build the app that I'm going to be using. So I think one of the things that I did want to chat about is how many of you are actually building AI applications like are you using um are you actually deploying anything with your apps? I don't know if maybe let me zoom in a little bit to the application that I am going to be showing or we can actually just look at it here as well.

[00:10:02]Um, let's see if we can Sometimes I don't know how to how to increase the size of my screen when I'm sharing uh so that it feel fits correctly here when you're sharing on but anyway um maybe if I do that it'll do it. Yes, I think so. Um but this is the application that I have been building and basically something we talked about last week was about AI agents and how most AI agents kind of operate the same way. They have this sort of agentic loop and then when you are using that agentic loop you can um so the aentic loop let me actually go to it. I think I have it in these slides as well. Oh, um, if we go back here.

[00:11:02]Oh, here we have it. Yeah, we have we saw this last week already, but I talked about how with the Gentic loop, most of the AI agents, so if you're using GitHub Copilot, for example, GitHub Copilot is a coding agent. So it is going to operate like any other coding agent would. And we talked about this last week that you can use or optimize your agents so they perform well. So you will as a user you'll send in a query to copilot and then copilot is going to the first thing in the agentic loop is it's going to try and gather context. So you want to make sure that it has all the context it needs. Um, and then it's going to take action with tools and then you're going to validate that whatever it did or help the agent validate for itself that whatever it did was correct.

[00:11:55]And um, kind of part of what I want to show on this stream is this validation step today and probably focus a bit more on that. But one thing that I think is is super cool as well is the fact that you can build your own agents. So you don't just have to use like a pre-built uh agent like Copilot. And actually GitHub Copilot if you go under the hood and it it it works just like a normal agent. And um so what I think is super interesting is that as a developer you can actually create your own agent using this agentic loop and you have different frameworks available for you to do that.

[00:12:38]One of the frameworks that we talked about last time was Copilot SDK, which I think is really interesting. But I'm also one of the maintainers for Langchain Azure, which is a um it is one of the popular Let me actually go I'll show you my slides. I'll just go back to show you my slide so we can we can talk about it. I am Yeah, I'm one of the maintainers for Lang Azure. And so this is like an integration and if you don't know what langchain is lang chain is a code it's a framework that's similar to copilot SDK and you can use it either in Python or in JavaScript and it tries to make it easy for developers to build AI powered applications. So I'm one of the maintainers for the Python version of this for our Azure integrations. So if you want to build out an application with lang chain, you'd go ahead and you'd use something called create agent and that would help you sort of build out this application. I actually have a file here, an agent file and it should be we can see this agent.py py file and you can see in the code um and I'm just I'm using before someone is going to be like oh my gosh you maybe you're going to show the keys or something but I'm using something called enter ID and actually let me go back to the chat let me go back to the chat and also ask how many of you have um how many of you are building your own agents and if you are building your own agents Um, have you struggled with making sure that your agent or your LLM doesn't look at your environment variables? So, I actually saw a tweet about this the other day. H, and I'm going to just stop and share some comments. I see Edwin saying, "Yeah, I use Azure cloud a lot. Azure is really good if you are someone that is interested in that." Um, I see.

[00:14:48]Okay. Okay, I see Nathan saying got multiple agents running in production for multiple different companies. Voice agents, multi- aents, of course, chatbot and a and other agents. How many of us are actually that's amazing and I think this is something that I feel you know a lot of people are feel that maybe are a little bit nervous about where the industry is going and there's lots of different like takes. I don't know if you've seen like some of these takes online where people are say have been saying that it's over for developers and you know things are not going to work out for us and they're going to you know AI is going to replace developers and I actually don't think that is really going to be true in the in at least for now from what we're seeing I don't think that's necessarily going to be true and I think one of the reasons for this is Because now that we h we know that AI is a thing and in lots of different industries, they're wanting us to build AI assistants for them. Whether that is going to be an AI assistant like Nathan is saying here, which is like a voice agent for a different company. So maybe I'm working for what's a good I don't know what's a good company. That's an example. Um maybe I'm working for Microsoft for example and at Microsoft I am going for a conference and I want to build a conference agent and the conference agent I want to build I want to have like someone use voice for that to interact with the agent and have the agent actually like do stuff based off of my voice or maybe tell me about different sessions based of my my voice.

[00:16:38]And one of the cool things is that we then can have these AI agent. We can build these agents for different companies by understanding how these agentic loops work and then using things like copilot SDK or using things like lang chain to be able to build AI into our applications. And so today, that's kind of what I want to focus on is I want to talk about what does it actually mean to build AI agents into an application like Nathan is saying for a different company. Maybe uh I do see someone is saying um okay someone is saying I am developing CAD agents. What is that? What are CAD agents? Let me Google that. What? Bing search. Sorry. Bing search. What is a CA agent?

[00:17:35]A CAD agent is What does that mean? H multimodal CAD agent. It's it's saying that this Okay, let me put it in. Let me put the description here so we can read a little bit uh a bit more about what CAD means or I don't Hang on. I don't want to.

[00:18:03]Okay. Okay. Let's let's just go to this one and I'm going to post the thing in the tab. Let's make a new tab for this.

[00:18:14]And let's see if we can or we'll replace this one.

[00:18:20]So, someone said they're building CAD agents. It says production ready CAD designs. An AI agent that builds production ready CAD designs directly into your browser. I still don't know what CAD means. What does that mean? AI build CAD.

[00:18:39]I guess it's like Blender stuff. Please in the chat uh let us know what what does CAD mean uh as in design. Ah okay.

[00:18:50]Okay. Yes. Thank you. Computer aided design. That's cool. I think that's pretty cool. Computer aided design.

[00:18:59]Someone is saying 3D graphics modeling.

[00:19:03]That's very cool. Wow. So you would be building like a an agent that can actually 3D model stuff for you. I think that's super cool. And okay, someone is uh Adion is saying it's an architectural language.

[00:19:23]Computer agent design again is what we are getting from here with 3D modeling.

[00:19:28]I didn't know that this was a thing and it's actually a very cool use case for building agents. So I think that is amazing.

[00:19:37]Uh I do see Wayne said he just soft launched a website monitor app open beta testing. That's very cool. Um yeah I think let's see uh okay I do see Alex is saying what do you mean by building AI agents like LM chat or autonomous code generation or testing review? I definitely have a look at lang first time you hear about it. Okay. So I mean about like similar to what we saw before where we heard from Nathan who Nathan said that he is building multiple agents running in production for different companies. So I'm going to show you an example in my VS code. This is an application that I have been building and that I'm actually going to present on today. So this is a sales agent, right? So the goal of this agent is that we want to automate some processes. So say for example, I want to start a business. And in this specific example that I'm going to be talking about, I'm showing a website where I've built this AI agent which is going to help me with sales.

[00:20:58]So I don't like selling talking to people and selling stuff. I like streams talking about code, but I don't love selling stuff. And so I would build a sales agent that can talk to customers or potential customers that have questions about my app. So here it says, "Welcome to Zava, your DIY sales supplies partner for Washington State pros and home owners. I am Zava sales concierge. I can help you compare pricing plans, see customer stories like yours, answer questions, or book a call.

[00:21:34]So, let's let's take a look at something. Let's take a look at an example. So, here I've put in some example um prompt. So, I would put a customer could come to my site and again this is a website an agent that's helping for a DIY supplies store called Zava. So this person is saying,"Hey, I run a 25 property management company. Do you work with teams like mine?" And so I've built this agent that can take that query and it's using under the hood, it's using lang. So the first thing it's going to do, it's going to try to understand, it's going to do a tool call to understand the user's intent. So what does the user actually want? And then it's going to search. It's noticed that okay the user wants some examples and see to see if that if we work with companies that are similar to them and then the agent is going to look through I've created an MCP server with a postgris database so the agent can access that database to look at different data like case studies of other clients. So here we answer and this is a debugging window. So we can look at what the agent is doing under the hood with each call. So the person has sent this question and then the response is absolutely let me pull up a couple of examples that look closest to your setup. Yes, we do. The closest match is this company. And then it also links to the case study if the person wants to read more. So then the person can actually go ahead and book a call.

[00:23:10]So I could do something like I could say great uh connect me to an account executive to book a meeting and I haven't yet like we have something at Microsoft called work IQ where you can actually automate this in teams in in Outlook. So, Work IQ like allows you to also provide your agents with tools that can connect directly to your calendar or to uh whatever it is within the M365 suite. And so, here our agent is going to respond. It's going to see the intent. It's going to say, "What email should I have the a the uh the executive use? What time zone etc to schedule the meeting?" And then I would want the agent to then use this tool to then set up a meeting between me and the sales exec.

[00:24:09]So, so this is what we are. This is kind of how an agent would work. And uh let me see what what people are saying in the chat.

[00:24:25]Okay, I do see Roman saying,"I just now built an AI recruiter." Again, all of these things, the idea is to take the AI that we would use with like GitHub Copilot, for example, and actually put it in an app that automates away some of the work that you don't want to do. So, if you're a recruiter and you want an assistant, so similar to how we have our coding assistants, then you would build an AI agent for that as well.

[00:24:55]Um, let's see what other people are saying. Okay, lots of people talking about um, lots of people. Okay. Okay, great.

[00:25:07]Someone shared a link for CAD. Thank you, Todd, for the link. Uh, Pontis said yes, she's on track. I did understand. I wasn't understanding what CAD was before, but I now understand that. And Angel also gave us an example of what CAD is. So let me go back as well to that CAD example that we saw on the screen which is super cool. So here like this example by the way I did not know about CAD before this but we are seeing from thank you for educating me on this topic. Angelo is saying CAD agents are AIdriven autonomous software designed to interact directly with CAD environments.

[00:25:53]So these are for designing. So it's computer AED design. And so this would be if you wanted to build out an application for a designer for example.

[00:26:05]And here we can see that it says in this specific app, shout out to the B build CAD AI people who I have only heard about today. And it says describe it. The AI agent builds it. So then it has this environment where you can design different things in 3D. Okay. So this is very cool and uh I do want to talk about GitHub copilot. have copiloted SDK and let's see if we can use this as an example of how we can use the SDK.

[00:26:48]And if I go here, I'm going to uh uh one second. Give me a sec.

[00:27:03]So now if we go here and if we take a look for example we talked about the CAD agent and this is an example of another thing you can use. So I talked about how in the code right now I'm using lang chain to build out this agent and add it to an application and there's different ways that you can do this. Lang chain is one of these ways and another way would be to use the GitHub copilot SDK. So regardless of the language you're using um you can use the copilot SDK to be able to do this. And so you'll build out your your the your app and then you'll plug in with different agents. Maybe I'll ask how many of us what computer what um uh languages are we using uh to code? Are we using Python? Are we using Go.NET Java. Let's see. Oh, okay.

[00:28:14]Alex is asking why it takes so long.

[00:28:17]I need to probably speed it up myself and I'm using Copilot and VS Code to help me speed it up. But at the same time as well, it does go and search for things. So it's going to the first thing it's going to do is it's going to figure out what the user needs and in this case it's taking some time or it was taking some time because it is actually quering a Postgress database. You don't really have to use Postgress and that would make it faster but just for the example that I'm using I'm using Postgress in this case.

[00:28:52]Uh okay.

[00:28:56]Uh I'm seeing someone saying I I don't like Zava. I like Zro Kent. I think Zava is uh I think Zava is supposed to be like a a brand for something else, but Microsoft was using the Zava brand for a while and I don't know people some people don't like it. Yes, I mentioned Copilot. What languages are people um using? Yes, Roman is using Python.

[00:29:26]Okay, Alex is using Rust, JavaScript, and Python. I think that's I'm personally a Python person. And so that is like one of the things we could use.

[00:29:38]So if we go back to our our uh if we go back for example here to GitHub and we look at the cookbook for Python, it has lots of different options for things to do. So actually I want us to see just the if we go to Python and we click here it will have like exactly the steps to get started with Copilot CLI and Python and GitHub copilot. So this is the code that you need to use. This is the code to get started to actually create um a version of the SDK in your code with GitHub Copilot. So, you know how here we can we can just say hello to GitHub copilot and then it's going to start a session or at the beginning of a session it'll start the session and then it's right now it's telling me information about stuff that I've just done because it's it's already part of the session. But if I wanted to create a GitHub copilot session like I already have in VS Code in my code, this is the code that I would use to do that in Python. So I would go ahead and the first thing I would do is I'm going to I was going to I go ahead and install the SDK. Um I actually think they should explicitly say what the SDK is but um the SDK is co GitHub copilot SDK on Pippi or with UV and you can also just come to this repository and install it directly here as well. Um but once you do you can import the copilot client and that copilot client is going to allow you to connect to different models. So here for example if we want to connect to G uh to GPT5 we would say create session. So just as how we would create a new session in VS code for example. So here I can you know maybe create a new chat or a new copilot CLI session. This is how we do that in the code as well.

[00:31:55]And then after that this is using async.

[00:31:58]So, uh, it's allowing Python to be a bit more efficient that way. But then we would send to that session our question.

[00:32:08]And so, we would ask, for example, what is 2 plus2? And then it would go from there. And this is a really nice way to actually build stuff into your application. Like I mentioned, I'm specifically using this is Copilot SDK to do this. Um, and there's lots of different examples here. I think there they have a manual resource management example that we could use. Um and that is one example in different tools.

[00:32:38]And then we also would have if we go back to my agent example, this would be how we would use it or initiate uh the models with lang chain.

[00:32:51]Let me go back and see if anyone has some questions. Okay, I see people are using different uh things.

[00:33:00]Okay, let me go back. Let me leave it on to the C-pilot SDK uh page so we can look at the very pretty banner while we chat. Uh someone says they're using TypeScript. It's fine to use TypeScript, by the way. I think Typescript is okay. I don't use TypeScript personally, but lots of people do.

[00:33:26]So, I'm not going to hate on you for using pi TypeScript. Please go ahead and use that. Um, someone is asking, is there a best practice or testing? What languages are better to use for? I don't think I think right now the ecosystem for AI agents is growing. So for example I showed with copilot SDK you have lots of language options. So you can use Python, you can use Go.NET, Java. The only thing I would say to keep in mind is that it's it's actually going to be depending on the language that you're using, you're going to maybe have access to more or fewer tools. So you might have some issues if you're using one ecosystem. Maybe there are not as many AI agent libraries in Java as compared to Python. I'm not sure because I'm not a Java coder. But for now, I would say, you know, TypeScript and Python are really well supported in the AI agent space. So those are the two I tend to recommend. But I also think net is also good.

[00:34:41]Um, okay. Pontis is saying, "You really never heard about it or just don't know what it is." In my opinion, there's a difference between I'm not sure what, but someone was saying they hadn't heard about Langchain or about GitHub Copilot SDK. So, um, oh no, someone said I'm early.

[00:35:07]It's 10:00 a.m. Oh no.

[00:35:11]I didn't realize that I started this stream a little bit early. No, this is not. By the way, thank you, sirly Dev.

[00:35:20]Hello from the UK. I'm good. I'm good.

[00:35:24]And I am in Germany.

[00:35:28]I'm in Germany. And this is why I didn't realize that I started this stream a bit early. I thought that Germany was an hour ahead of so you know our usual stream our usual stream time is 11:00 a.m. And so I already thought the stream I thought the stream was I thought that it was like 11:00 a.m. right now in the UK.

[00:35:57]Let's say what time is it in the UK now?

[00:36:05]Oh. Uh, yeah. I think I might be a little bit early. I think I might be a bit early. Um, I'm usually in the UK time, but it's okay. I'm a bit early because I'm in Cologne, Germany today for a conference, which is why um which is why I am streaming.

[00:36:25]Uh, someone's saying, are we getting a video of this live session so that we can re-watch and learn? Yes, all of the live sessions will stay on the GitHub YouTube channels, will stay on the LinkedIn if you're watching from LinkedIn. So, don't you can always watch this back to take a look at what I said.

[00:36:45]No worries, Alex. I'm happy to answer the question. Oh, someone said, "Very nice. I use Entra a lot, too."

[00:36:54]And Okay. Okay. Yes. Sly Dev is telling me that Germany is one hour ahead of the UK. The problem is, do you know my issue is that I think I didn't change I didn't I didn't change my settings and so I feel like I'm still in the UK, but I'm not. I'm in Germany.

[00:37:14]It's 11:30 in the UK. Thank you for letting me know I have uh at the wrong time a little bit recently, so it's okay. Some other people are joining us from different places. Uh, we see here someone's joining us from NYC. It's 6:30 in New York. Raise the roof from New York. I like New York, by the way. I have a bunch of co-workers that are based in New York. And I love it. I think it's such a fun city. Like one of the best cities in the world for sure. I also see someone is saying it is 3:00 am or are you okay to be awake at 3:00 am in the morning in California if they're California people watching the stream it's too early please go to sleep okay David is letting us know it's 1:30 p.m. in Kenya. Let me know if you're watching u from Kenya.

[00:38:19]Okay. Okay. Wait. Uh Deb is saying this is why you have to use the time zone feature when putting calendar events in your phone in C. Yeah, truly. I really should have checked. The problem is I'm like, no, this is not Yeah, I'm not great with time zone changes. And this is a problem. Even when you're building apps as developers, it's it's it can be really tough. And we also have a lot of other people greeting us. We have someone from Azar Bjan. Azaraijan. I'm not sure if I'm saying that correctly.

[00:38:53]Todd is letting us know it's 4:00 a.m.

[00:38:55]in the States. And programmers don't sleep.

[00:39:01]Okay. If you are a programmer watching this at 4:00 a.m. or 3:00 a.m., please go to sleep. Please adjust your sleep habits. One of the things that I personally feel, I don't know how other people feel about this, but I feel like just to program well, I typically have to like sleep a lot. And I feel like I really need sleep or else I'm not going to be okay. I'm gonna not be performing well. So, I don't know with Todd's I don't know how people feel about Todd's statement or the fact that other people in California are already up. Uh, someone's in Italy and it's 12:30 in Italy. That's that's good. Okay.

[00:39:51]Yes, I am in Aush. I am in Cologne just a couple of of miles away, which is great. Oh my gosh, we have someone from YouTube. Hi, Mustafa from from Egypt.

[00:40:01]Sorry. Hi. Good to see you. Good to see you. Wow, we do have a lot of people here from lots of different places. We have someone from France. We have someone from Australia, which is like 8:30 in Australia, which is amazing. I know.

[00:40:20]Morocco, it is 11:30. Okay, let's let's go back to our conversation.

[00:40:28]Someone says, "Yeah, I slept bad. My brain drifts when I sleep bad. My brain drifts so much to program. So much to programming." Yes, as programmers, let's be getting a good night's rest. Let's be sleeping and and resting as programmers so that we can we can be okay.

[00:40:49]Okay, it is 6:30 for Victor in in um uh in Maryland. That's in the in America. I'm gonna just scroll up to our co-pilot SDK logo so people can see it.

[00:41:05]But um good time. And I agree with Mustafa that if you sleep one hour, you're you're out of the loop and you're not okay. Personally, I'm not okay if I don't get to bed.

[00:41:20]Okay, Todd is giving a good excuse saying, "I went to bed at 700 p.m. Thank you for your comment about rest." I'm glad you're well rested.

[00:41:30]And uh that's a really good thing as long as you are evening out. You got to even out the the sleep. So if you sleep early, then you got to also you can wake up early and it's it's actually okay.

[00:41:44]All right, everyone. I am going to go back to what we were chatting about.

[00:41:50]Okay, someone is saying it's almost 7 PM in the Philippines. Yeah, feel free to cook some dinner while we chat. I was talking earlier about the app that I'm building and how you can in your code you can go ahead and I did see someone mention this. So I loved this statement uh about intra ID. So I don't know how many of you have um if you have used intra ID before to be able to protect your secrets. So one of the things I seen people let's see if we can find it.

[00:42:31]There's lots of comments now about what time people are sleeping.

[00:42:36]So I don't know if I'll find the comment but someone talked about ID. Let me know in the chat if you have um um let me know in the chat if you've heard of intra ID. If you haven't, Entra ID is a very cool way. So you don't have to use environment variables. So you can build an application. Something that I've seen people complaining about recently is that their environment variables or their ENV files, the agent has access to those files and maybe you don't want to give, you know, your agent access to those files. And so using something like inraid is a great option to be able to like stop your agent because you don't use uh keys in your code. And I think that's really helpful.

[00:43:27]Um, so in my code that you can see on the screen, I'm actually using enter ID and you get something called a token. So you have this token provider line here.

[00:43:37]So the first thing you do is you want to set a default credential. So you have to be logged into you have to create an Azure account and then you have to like uh log into AZD uh in the CLI. And when you do that, it is it's going to get the default credential that's in your CLI. So, GitHub Copilot SDK has a similar thing.

[00:44:00]So, you're not using keys at all and instead you'll get a token and that token is something that will refresh all the time so that it's not able to be exploited by people for example. And then this is the code that I used in lang chain. I have a main uh agent or LLM agent uh LLM connection that I'm creating. So I would go to Langchain.

[00:44:26]I'd import this chat open AAI and I can connect directly to an Azure endpoint which has GPT 5.5 mini or I'm also having a second one with GPT5 nano and then I would create the agent which looks like this in lang chain and I would give it lots of different tools um I would give it something called the state where it just monitors the conversation like we saw here it was it's going to be tracking the conversation as it goes along in that state and uh after that it also has stuff called middleware as well. So all of these are things you can add to your agent in the code. And again, this is just go if you want to do something like take your agent in copilot and actually build one that you put in the code, that should also work as well. Um, exactly like not a lot of people are using intra ID. It's a very important topic because it allows us to be able if you want to look, let me look for intra ID.

[00:45:31]Let me look for an article.

[00:45:34]Um, yeah, let me look for an article that I'll show on the screen for us to see about Entra ID.

[00:45:44]Um, let me put it in the in the thing as well. Let's do this.

[00:45:50]And let's make a new tab. I have so many tabs open. And this uh link there I'll also paste it will have some information about identity management and different ways you can use entra ID and be secure or make your app more secure. I'll also post let me post the link in the chat as well for us to look at. And again like what I said enter ID is a great way for you to just protect your application.

[00:46:20]So, um, it is it it's great.

[00:46:25]Uh, Todd is letting us know that he's heard of intra ID, but I haven't had much time to delve into it. This is very informative and your explanation was easy. Thank you, Todd. I appreciate that. And I do think you should check it out. It is something that could be a little like it's not always easy. So, you do have to put some effort into initially like setting it up. But once you have it set up, you can use the same sort of system over and over again. Or you can even have copilot. You can have copilot in VS Code or Copilot in the CLI help you install intra ID and that should that should work as well. Um so going back to our application because we don't have much time. This was the application. I'm actually about to give a talk about this. So wish me luck on the talk. But this is the app that I'm going to be talking about. And then I'm also going to be doing similar things to like talking about MCP and how the app works in general. Um, and I'm also going to Oh, something I did want to show you is that I only noticed yesterday that you can have your Okay. Okay. Wait, let me show you this. Let me show you this because this is cool. This is cool. You can have your agent actually test your app in real time using simple browser.

[00:47:51]So if you can see here, I have this like cool button here that says sharing with agent. And what it means is that I'm sharing my different um I'm sharing my different tabs in simple browser with the agent and the agent can test them.

[00:48:07]So I want to say uh can you open the app in simple browser and run tests on uh based on each uh each um chip prompt and this is something I saw so playright we talked about playright last I don't know if we talked about playright but this is I was surprised by this because VS code has this built in. And basically what can happen is that if you're building an application like this. So I've built this app and I want the agent to validate that the app is working as expected. The agent can actually do that for you. So we can see that I'm I'm not touching the keyboard. The agent has opened the application on its own locally. And then now what it's going to do is it's going to use playwright in VS Code to start clicking on buttons and testing things. Let me see. Oh, no, no, no, wait. Why is it doing that? It's saying now if I up the No, I don't want to use the API. Hang on, hang on. It's Let me say use player. Okay, now when I'm on the stream, it's not doing the way that I wanted it to do. Let me say use playright to click on stuff and test.

[00:49:39]Okay, I I you see this is what happens when you whenever you're doing something live, it does the wrong thing. But I'm excited to show you this because I actually think it's really cool. Um, and basically what I liked about it is that you can actually see the agent clicking on stuff and you can see it like running tests in real time and that was something I really enjoyed seeing. I don't know if it's going to do it. Okay, it did. Okay. Yeah, there we go. I don't know if you saw that, but it says clicked on this chip and it's activating it. I don't know why it's saying it's activating Cortana, but it clicked on the app and it's doing all of that by itself. Um, and basically what it's using is the it is actually using the chate the Playright MCP server that I already have installed. And then it's going to just like click around, test the different things on the page, try different tests, and then click them in real time. And then this is a great way to actually test that your app is working and the agent gets the response that it wants. How many of you have seen that? I don't know if I don't know if it's just me that's the only one that like thinks that's super cool, but I just think that's amazing because you can actually see it like clicking around and testing on its own. So someone said, "Ah, looks so fun. Looks fun. So a machine is testing the machine." Yeah, you get and this is the thing you need to get the agent to be able to test its own work because one of the things is that we don't sometimes the agents make mistake and it's hard for you as a human to always go and validate that the agent has done things correctly and so using something like this like playright and instructing the agent to do that is is really fun.

[00:51:39]Um ah look at this comment. Nate is saying any other trolls here? I don't really want to learn SDK. More likely I want to criticize the whole approach.

[00:51:54]Nevertheless, I might try it out sometime. Please try it out. Please don't be a troll.

[00:52:00]Um, I will share a link to an article about how to use the playright to the Playright MCP server, but basically it's it's just playright. You can actually go to the MCP store. Actually, let me show you how to do this. uh you can go here to the uh installation tab and then you can say at MCP and that's going to bring up different MCP servers and then you're going to click on the playright one. So that's the one you want to install and this is what is going to allow you to be able to do everything that I just showed as well. Yes, I think this is super helpful. Um I think it's it's it's it's great. Pontis is saying, "Good idea to bring the article up. It's easy to make a mistake by not knowing." Yes, 100%.

[00:52:52]Um, I don't know why the agent said Cortana.

[00:52:57]It really Yeah, I don't know why it said that. It should not have said that. But I do think the approach is super cool. I think that it's nice to be able to see the agent testing in real time. And I like that we can validate that the app is working and that we can see the agent like yeah working in real time. Um okay let's let's just chat now. We have only a few minutes. We have about 10 minutes. Does every anyone have any questions? Pontis is saying when I program it's all about flow. Sometimes I barely make a line of code and that's so sad. But hey that's just me. I like it.

[00:53:39]Especially in Spain, I'll sleep two two two nights and let let all the agents go ham 100%.

[00:53:48]I did see someone saying how can we use the guardrail and callbacks for these agents. So I do mention I mentioned that lang chain has a middleware layer and one of the things middleware is basically a way that you can get your agent to run certain functions before every any like LLM call and uh you can actually add guardrails then. But I also want to mention that you you can also have your AI agent go ahead and like if you're building your agent and you decide to use Azure. So for example in our code, let me go to the agent.py again.

[00:54:38]In our agent code, we use the Azure OpenAI API. And this is a great API to use because it does come built in with some features already that are like guardrail features. So I can definitely recommend using Microsoft Foundry to deploy models because it comes with some guardrails already. Um and middleware is also a great way to do that. Does anyone have any questions? We can we're about to close off for the day.

[00:55:11]Thank you Mustafa. I'm glad the session has been good. Um, the app has already finished running.

[00:55:20]So, I did it already tested all of the things. And if you want to check out the app, I have a link to it. This is the app that I was running. I'll put it on this screen as well in case. Oh, that took me to the wrong page. But this is the link.

[00:55:41]I've made it as an Azure sample. So if you want to actually no worries. Um yeah, I think this is the thing with these applications. It's like you never know. But if you want to test out the app that I showed, which is this one here, let me go back to the app to the sales app that we were also having the agent test. Um, that is the link. I've put the link on the screen.

[00:56:10]You can try it out yourself. I'm making some changes to the application. So, as we go along, it will update. And something I'm trying as well is to get different personas to have different access to uh different levels of information as well. I also have a customer service agent I'm building um with it too.

[00:56:33]No problem.

[00:56:38]Yeah.

[00:56:41]Okay. I see I see I I see some some funny things there.

[00:56:47]Uh Mustafa is asking React and Typescript for AI apps. React and TypeScript work great. A lot of people use that. For me, I like a mix of Python for the API stuff that I showed you with like uh the SDK or like Copilot SDK or an engine copilot SDK and then I also mentioned lang chain. Sorry. So, I like that AI bit. I like to use Python for that and then I actually use uh React and JavaScript in the front.

[00:57:21]Um, okay. Todd is saying, "Are there any frameworks you have used besides the copilot SDK and some other MCP that pairs well with Copilot SDK?" I use Langchain. So, actually a lot the app that I've showed on the screen is lang chain. Um, that's what I use to build it or the agent in it. And then, um, so that's the other framework that I use.

[00:57:46]Microsoft also has a framework called Microsoft agent framework. So you could try that out. But I like copilot SDK or lang chain. Microsoft agent framework is also good by the way. But yeah, I just use uh I use lang chain a lot and also copilot SDK. MCP servers. There's so many great MCP servers. I just recommend going to the extensions, typing in that MCP and scrolling. I really like Playright. That's my favorite.

[00:58:16]Obviously, the GitHub MCP server, it comes in, I think it comes pre-installed with uh Copilot CLI and Copilot SDK, which is why uh it's showing that it's not installed there, but it it it comes installed, pre-installed, I think. So, this is my other favorite is the GitHub MCP. You also have like a Unity MCP server, which is cool. That could be fun to test. Azure MCP server is great in case you want to do deployments. I have a bicep MCP server already installed for example. I would recommend just typing out that MCP and then scrolling for yourself to see like what is available.

[00:58:59]I think that could be cool to do. All right, I think we are almost at time everyone.

[00:59:06]Uh no worries. Thank you for joining me today. I am next time we call I'm going to make sure that I am at home and we'll be able to chat. Uh Stefan is saying thanks for the input. Next meeting is waiting. Yes, I hopefully I will see you all next time. I will be streaming at the correct time. I'm going to double check what the time is and the time zone.

[00:59:35]Um okay. Okay. Hey, I do see a question from Ada saying, "How can a non-programmer like me who has interest in AI agents and be able to build my own AI agents?" I do think you could should try GitHub Copilot. I think actually if you want to, I can recommend downloading VS Code and and trying uh Copilot in VS Code. I think that works for anyone even if you're not a um even if you're not a a programmer. So great answer, but I am so happy that we were able to chat today. Thank you for hanging out with me. Wish me luck on my talk today and I will definitely see you all next time.

[01:00:20]Thank you for joining. Bye.