Rubber Duck Thursdays

Hello. Hello. Hi everyone. Good morning, good afternoon, good evening depending on where you are joining us from. Welcome to today's rubber duck Thursday. So, it's yet another beautiful Thursday where we get together and talk about what's new in the world of GitHub. So, yeah, I already see so many comments on the on the chat. So, keep them coming and thank you to everyone who's contributing. Um, it's good to see everyone joining the stream. So, for our regulars, welcome back to yet another session. And if you're new to Rubber Duck Thursday, then welcome to this awesome space. Um, I saw a question already and um, so let me just scroll up that question. Um, yes, I agree. We'll have tons of fun in today's session. There is quite a lot to go over. So do stick around to the very end. Um, so Ratandep here is asking if anyone knows the agenda. So rubber duck Thursdays is basically a live stream where we come here and talk about what's new in the world of GitHub. So that's exactly what we're going to do. We'll go through the change log, see the new announcements that we have this week and then we'll probably go through some demos and see the CLI in action. So that's what I have in mind. But for this live stream, the agenda is rather flexible. So if you have any questions, if you have any ideas and projects that you'd want to talk about, always open to covering that. So um Rotten Deep, if you have any questions, if you have anything cool that you're working on and you want to share with the community, then that's something that you can as well do on this live stream. Right. Um so we see a comment here from Vina Sharma. Thanks Satan for giving me an opportunity to learn and deep dive into new things and updates of GitHub. That is exactly why we are here. So talking about what's new, what's exciting and what's not. So this is also a place for us to get feedback from you as the community in terms of how GitHub can continue to improve. Hi Ali, welcome to the stream. Hi Ashok from LinkedIn. So, we're streaming this on Twitch, LinkedIn, and YouTube. So, you can join on your platform of choice. Hello. I see so many people joining the stream. This is so good to see you all. Hi, Dak. Hey, Julie. Hi, everyone. Good to see you. All right, so I hope I've covered on the agenda, but as I said, we can have this be flexible. So if there's a specific topic you want to dive deeper into, happy to do that. Um but otherwise we'll just go through the change log, discuss what's new, and then just do some bit of um CLI stuff. Hello, good to see everyone. This is really nice, right? So um yeah, so Pandy, I hope that overview sort of makes sense in terms of what we do here. and it's your first time on Rubber Duck Thursday. So, welcome, welcome, welcome. We have this live stream every week on Thursdays. We have it across three different time zones. So, if you go on the GitHub YouTube channel, you'll see two other streams that will come after this one across different time zones. So, you can always join whichever um session makes sense, makes most sense for you. All right. Um good, good. So, we'll just get started. I see so many people on the on the chat excited to share what we have. Um, and I know it's been a busy week for GitHub and for the industry. So, I know that we have we might have so many questions. Please keep them coming. I will try my level best to answer them to the best of my ability. If not, I know that we'll have so many people to join us and to share um who are going to share some insights as well. All right. So Kumar is saying please start. It's exactly what I'm going to do. So let me just pop my screen and uh yeah. So we're going to get started with the in the on the change log. So this is basically um the new releases, any new improvements and any announcements on anything that has been retired this week. So I'm going to refresh this because the change log is one of the pages that changes quite fast. So we can see today is the 23rd of April and all of these releases were announced just yesterday. So there's quite a lot that we're going to go over and so I agree with you all. Let's just start right. Um so the first one is an improvement. I'm going to open this up and um for the customers and for you on the live stream here who are on Jira, GitHub Copilot now has some latest enhancements. So let's just quickly talk over what this is about. So if you use GitHub copilot in Jira. Um so looks like we have some more powerful and more um customizations that allow you to use the cloud agents from Jira. So let's talk about this briefly. Um so number one, we have support for um custom agents. So now you can specify in the Jira ticket a custom agent uh from your repository when fulfilling the task. So if you use Jira then it looks like right now you can bring in and assign Jira um tickets to custom um copilot agents. So this will allow you to have specialized agents for specialized tasks and yeah so looks like that's an enhancement that has been published this week. So again uh the link is github.blog/changelo if you want to follow along. So for those who use Jira this must be some exciting news for you. Then number two we have support for custom fields. Um copilot cloud agent can now read the content of a classian custom fields such as the acceptance criteria and include that context when working on your issue. That's fantastic. So now you are no longer limited to the default fields fields. You can now have the copilot cloud agent read through your customdefined fields and then pull that into the context. This way you'll ideally get better results while working with the cloud agent. So for again for those um using Jira, you do have support for pulling in the data in your custom fields as part of the context. So that's another exciting update. Then we see here you have custom branching rules. So copilot cloud agent now respects branch naming rules that are specified in your atlasian ticket. So here it's all about customizations and making copilot work for your organization, work for your team. So if you have existing naming conventions, so in this case if you have um existing rules on how you name your branches, then the copilot cloud agent will pick that up automatically and work in accordance to your existing convention. So that's also quite a good update. Custom instructions, you can now define um custom instructions for the copilot cloud agent. Again, this is for the users working on Jira. So, this allows you to define any repetitive configurations in a single instructions file and then the cloud agent will pick that up automatically with every request. So, this will just make it more useful. So instead of you repeating the same set of instructions every single time, you can define that in custom instructions and then that will basically work. All right, so looks like that's quite a lot of enhancements and improvements this week. So for those who are working on uh Jira with access to GitHub copilot, please have a look at the change log. There's quite a lot of announcements for you. Okay. Uh, I see some questions on the chat. Let me just scroll through. Hi, Marcos, welcome to the show. Uh, Kish is asking, "Why will Jira integrate AI models?" Um, why would it not? I feel like today models um AI models can really streamline a lot of um and make processes efficient on across different tools. So I think it's a question on why wouldn't you have a model basically come in and try and streamline some of the processes. So yeah that's that's it. Um so we have a question. How long do these streams usually last? About an hour. So about 60 minutes uh depending on how much you have to cover and the interaction from the community. All right. Yes. So I see John also um responded there. Why wouldn't they? so much domain knowledge in Jira and again um models are only powerful uh given the context that they have access to. So if you can bring these models close to where your data sits, close to where your processes are, then these models will easily just pick up on how you're already working, making them um making making it making it easier for them to um fit into your existing workflows. So that's a good question. All right. So the other improvement is on copilot code review. user accounts now aggregate in usage metrics API. Let's have a look at that at what that is about. But it looks like it's um uh some new parameters have been introduced to give admins more visibility in terms of how the co-pilot code review is being used across your enterprise. Okay. Right. So it looks like uh following the launch of the copilot code review active and passive user identification enterprise and organizational um usage reports in the copilot usage metrics API now include aggregated active and passive user accounts for c-ilot code review. Yeah. So for businesses for organizations you have allocated several seats across your company. So now you can basically have a good aggregation in the metrics in terms of how the developers in your organization are using copilot code review. So with this improvement we see that now you can have access to metrics in terms of the active users who use the co-pilot code review on a given day or across a 7-day window as well as a 28 day window. And that cuts across from active users and passive users. So if you're wondering in your organization, how do you define an active user versus a passive user of co-pilot code review? I believe that's defined here. So anyone is an active user if they one manually request a co-pilot code review or if they apply a copilot suggestion. So within your organization, users who do those two things will fall under the active users category. And for anyone who probably just has copilot doing a review as part of the default policies. So if you create a pull request, then you have some autot triggered reviews. That doesn't count as an active user, but it will count as an active user if you interact with the suggestions from copilot. So again, if you want to understand more about how you can get even more um specific metrics in terms of how copilot is being adopted in your organization, this is a good um update to to read through and again this is available on the github.blog/changelog. All right. Um so the next one here is quite interesting. So we have support now for bring your own model key in VS code. This is now available. Let's read through this. I I thought this was already supported. So let's just read through and understand what it's about. So copilot business and enterprise users can now bring their own keys into visual studio code. So this allows you to bring in your own your own API key and you can connect GitHub copilot to be serviced by your own provider of choice. So could be Anthropic, could be Gemini, OpenAI, Open Routter as well as Azure. And this also extends to support models that are running locally either through Olama and Foundry local. So this one here is exciting. Looks like it's for uh those of you who are on copilot business and enterprise. So this means that you can now bring in your own API key and connect to models that are hosted by your provider of choice. And this uh as you use copilot, those requests won't be um won't count towards your copilot subscription. this will um be charged to your model provider. Right? So that's that's an interesting update. So for those on business and enterprise subscriptions, be sure to check this one out. Um so Majid, I see your question. Is there any change for upcoming release? Could you clarify that question a little bit? Um I don't understand it. So kindly just um explain what you mean by that and then I'll come back to it. All right. So that's it. Let's see what else. Um GitHub CLI opt out usage telemetry. So I'm hoping that everyone here has had a chance to try out the GitHub CLI. If you haven't, we are going to do a demo on it today. So you can stick around for that. And if you've been using the um copilot CLI, please let me know in the chat how has your experience been like and um yeah any tips that you've picked up on that you can share with the community. So as of yesterday looks like um you have the option now to opt out of um the collection of usage telemetry again. Let's see what that's about. So yeah, we start by understanding why is it that GitHub collects telemetry. As usual, this is to help in the growth of the tool. So, we collect some um telemetry to help understand what features work for you as a community, what don't work. Uh and we have an example here. So, once the team ships a new subcomand, we want to know whether you're using it and how you're using it. If the adoption of certain subcomands are low then the team revisits the discoverability or design of the command and then positions it in a way that it will be more useful to you as user. So that is why um the team here is collecting the telemetry. So what has changed according to this announcement? So it looks like now you can actually review what the team collects the CLI is open source. So if you visit this repository, you should be able to view every single data usage data that is sent back to the team. So you have full visibility in terms of what gets collected. Then if you want to see exactly what the team collects without sending it. So you're curious to see what form of metrics are or what form of usage data is collected. Then we have these instructions in terms of how you can basically um configure an environment variable here and then set up this configuration to just preview what gets sent to the team. So this will allow you to preview what is collected without actually sending it. So that's something you can actually try right now. Um, so once you get into this logging mode, the CLI will print out the JSON payload that would normally be sent back to the team and then you can inspect every field before deciding on whether to keep it enabled or to opt out. So again, full transparency, you get to see what gets collected. And I believe now you have the option to opt out. So if you review the payload and if you um review that um information and decide that you want to opt out again this is the instruction. This is exactly what you need to do in order to opt out of sending your telemetry data. Then lastly they do have a note here in terms of where that data actually goes and you see that the CLI sends the telemetry events to GitHub's internal analytics infrastructure. So this again is for the sole purpose of improving the product. All right. Um let me see. So Madrid, you can just expound more on the question you asked earlier. Um Smith says, "I'm using codeex and GitHub CLI for project automation and it's amazing." I'm glad to hear that. Again, we're seeing a lot of cross tool collaboration. So if you find a workflow that works well for you, go for it. So I'm happy to hear that. Um yeah, I have another comment here. Um I use the the GitHub or get CLA for almost everything creating, deleting, cloning, and editing repos. That's good to hear. And I hope that everything is working well so far. If not, please do drop your feedback on the comments. Looks like this has been quite a busy week. Um, all right. So, there's a release here on C++ code intelligence for Copilot CLA in public preview. We won't go into that, but if it's an area of interest, have a look at that. CodeQL now supports sanitizers and validators in models as data. Again, um, we won't dive deeper into that one, but it's also an interesting one. um to cover. So there's one here that I wanted to just briefly talk about because we've seen so many reactions and um I want you to know that you know the team here really knows that this has been a disruptive announcement. So what I'll do is I'll just summarize this um announcement around the changes that have come into the co-pilot plans for individuals just to give you an understanding of what it means and why we are doing it. So if you have any questions again just put them on the chat and happy to talk um through them. So one of the biggest announcement this week we've seen is that uh we are seeing some changes into the co-pilot plans for individuals and I will just summarize them as they are on this blog. The first one is GitHub has paused um the sign up or the enrollment of new users that is for the pro, pro plus and the student subscription types and uh the reason why this happens is because um the free trial system for the generous co-pilot um trials has been we've seen a significant um rise in in abuse of that free of that free trial system. So in order to have some time go back to the drawing board and redefine a system that will work. Uh so we've now posted the signups for new subscriptions. So that's something to have at the back of your mind that that's a change that has been rolled out this week. Then in addition to that just to also ensure that for the existing copilot users we are having a rather predictable and sustainable model to deliver the amazing services through um through GitHub copilot. We now have some tighter usage limits for individual plans. So in a way to summarize this we have two big changes. So the first one is that we've introduced session limits. So once you run a session for some time, you'll hit a limit and you'll get a warning either on VS Code or on the CLI and that will resolve when that session resets resets. So that's something you might note as you keep using copilot that we've introduced some tighter session limits. And then we also have a weekly limit which will basically introduce a cap on the tokens that are consumed in a week. So again there's a blog uh let me see there should be a blog that captures all these details but um that means that you'll see some titerate limits and um you'll basically just have a warning that will let you know as you're approaching the as you're approaching the the ceiling. This way you won't have some surprise um termination of the service. But again for the weekly rem limit for instance um that will just reset after after the week. Then the other update here is that the opus models are removed from the pro subscription. So opus models are no longer going to be available in copilot pro but oppus 4.7 will remain to be available on pro plus. So that's the other change that we saw this week. And again, I do acknowledge that these are disruptive changes. I'll I I'll just say that these are all in efforts to ensure that the service remains to be um to be predictable. It continues to be reliable for existing customers. So also be on the lookout um in case we have any further updates on the same. Let me see on the chat if you have if you have any questions. Okay, Madrid is asking so it means pro version exist more feature. I don't know what I'm missing but I don't quite get the question but from this announcement um yeah the existing subscriptions still remain. you just can't sign up for the new uh you can't sign up for a pro pro plus or a student subscription at the moment. Right? So that is what um has been announced this week. All right. I think that's quite a lot. Uh so again just visit the change log if you want to see anything that's that's new. So what I'll do is I'm going to jump into my terminal. If you have any questions, if you have um anything you'd want us to talk about more, any demo you'd want to see, I'm going to put myself on the spot there and uh and just say that you can request for that. But yeah, I'm going to do a quick demo here just to show um Copilot in action. So again, if you've not used the Copilot um CLI, please do check it out. But as soon as you have it installed on your terminal, you can easily type You can easily type copilot and that will spin up copilot CLA right there on the terminal. So this is your interface for you to connect with an agent that can carry out um development tasks. So just as you would use copilot on any other interface on github.com on github mobile on VS code and the other idees you have different ways that you can configure and interact with different settings for your agents. So for example you can browse through any available agents. So you can create custom agents. Um you can switch between different specialized agents for different tasks. Then uh you can you can also let me see you can also check on the models that you have access to. Now this list here would be different depending on the subscription type that you have but you can see all the models that you have access to. You can also see as of last week I believe um we had support for auto mode that is having um the option to just have one selection that is auto mode and then copilot will intelligently decide which model to use based on service health and availability. So with auto mode support this means you don't have to think so much about which model you want to use for which type of task. you can simply just use copilot CLI in auto mode and then it's going to reroute based on the model that will will probably um give you the best results. Right? So that's something that I believe that was announced last week if I'm not wrong. So if you haven't uh tried it out, you can use auto. And something else I'll call out is if you use the CLI in auto mode, this means you'll get a small discount in terms of the service charge for the different models. So if you use auto mode and it routes to a model that's charged at 1x, then you'll basically uh get a discount of I believe it's 10% of the charge for that particular model. So it's also a cost effective um approach to using the CLI. So that's something that I wanted to point out. Right. So for this demo, um I'm in this folder. It's called Zava and it only has one file. It's product data.json. So I'm going to open this file in VS Code just to see the contents of this particular um JSON file. So I'm I'm going to open this in VS code insiders. I default to using insiders just to get the most upto-date um features, experimental features. And if I open this product data um JSON file, you'll see that it's just some some product information. So we have product categories, we have some product details, name, SKU, price, description, stock level. So what I want us to do is to basically see if we can create an interactive dashboard to work with this particular data. All right. So one is you can see this data is quite a lot. Um yeah this is a lot of information. So what I'll do and just because we can do that I'm going to ask copilot right here in the CLI to convert this into a an Excel file right I mean it would be easier to um digest the information that we have in that JSON file. Then think of an instance where you have data in this format and then someone from a different department say the sales department wants to have an understanding of that product data. It's a bit painful to look at it in JSON format. So probably we could have we have this in a format that is quite easy to share. That would be nice. So why don't we just put Copilot in the CLI to the test here and see if we can get some some results. So I'm going to ask copilot. I'm going to use sonet 4.6. So I can ask it um can you convert the data the data in I'm going to use the add symbol to click on that file. So can you convert the data in this to a spreadsheet? Then I'll add ask it to use different sheets and add visualizations to help understand the business. All right. So, let's let's try that. Um, okay. So, Adam says, uh, you have no idea what's going on here. So, I don't know at what point you joined, but we just went through the change log and right now we're just doing a demo on the CLI just to see the CLI working, seeing it in action for those who who are yet to try it out. So, for the demo part, we have a JSON file here with some product data. We want to first see if we can convert it into a presentable format. So assume you've just been given this data and you want to just have some insights, have some business understanding on what this data means, then probably Excel would be um a nice way to represent the data. And so we want to see if the CLI can help with such a task. So yeah, we've just sent our prompt. Can you convert the data into a spreadsheet and just use some different visualizations? And we see that the CLI agent is now picking up on the task. Uh it has a good understanding of the data structure. So it's going to create a comprehensive Excel spreadsheet. So you can see here you can just read through the progress what the agent intends on doing. So if you read something uh in its reasoning logs and then you pick something that you don't want it to do, you can easily just follow along with a prompt to correct it. Or in this case, uh for a start, we can just work we can just have it work through and I expect it to just create like a simple Excel file to help me understand what's in that data. Right? So that's basically what we're trying to do here. And then after we have that, we'll just take it a step further and see if we can also have the agent build an interactive dashboard just to again see if we can convert this from rojson data to a format that we can probably share across departments and then eventually get to see if we can convert that into an interactive app. Right? So, and in the process, we're just going to pick up a few tips in terms of how we can use the CLI. All right, so it's still working. Let me go through the chat. Okay, so Jar has a question. Is there a way to check the rate limit in GitHub Copilot? In Cloud and Codeex, there's a 5hour window. How does it work in Copilot? Right. Um, so I know we have a usage command. Let me just start a different instance here. So with a slash usage. So with a slash usage, you should be able to see your um you should be able to see the requests that you use. So in this case, I've just um opened a new chat that doesn't really show much, but after a working session with the CLI, then you can basically just have a breakdown of the premium requests you've used and you can check that against the allocation that you have. This should give you um this should give you an idea of how much you've consumed and how much you have remaining. And as we also saw in the change log uh with the newly introduced session limits on the CLI as well as in BS code, GitHub copilot will issue a warning. Myself, I am yet to hit that um that limit. So, I haven't really seen that warning, but from the change log, it sounds like once you approach the ceiling of your um usage, you will receive a warning just to tell you that hey, this is how much you've used up and probably you should now probably just think about how you can wrap up since you'll be hitting your limit soon. So, yeah. So, Jamar, I think you will get a notification. I just haven't seen that notification because myself, I'm yet to hit it. But before you hit your limit, you're going to get a notification that alerts you ahead of time once you approach it. All right, good question there. Okay, and that's through the CLI on VS Code, I believe. Um, let me just switch back to VS Code. We do have this Yeah, we do have this copilot icon and if you hover on that icon, we have this um we have this manage chart icon and if you click on that it will basically take you to your copilot settings and you can have a better visual in terms of what you have remaining in terms of the the the quarter that you have, right? Yep. And more specifically, I think this is a new UI, we have this new button that allows you to also see the rate and your usage of co-pilot across the day. So you can basically see how many suggestions are you accepting and how many are you not accepting. So that's also some good metrics to have. Okay, so let me switch back to the terminal. Um, we see the agent is still building the spreadsheet. So, let's give it a few a few minutes. Um, all right. Massive first time here. Massive edits. Welcome to the show. We're just trying to work with Copilot CLI. Um, quickly, let me let me hit yes. So, I'll just give a summary of the scenario we're working with here. Salv uh good morning. No worries. Welcome. Welcome. So what we're doing here again you can see in the background here I have a very long um product data JSON file. So this is just some product data. We want to see if we can have the copilot CLI help create some easy to digest format for this particular data. So I've asked it to create a spreadsheet and again the scenario that um we are working with here is assume you have this data and you want to share it with a colleague from a a different department say sales right so it would be difficult to share that uh as as a JSON file so we're trying to see if we can have the CLI work its way to build a spreadsheet from that data All right. So, we can see it's generating multiple sheets and multiple charts. Continue. I don't know if I've just stopped it by mistake. Let's allow that. So, it's generating our spreadsheet. It's hitting some errors, but it's also resolving that. Again, that agent um agent loop, it's able to work through its errors, find out what issues it's hitting, diagnose, find fixes, etc. So we see that the agent has created an Excel file. It's saved that into our current folder. And here is a summary on what it has built. So let's just open that. I'll open it with Excel. So let's see what the agent was able to come up with. Okay. So I have it here and there we can see that in a few minutes we moved from we moved from data in JSON format. Again this works if you're building solutions but think of a case where you now want to share this with different departments. again with the use of um AI coding tools with the use of different agents we're seeing a lot more cross team cross department collaboration so you don't have to spend so much time you know like working um working on different tools you can easily you can easily get a tool like the copilot CLI to help with some of these tasks um so Moid is asking what agent is this this is the copilot CLI. So we're using the GitHub copilot CLI for this task. We've given it a file in JSON and as a result you can see we have the summarized product information in form of an Excel spreadsheet and we it has actually done um quite a decent job in terms of creating the visualizations to help understand what's happening here. It has categorized the the product per category, subcategory, SKU. We have the price and yeah, I see we have a couple of more um product sheets. We have a couple of more visualizations. So, all this is good. I won't spend so much time of this because I actually want us to do something else with this data in this format. So, I'm going to save this. I'm going to close this. Right. And so at this point, I'm actually going to delete this JSON file. Bear with me. I'm going to delete the JSON file. And so we have this Excel file. And I'll go back to the terminal where I have a copilot CLA running. And what I want us to do is to now just build an interactive dashboard to help us have some interactions with that data. So right now we have it in Excel. Now let's see how we can easily just again get back into our development um workflow and build something real quick out of that. So before building out the dashboard, I normally start by working on some designs, but I am no designer myself. So what I'll do is I'm actually going to ask Copilot. I know about a recent tool uh that is called pencil. I don't know how many how many of you have used pencil. It's a design tool. I have I am yet to try it but it sounds like it's a good alternative to connect agents to a design interface and it's able to build out some designs for you. So what I'll do as a first step is I'm going to ask the copilot CLI to do some research and we have um a research command. So I'll just type slash research and ask it to find out all there is to know about and then I'll paste in the link to the documentation for pencil. Right? And before that I should have cleared my session just to reset the context to window. That's something that is recommended. So what I'm doing here is before using a different tool with a CLI, I like to start with this research research step. So this way it will go off and do a deep research on that particular technology, the framework, the service that you want to use and this will give it enough information to know how to effectively work with that particular tool. So in this case I myself am new to pencil. I've just had some good reviews about how you can easily convert ideas into designs instantly with agents, but I haven't really tried it out as much. I'll also assume that my agent needs to build some form of knowledge about this particular tool and how to use it. So I start by that first initial step which is hey can you do some bit of research around this tool right and that I believe you'll see some improvements in terms of how your agent eventually ends up completing the task that you gave it. So you'll see that the agent in research mode went off and did some deep research and then it now has some understanding of pencil and it's going to build a comprehensive report based on the findings of its research. So again it's part of my workflow before starting anything new. We've just started in an empty folder. I want to actually build the context with the agent hand in hand. All right, let me see. Do we have any questions? Okay, so Naral is asking GitHub Copilot allows review code once code changes are completed based on the requirements. How do we control this? So I believe you're talking about the dedicated code review um feature which is available both on VS code and the other IDs I'm assuming as well as on the CLI. I believe on the CLI if you type slashcode review. Let me see review. Yeah, code review. This will run a code review of the changes that you've made. And then we also have uh on VS code where once you've done your changes you can instantly uh before committing them you can have the copilot come in and perform a quick review. So that's built in as part of that process where you work on a feature and then just before you push that into the cloud you basically bring in a an agent to do a surface level review. But part two of your question is how do you control this? You can always ask the agent to do a code review at any step of the process. Right? So you can be working on one particular feature and then you ask the agent to review that part of the code. So you don't have to wait until you're done with your session or you're done with your implementation and then invoke the code review. Code review is just a builtin feature that you can um you can opt to use just before pushing the code from your local device. Again, that's to just do this last minute review or last step review of your code before you push it. But you're not limited to that. Every single time you're working on your project, you can always initiate a code review with your agent of choice. Right? So that's something you can do. And again with custom agents, you can always define handoff scenarios. So where you can work with two agents, you can probably have a design agent, a builder agent, and a code review agent. And then just look through the three for your project. So for a new feature, your entry point is always the design review which builds out the designs, hands those designs over to the builder agent, and then the builder agent hands over the work to the code review agent. So you have so many ways you can customize that experience. You don't have to work with the features that you get out of the box. So I hope that answers your question. Um what is Robert duck? Uh yeah so uh this session is basically just a weekly live stream where we come together and demo talk about what's new in the world of GitHub. And so for today, we're just looking at Copilot CLI. For now, it's finalizing on the research on this new tool that we want to have it learn and find out all there is to find out about Pencil before we even start using the agent for this particular task. So, we'll just give it a few more seconds. I'm hoping it will be done shortly so that we can get to the next part which is to now have it build out the actual designs. All right, questions. What What is everyone building? What are you guys working on? Uh so we have we have a question which areas to use can you use it for education? Yes absolutely. So you can basically apply these agents to uh to your areas of choice. Just give them the right context give them access to the right tools and they will get the job done. So our research is done right. So we can see that we started with an empty folder, an agent that probably didn't know about um this tool that is pencil. We pointed it to the documentation and asked it to do a thorough and a deep research on it. It has now come back with some research findings. So it now has an understanding of what the tool is, the defining pillars, how to use it, etc. So what I'll do is right now this research is saved in the memory of this session. Right? So you can see this has been saved in a do copilot folder. So this is where the CLI configuration is mostly set. But in this case I want to pull this research and have it in my work base or in my workspace. So I'm going to use the slash share um command. I will use that and then specify that I want the agent to share the file and then this file is as a result of a research activity. So I wanted to share a file and I wanted to call it research.md. And I'll specify the research report that I wanted to save. And so now you'll see that in our folder we started with an empty folder and right now we should have a research.md file which of course contains the research findings from the um from the agent that that research session that we've just had. It has summarized and as you can see it's a very comprehensive report and now it has saved that into my workspace. So I can easily go off and start a new session. So this will allow me to reset the context. I have the research. Now the next step is I want the agent to actually use that research and use that tool to build out some designs for me. Okay. So we're going to clear that. And again just using the same model I am going to ask it. I'm going to pass in a prompt. So I want to build an interactive dashboard for I'm going to select the Excel file that we generated. That's the Zava product catalog Excel file. And I'll explicitly ask it to start with the complete and to end designs using and then I'll point to our research finding. Right? So what am I doing here? Again the workflow starts from there's a project we want to do. We do some research on the tool that we want to use. The agent saves that research and then I'm telling it that hey, we have an Excel file. I want to build an interactive um dashboard for it. So, let's do that. But before we build anything, let's start by working on the designs. And for these designs, I want you to leverage the research that you've just concluded. Right? So in this case we'll see the agent um acknowledging the assignment the task that we have just given it. It's going to go through the data that we have as part of our project. So that's why I keep um resetting the context. This way we're not um polluting the context. We're having it focus on the things that matter. Um it's going to look through the tools that it has um available. So actually something that I forgot to do was um I wanted to do this in plan mode. So I'm just going to clear this session. I'm going to clear this session because I first want to build a plan. Apologies, I I missed on that. So, I'm going to switch to plan mode and then I'm going to use the same prompt again, but I'm going to do it in plan mode. So in plan mode, it's not going to go off and start with the implementation, but it's going to analyze the resources that I've given it and then just build a plan. So you can see it's trying to see if it has access to any skills, any custom agents, any customizations to help it do the job better. But no, it didn't find anything. So let's see what it comes up with. All right. Uh going through the the chat. so I have created a project but want to publish through GitHub. Please help me. Um yeah. So PL um allow me to redirect you to the GitHub page. We do have a dedicated series. So let me just show So if you go to the GitHub YouTube page, we do have a dedicated GitHub for beginners playlist. I believe that this will have all the information that you need to quickly set up your project and successfully push it on GitHub. It's self-paced so you can go through it um whenever you have the time. So just allow me to redirect you to that. Uh so it's YouTube.com/github. Okay. So let's see. Um so now in plan mode uh the agent is trying to understand okay this user is trying to complete a certain task. Let me ask a few clarifying questions to understand hey what's the vision that they have. So in this case it has analyzed the Excel file. It has analyzed the research about the pencil tool that we wanted to use. And so now it's asking what should the plan cover. I want it to cover everything. That's design plus code. the visual theme. Let's let's work with light theme for now. Then which dashboard or screens do you want to have in the designs? I'll go with everything. So everything is selected. That's fine. The design uh style or brand. Um in this case, let's go with there's modern and clean. There's bold and industrial. Let's go with that for the sake. Yeah, strong colors and the pop. Let's Let's go with that and see how it goes. So, it's not going to build out the plan. I expected to split it into a design phase because I explicitly asked it and told it that, hey, this is how I work. I love to build out designs before building anything. So, I'm hoping that it's going to split that into at least the two faces, if not more. So, we'll let it cook. I'll go back on the chat. Um, so Stefan says, "I am working on a few things. Please take a look at my personal site." Okay, so I believe that's for everyone on the chat. You can give Stefan some feedback on on his work. All right. So I'm recently I recently started using coding CLI agents on VS Code. What are your best suggestions to choose the best coding CLI based on fact and requirements? Good question and allow me to come back to that in a minute. Looks like our plan has already been generated. So we have the plan here. It has um the summarize that we want to build a hardware dashboard. This is phase one, the designs. And as you can see, it's not just randomly guessing how to approach the design phase. I gave it research. And so it's grounding its response and this plan on the research. And then phase two is to now build that React app. So in this case, I'm going to change one thing. You can accept the plan as is and build it with autopilot. Autopilot is a mode where the agent will basically ask for no input from me. It's just going to go on and on and on up until it completes the task. I can either accept the plan manually and then have it build on my default permissions or I can exit plan mode and then prompt this myself. But then I'm going to suggest a quick change. So instead of react react let's do a desktop app. So I'll just prompt it to revise the plan. So again this is a level of control that you have. You can have the agent generate a plan. Spend some time on that plan. Review it. Change what you don't like. this case, let's just go with electron. And then once you're you have a solid plan, something that you're comfortable with, then you can just let the agent work. So it has asked which stock do you want? You can give it point it in a specific direction or just have it work with its own recommendation. So it's hopefully going to quickly update the plan. So you can see it's editing the plan. And again, just like the research that we did before, this plan is being saved in the copilot folder. So if you want to have it referenceable from your workspace, you can always um export that plan and just bring it into your workspace. I like to do that because as soon as I get out of this session, I'll basically just lose all that context. So, I'm going to exit the plan manually and I'll ask it to save the plan in the root uh in the root. Okay, let's let's just give it a name. Plan MD. Oops. Plan MD. Yeah. So hopefully it's just going to grab the entire plan and then just save it in my current working directory. This way I can always just pull it in and reset the context without um worrying about um losing what's logged or saved in my session logs. All right. Uh so the question so my best suggestions um I I think this is different across different users. I mean one the biggest thing right now is access. We've seen so many changes um around how how you can access the different um coding agents that are available. So once you have access to your set of tools I believe this varies from user to user. This varies depending on how you want to use these agents. At the moment, I believe that most agents sort of cut across the same tools. We have defined specs that allows these agents to work with different sets of external platforms and services. So, I would say that part of my answer here won't probably be what you're looking for because this depends uh and varies from user to user. So for now, if you could try the different CLI based tools that you have access to and then find one that you know fits with your rhythm, is able to seamlessly connect to the services that you care about, is able to uh you know addresses the security concerns. So, I'd probably just bounce that back and say that you're better off trying out the different agents because so many suggestions that I even come across on the internet are just based on individual configurations, individual experiences and that doesn't cut across for everyone. So, Nimal, I would probably say that from from my perspective, I work with the CLI most and it has worked for me so far. this would be different for you for someone else. So, it's just a matter of you finding a tool, finding um a set of tools that fit within your existing rhythm, right? And I know that's not the best answer. I apologize, but uh that's that's the best that I can give for for now, right? So, we'll double check. Uh so, I'm going to just confirm that our plan has been saved. So yeah, we see we have a plan MD file that has been added in our directory. So I can basically just clear this session, right? So again, I'm just going to reset the context. So we are working with a fresh context for this next step. Okay. And so the prompt that I'll go in with is um that hey uh let's now start with the designs. Do it one by one and then save all the screenshots in a new folder. Designs. So for the model, let's just stick with set 4.6. And yeah, so I'm going to hit send. I'm going to follow that prompt with um use plan MD. So, I'm assuming that it's going to pick it up um by by default since it's in my working directory, but I'm just um following along with this prompt to use the specific plan that we have just worked on previously because remember this is a fresh context window. So, it doesn't have our previous conversation as context. It's going to try and work with what it sees in the folder. So in this case I have basically asked it to focus on the plan and then build out the designs. Now before that you're probably asking hey so how does it actually connect to pencil? So something that I did before was install the pencil MCP server. So if I use the / env command on copilot cla you will see that one of the mcp servers that I have is pencil and I also have the status which is connected. So you can use the / env um command on the CLI to see a status of the different configurations that you have from MCP servers skills plugins etc. So, I have the pencil tool already. Um, MCPS have already installed. It has gone off and opened the pencil app, right? I believe that's how it's supposed to work. So, let's see. So, this is what just popped up on a different screen. I've just moved it over to this shared screen. I'm going to switch back to my terminal. And you can see that now the agent is setting design tokens first and then it's going to build out the components and the screens. All right. So let's see. Let's see if it's able to complete the task. I wish you could see this side by side. So let me minimize let me minimize my terminal. I'm hoping you can still read this. So you can see it working and I believe this is the window that it's working on this. Okay, I don't have much room on my screen. So let's let's see if we see any designs coming in. I'll go back in the chat. Yeah. So, Patchy says, um, created an account on GitHub, but I don't know how to use this app. Please, uh, tell me something. So, Patchy, I'll probably share this same um, tip that I shared earlier. I would redirect you to the GitHub for beginners playlist on the GitHub channel. Trust me, this has everything you need to know from setting up a GitHub account and all the different features that you can access on GitHub. Not just copilot or AI features, but everything across the GitHub platform. So, I'm actually just going to click on the playlist itself and drop this on the chat just for anyone else who's starting off with GitHub and probably just needs some um some resources to help get started. So patchy I would redirect you to that playlist. Okay. So all right let me just approve again for tool calls you have to give explicit permissions. So you can either allow it to use the tool in this session or always allow it. So you again you have control in terms of um in terms of the permissions that you want to set. And so we see our designs are coming in. I don't know if you can see this. Let me try and zoom in a little bit. All right. So, looks like we have some designs coming in. So, the agent is actually working on the pencil app in real time. Uh, you can see here it's asking for some additional permissions. So, let me let me just accept them all. I don't know if this will allow me to grab the screenshots, but yeah, you can see now it's starting to build some designs again. It's not making anything up. It is working off from the research we did initially. Then the plan that we created and now it has access to the tool to start the work and you can see in real time our designs coming in. Uh yeah, so Numal, you've used cloud code, you've used codeex, open code, and you're new to copilot CLI. Yes, I agree with you. It is interesting. So maybe just give it a try and see how this compares and if it actually um extends and meets any new any needs that you might have. But yeah, go off and uh yeah, experiment with the CLI. I'm sure you'll have tons of fun. All right. Um, feature two few words about using fleet orchestration. Um, good question. So, let's see. So, we'll let the agent work on the design. We can see that's happening right here. So, we have house. Okay, is asking about the fleet command. Let me check. So, we have again for those who are new to the CLI and you can see the designs coming in in the background. Um, our agent needs some approval. So, let me just hit yes. sorry. I'm just trying to I'm trying to um balance just so we also wrap up uh in time. I know we are a bit over the 60 minute mark, but let's see how far we go with this. Uh, so you can see now it's trying to take the screenshot. So, I'm also going to say yes. Looks like it's grabbing the screenshot. I asked it to save that in a designs folder. And I'm going to see if that's been done. So, it did create a designs folder. Let me let me expand this a bit. So, it created a designs folder. Then, let's see if it has saved anything. Yeah, we do have a file. Let me open VS Code so we can just preview what has been saved here. So yeah, we have our first design. It's been saved in our project directory. So that's good. So far it has done everything we asked it to. Okay. So yeah. So hopefully it's going to start working on on this. So it has verified that the screenshot looks good. It's seeing some bit of issues. So you can see it's also selfhealing. So it's going to on its own inspect the work that it has done and do anything that it needs to fix this. So I'm not sure why it keeps coming back with a request to approve the pull calls, but that's that's fine. Yeah. So there was a question around the fleet command and if you're not familiar with that. So the fleet command allows you to to run different agents in parallel. Okay. And um this would have actually been a good demo for this session but I needed the designs to be built first before you know getting another agent to now start implementing this application. part. Think of a scenario where we have the designs coming in and then we also have an agent ready to work. We can use the /fleet command and then the CLI will dispatch two agents running side by side in parallel and then it's basically going to assign each agent a specified task. So we could have one working on the designs and then we could have the other picking up the designs and implementing them. Right? So this case you're having the CLI as an overall orchestrator. So it's going to balance the tasks and if it can pick output from agent one and then pass that as input to agent two it does that and then at the end of that session it's going to give you a summary of what has been accomplished as well as the agents that were involved. So for the fleet command, it's also an interesting one that you could experiment especially if you want to run um tasks in in parallel. Uh let let's go back to our agent. So we can see it has worked on the first design. It's now working on the second one. Right? So you can see a product catalog design has just come in. It's asking for my permissions. Let's just approve that. Okay. Um, when should I use the GitHub Copilot CLA instead of GitHub Copilot in VS Code? What are the main differences in use cases, capabilities, and workflow between the two? Maro, that's a good question. Um, so number one, it can be as simple as a matter of preference. Do you prefer working on the terminal or do you prefer a more GUI approach where you're clicking on buttons, you're seeing progress bars. So it can be as easy as I just prefer to click buttons so I'll work more on VS Code or I just prefer staying on the terminal and you can go for the copilot CLI. In terms of features, Copilot in VS Code has more features compared to what you get in the CLI. So for example, or more specifically with Copilot in VS Code, you also get the integrations with the IDE itself. So if you have any context that sits within the IDE itself, Copilot is able to pull that as part of the context. So for example, if you're working in VS Code and then the agent tries to run the application, it will run that using the terminal and then probably the issues will be captured in the problems tab. Then these problems can just be captured as context as you're working with copilot in VS Code. While on the other hand working with copilot in the terminal the terminal will not necessarily have access or visibility into the context that is um native to the IDE. Right? So to your question, two things matter of preference and then the other one is if you are interested in having the overall context that is captured within the IDE as part of your sessions, then you're better off working with co-pilot in the IDE as opposed to the CLI. So I hope that answers the question. But in terms of workflows, you can work with any form of customizations on both interfaces. So you can work with custom agents, bring in your custom instructions, bring in skills both from the IDE as well as on the terminal. So um you can let me know if that answers your question. Uh Arian is asking is GitHub all functions are paid or it's a free of cost platform. So for what I'm demoing today, this is this is part of your GitHub copilot offering. So you would need to have a GitHub copilot subscription for you to access um the features that we are seeing and what we are seeing um working in the background. So not all functions are available for free. Some are part of your different GitHub subscriptions. And so for what what I'm demoing here today, you would need to have access to a GitHub um to to a GitHub subscription. And even if you're on Copilot free, you will have access to Copilot CLI. All right. So we can see the agent is still working on design. So we have several designs that have come in. So we have different pages. And as you can see, this is a pretty decent job to be honest. It's taking screenshots. It's saving those in in my project. So that's absolutely fine. So what we can do is in a different session. I'm actually going to switch models. So let me work with GPT 5.4 here. Medium medium reasoning effort. And then I could ask it to Yeah, since we had a question on fleet, let's let's just try this. All right. And and see if it works. So we'll kick off agents in parallel. So I'll ask the CLI. Um can you dispatch one agent to start working on the desktop app? following. Remember, we created all this in the plan. So, I'm just going to tag my plan. So, dispatch one agent to start working on the app while the other confirms the designs that are ready for implementation. and I'll tag in the designs folder. Okay. So, let's let's see how that's how it handles this task. So, our designs are still being created. They're still being saved in our project. So, what I've just kicked off is what I hope will be um agents now running in parallel. We should see an agent to confirm the designs that come in and then we should have one now start working on the project. All right. So, I found the workspace structure. There's a ripple label plan and some designs. Okay. So I'm hoping it's also going to let us know the two agents that uh that will be working. So yeah, so you can see I've turned these into two independent work streams and I'm dispatching both now. So one to begin on the electron app from the repo plan and then one to inspect the four designs and confirm implementation readiness. Okay, I'm not sure I came off clear in in that ask uh because it looks like it will try and inspect the designs themselves. So, yeah, I might have captured that wrongly in my prompt, but let's just see if it's able to get the job done. All right, so you can see we now have two agents. We have the desktop app builder as well as the design readiness reviewer agent. They're both running in parallel. And now we have the CLI basically just monitoring and sort of just acting as the orchestrator for the two um agents that are running in parallel. And you can do that with a fleet command. Okay. So let's let's just approve for this tool called. So you can see we have three I would say we have three work streams running. We have one that's building out the designs. It's connected to pencil and it's creating it's creating our designs in real time. And then now we have we have these two that are running and as you can see now it's starting to build out the actual application. Um so Bazal is asking what is the best code? I'm assuming you meant IDE to work with GitHub. this depends on what you're after. Um, Copilot is accessible across a variety of IDE. So, we have this um this notion of we go where the developers are. So, whichever IDE um your company or you adopt, we try as much as possible to bring copilot functionality within those idees. But in terms of feature parity, we have most of the features available on VS Code. So VS Code, you'll find it often has more features compared to the other um idees and editors. And yeah, so we have the documentation that fully captures what features are available just on VS Code and how they distribute across the other platforms. So for your question, my personal answer would be VS Code simply because it exposes me to more and more um native experiences working with GitHub copilot and it basically has solved most of the problems for me so far. So I have found no reason to shift from VS Code. Uh Elvis, yes, I believe that he meant IDE. uh Kieran felt it's a bit difficult to understand but for sure when one learns it gets easier. I I do get I understand what you mean by that feels like there's always something new to learn. There's always something new that you should know how it works. But the biggest um the biggest favor you can do uh for yourself right now is just experiment. Just try everything out. pick out what doesn't work. Just feel free to cross that off your list and whatever works, just work on improving it, making it feel like it's actually making you more pro productive. But for now, yes, there's so much to learn and you can only understand it once you get your hands dirty, once you start building and play playing around with the many tools and the many agents that we're seeing out there. So if you yet to start with copilot CLI, it's quite easy to get started. Just install the CLI and then you can basically run the different um you know that you can just basically do such uh scenarios. You can cover such scenarios as we have covered today. This goes from integrating with different tools MCP support. So you can connect to your external data sources, pull in the context, and basically have these agents do a bunch of your work for you. All right, so we have our fleet session here. Um, copilot is giving me an update. So it's telling me that, hey, the review designs readiness agent completed. Again, I mentioned that I probably wasn't clear in my request here, but yeah, so the agent has done its design reviews and the co the main copilot CLI agent here is checking on the results and then I believe it's also going to uh take into account any results that we get uh or it gets and then it's going to basically just feed that back into the context. If there's anything that needs to be redone, then you can probably just have the designer agent pick that up. Right. So, the review is done. Um, most of the designs are ready. So, that's fine. Uh, the desktop app is still in progress. So, we didn't even have to wait for everything to complete. As we speak, our app is also being built. If I can open VS Code. Okay. So nothing yet, but I'm sure that the agent is still cooking and shortly we'll probably just find uh we'll probably just find the implementation done. Okay, so our design agent is done. All the screens have been completed. Okay, so app is still being built. I'm trying to multitask a lot, quite a lot. But yeah, you can see here that we have the different screens that have been created. Again, this is on top of the Excel sheet that we started off with. So, if you started the live stream with us, you do remember that we had an Excel file with some product data. And as you can see here, these designs are not pulling any dummy data. This is actually designs that reflect the exact data that we had in our Excel sheet. So think of a case where uh you know that you have some data sitting within your organization. You have no idea what that data is about. You can instantly work with the CLI to have it understand the data and translate that into designs which again now you can easily implement into the required format. Okay, we are so much over time. I apologize for keeping you all more than I should have but I think this was great. Um, we've seen we've seen so many workflows. We we've seen the copilot CLI integrate with pencil. So I think it's a tool worth trying out. It has seamlessly just integrated with the platform built out designs. Uh it's for now it's capturing the screenshots and saving them in my working directory. But if you are doing this workflow from inside VS code, the workflow is a bit different because pencil actually integrates with the IDE itself. So if you install the pencil extension, so that's pencil. If you install the pencil instruction, then you should be able to work on the designs directly from VS Code without having to use the actual pencil app. That I'll save for a different live stream. So probably when we come back next we can try and do a similar format but use the integration that's built into VS Code. But for now since we are working with the terminal we had to take that extra step of grabbing the screenshots for the different designs and then just pulling them into our into our context and our project. All right. Uh thank you for joining. I'm glad you found this to be insightful. Kiran, uh, thank you for the session. Oh, fantastic. Wishing you the very best. Um, so what are the things that make a GitHub project look professional to people who visit it? you know what? That's actually a good assignment to to give um to the CLI or to any coding agent. Just be like, "Hey, here's my project. Make it pretty or improve the documentation, improve the read me, um make it uh make it more appealing to those who visit. So Audrey, maybe depending on the exact um level of professionalism that you are referring to in your message there, you can easily just point an agent and be like, "Hey, I'm looking for this project to be probably clearer. I want you to document the code. I want you to improve the actual documentation." And I'm sure that's probably a task that these models can excel in. All right. Right. So I'm just uh just going to check our implementation is yet. Oh actually we do have it created as our dashboard folder. So we can see the implementation kicked off as we are working or finalizing on our designs. And this is actually running as uh as a fleet job. So we have two um agents running in parallel. And as you can see here the agent is still just trying to give me some updates. The desktop app agent is still in progress. So probably give it some time. We could try and see what it has just because I want us to wrap up in the next one 2 minutes. So we can just see what it has. So we can get into the Zava dashboard. Let's do an npm rundev, I'm assuming, is what we need to do. Yep. So, the app is running. Let's see where it's going to pop up. There we go. So, yeah, I've still have the boiler plate code. So, as we saw, the agent is still working on the code. So, this will probably take some time. We might not be able to see the final look. I would have wished that we had gotten to that part in this live stream, but it's it's okay. So, the expectation here is that it's basically just going to finish that workflow and we should have the data that we had in Excel in this inter interactive format. So, I'll probably wrap up here. I will share a screenshot um in a post just summarizing what we did in today's live stream. But I hope this was insightful. If you learned something cool, something new. Um, tag us. Just tag GitHub and share what you learned. But yeah, thank you so much for sticking around. I am glad that this was helpful. I will see you on the next one. We do have uh different rubber duck um Thursday sessions running across different time zones. So, we probably have one happening in a couple of hours. We'll have one happening later into the night. Well, for my time zone, but you can just check the GitHub um channel on YouTube just to see if you can catch a different session with probably a different host. All amazing people across the team. So, you have nothing to worry about there. All right. So, our agent is still cooking. Let's see. Let's see. Yeah, not yet. I'm sure it's almost done, but I don't want to keep everyone waiting. So, yeah, I'm tempted to ask the agent to hurry up, but it's probably not doesn't care so much about that. But hey, I believe that you've seen that workflow. You can do research, deep research with Copilot CLI. You can uh build implementation plans, you can connect to different tools and then just keep piling up on top of the context and have the agents um build the applications and the systems that you're working on. Okay, so I think I'll just wrap up. Uh but as I said, I'm just going to share like a summary post with uh what the agent was able to do and probably in the next live stream uh we'll we'll try and do the same workflow but from VS Code. Let's see. All right. Not yet. So, I'll probably leave it at that. Thank you all so much for joining. I will see you all on the next one. Bye.