Rubber Duck Thursdays!

GitHub| 01:06:13|May 15, 2026
Chapters10
Host greets viewers, explains Rubber Duck Thursday, and invites audience questions to shape the session.

GitHub’s Christopher Harrison riffed live on Copilot CLI, coding workflows, and AI-integration tips—plus remote sessions and practical UX for front-end devs.

Summary

Christopher Harrison (GitHub) hosts Rubber Duck Thursdays, a relaxed, improv-style session where viewers steer the agenda. He focuses on Copilot CLI as a productivity accelerator, explaining when to use CLI over VS Code and how to couple it with TypeScript, CSS, and Tailwind for real-time feedback. Harrison shares concrete workflows: build features in Copilot CLI to make it work first, then polish in VS Code, and finally push for production-readiness with linters and code reviews. He demonstrates practical tools like Copilot Code Review, Tailwind instruction packs, and the awesome-copilot learning hub to bootstrap AI-assisted development. The chat-driven format covers real-world questions—from JavaScript vs. TypeScript preferences to remote Copilot sessions and cross-project AI infrastructure. He also touches on AI governance basics, instruction files, and how to keep fundamentals intact while leveraging AI accelerators. The session blends humor (spring in Seattle, sound effects) with actionable guidance for developers exploring AI-enabled workflows. By the end, Harrison emphasizes context, iterative prompts, and the value of sharing learnings inside teams via gist shares and internal sessions.

Key Takeaways

  • Copilot CLI is best used when you want to iterate on functionality first and move to code readability later, especially for front-end CSS tasks where real-time feedback shines.
  • In Harrison’s workflow, you build in Copilot CLI to get the MVP working, then switch to VS Code to ensure readability, maintainability, and proper testing.
  • Copilot Code Review remains a foundational tool: it reads your instruction files (policies) and provides feedback aligned with your project standards (prettier, linters, tests).
  • Remote Copilot sessions preserve model continuity and context; you can steer and interact with your local/hub instance from a browser or mobile device without losing state.
  • The session highlights practical AI scaffolding—instruction files, custom agents, and agent skills—stored as Markdown assets inside the repo for reproducible AI-assisted workflows.
  • The community learning ecosystem (awesome-copilot.github.com) provides ready-made agents, instructions, and tests (Tailwind instructions, accessibility agent, Playwright tests) to jumpstart AI-assisted coding.
  • Fundamentals stay constant: humans remain in the loop for security, testing, and architecture decisions even as Copilot accelerates development.

Who Is This For?

Front-end developers, DevOps engineers, and code managers exploring AI-assisted workflows with Copilot. This is essential viewing for teams trying to balance MVP speed with maintainable, production-grade code and for anyone curious about remote AI-enabled collaboration.

Notable Quotes

""I use Copilot CLI when I don't care about the code.""
Christopher Harrison contrasts fast MVP work in Copilot CLI with deeper readability work in VS Code.
""Context is key... the fundamentals don't change.""
He emphasizes maintaining core software practices while leveraging AI tools.
""Ask Copilot... just when in doubt, ask Copilot.""
A practical tip to leverage the AI’s self-awareness and grounding in docs.
""Remote sessions... I can steer it from my phone or browser.""
Demonstrates cross-device AI orchestration and collaboration.
""Copilot is an accelerator that builds on your existing developer experience.""
Frames Copilot as a force multiplier, not a replacement.

Questions This Video Answers

  • How do I decide when to use Copilot CLI versus VS Code for a frontend project?
  • What are instruction files and agent skills in Copilot, and how do I organize them in a repo?
  • Can Copilot Code Review enforce project-specific standards like Tailwind or accessibility guidelines?
  • How can remote Copilot sessions be securely enabled for a distributed team?
  • What are the best beginner resources for Copilot CLI and Copilot for beginners?
Copilot CLIGitHub CopilotTypeScriptCSS with Copilot CLITailwind CSSVS CodeCopilot Code ReviewRemote Copilot sessionsAgent skills and instruction filesAwesome Copilot learning hub
Full Transcript
Let's see. I think I can figure this out. Hi, good morning, good afternoon, good evening, depending on where in the world you happen to be. Um, my name is Christopher Harrison. I'm a senior developer advocate at GitHub and uh this is rubberduck Thursday. So I shared this out on LinkedIn which is that basically all the adults are gone um and they left me in charge and basically gave me no instructions. So hi. Um let's have a little bit of fun. Um y'all are going to guide this. We're going to do something today. We might build. We might just chitchat. might kind of do whatever it is. I genuinely have no agenda and I have very little idea as to what it is that I'm going to be doing today. So, we're going to find out. Yeah, today is today is Thursday. Um I see people already kind of saying hi in in in the chat. I I love to see the uh uh the engagement there. So, you know, hello to uh to Kamari Nebron and to Amy and to uh LinkedIn user and to Ryan and B Samuel and Sassan and everybody else that's uh showing up in in the chat. So, let me just kind of put this out there, which is hi. Um do you have questions? What are you all interested in? What are you all working on? Like let's get in and and kind of poke around and and see what we can do. Um what I will say is that I have been um focusing uh the vast majority of my attention these days on Copilot CLI which if you haven't haven't already played with it's copilot but um in the in the CLI thus the the name um copilot CLI see it's clever um and uh honestly in particular like as a web dev I really like it because I'm able to just kind of describe hey this is what I'm And then with Vit and the automatic updates and so forth, I can then kind of see those impacts and work back and forth. Um, and then when it comes time for me to finally actually look at the code, then I go ahead and I fire up um, VS Code and away I go from there. So, let's sort of hear it. Chat, what do you what are you working on? Have you played around with Copilot CLI? Do you have questions on Copilot CLI, etc.? Put it Put it in the chat. You all are going to drive this. I you you probably think that I'm kidding like oh no like they wouldn't have let somebody come into here and and not giving them a specific uh uh agenda or anything like that but but no. Um as far as going in and like specking out prototypes, I'm not going to be clicking on any links. Um I'm uh yeah, not not going to be doing that. Um but u but if you have questions um like specific questions or things like that, please by all means go in and uh and ask. So yeah. Um, cool. So Brown asked this great question um which is one that comes up um quite frequently which is why use the CLI instead of Copilot inside of VS Code. And this is something that I get uh asked quite frequently and honestly it's a little funny because I I I'm really amazed at how quickly my flow shifted. um honestly in a lot of ways away from VS Code um and far more into Copilot um CLI. So the primary use case that I see for CLI over VS Code and I'm going to say this sentence this way on purpose. So I'm I'm saying this for attention. I'm saying this to to to like be a little bit controversial here, which is I use Copilot CLI when I don't care about the code. Now, let's talk about that for a second. Let's let's let's unpack this. Why is it that um I would be in a situation where I don't care about the code? Well, I find myself um there in a couple of different cases. Number one is that when I'm writing code, and this has been true pretty much since day one, that step a make it work. Step B is actually make it pretty. Actually make it look good. Because like when I'm getting in and I'm debugging or I'm adding in a feature, the only thing that I really care about at that exact moment is does this work? Can I actually make this particular thing sing? And then once I get that done of, okay, now it works, now let me get in and actually make it pretty. So now let me go in and start to refactor and make sure that we've got good tests and all of that. And it's at that moment that I care about care about my code. And so the way that my flow works is I'll go into copilot CLI and I'll build in there and I'll iterate in there and I'll like um be giving it specific features and this is what I want and then finally once I reach a point where I care about the code then I'll go ahead and I'll open that up inside of VS Code and then I'll start to review it and and go through um go through all of that from uh from there. The other uh really big place where I will uh primarily focus my attention in copilot CLI rather than in VS Code is when I'm doing uh CSS that I've been doing web development now for oh my gosh it's probably been close to 30 years now. Wow that's been a spell. Um, I've been doing web development for for for quite a bit of time and I can say as a seasoned web developer with confidence, I don't know anything at all about CSS. There, I said it. I don't know anything about CSS. Um, now I'm exaggerating a little bit there, but honestly, so frequently whenever I'm working in CSS, I um feel like I'm tricking the browser into doing what I want it to do. And so this is again where Copilot CLI really shines. And again, especially when you're using something like V where you're getting that automatic refresh because I can describe inside of Copilot CLI, this is what I'm looking for. This is what I want this to look like. So, I want this shade. Make it a little bit more blue. Uh, make it a little lighter, a little darker. Um, move this over here. Rearrange this. And so, I can tell Copilot CLI all of that. I can then see the results in real time. And then finally, once I reach a point where again, everything's laid out. Now, let me go open up VS Code. And now, let me actually go look at the CSS because I can kind of read it and I can structure it well. But like writing it, again, I'm terrible at that. And so VS Code really shines uh when it comes to uh to that. And so that's really my primary workflow um is I will go into Copilot CLI, let me make it work, and then let me go into VS Code and let me make sure that that code is actually pretty readable, maintainable, um and uh and all of that good stuff. And Ron Bell is up here with uh that's got shippet vibes. And I absolutely hear that. And there are certainly times when like maybe I'm just creating um an MVP like a minimal viable product or just a p a proof of concept. Um I I I I agree completely Jeremiah CSS is voodoo don't trust it. Yeah. Um but in uh in any event um going back to the the just ship it vibes I absolutely hear that. And the the couple of things I always like to highlight is again like if I'm just looking for an MVP, if I'm just looking for a proof of concept, I again I really don't care. So in those situations, I might not even bother stopping to look at the code. But um when I go to actually then deploy that out to production where now I actually care about it, now I'm going to go in and review that code. Now I'm going to actually read that code and make sure that um that it actually uh actually works there. So um yeah. So um Dylan asked the question, what about uh what about JavaScript? Am I proficient in that? Um I am proficient in that. I have no idea what this is. Um which is a a a JavaScript joke for probably a very small audience. Um but in any event um uh I do like JavaScript. I personally am a huge TypeScript fan. uh that I really like the um the the the typing capabilities in there that to me because I I really got into doing bigger projects I would say um when I got into C and I find that TypeScript is a perfect combination of like JavaScript and like the loosey goosey approach that it has and the strong typing that we get out of of um C# without being overbearing. Um and so I find that that's just like a perfect blend. So as far as I'm concerned, TypeScript is is the perfect language. Um everybody has their own opinion. Uh but I uh I really like that. So yes, so I can actually read, write and and work in JavaScript. Um although I genuinely prefer uh doing everything in um uh in Typescript uh instead. So uh yeah. And Robert here says um be clear on the end result. reverse engineer, get it functioning, then clean that up. Absolutely. And and again, you know, this I'm going to readjust, make a bigger point here, pull up my sleeves. Here we go. Um, so one of the things that I always like to highlight whenever we're talking about co-pilot and where whenever we're talking about AI is that the fundamentals don't change. that I still need humans in the loop, that I'm still going to build and go through security testing and and and my llinters and my units. It's like all of that fundamental doesn't change. And the way that I've been writing code in a lot of ways hasn't changed. like at at at its base core, I'm still step one, make it work, step two, make it pretty, which is how I've been working since I first started uh ever writing code. The part that's changed is that now I have new tools that help me be more productive and help me be more effective in that regard. So yeah, absolutely. Be clear on that end result, have a really good prompt, um, iterate, build, and then once you've got it functioning, then go ahead and clean that up. So once you actually have something, then go ahead like make it something that you would take home and introduce to your parents, so to speak, like make it make it pretty at uh at that point. So um yeah um boy this I this is a huge question. So um how would I architect a GitHub repo for a multi-product ecosystem where code documentation assets and AI generated artifacts all need to stay in sync building a product factory want to structure it the way a senior engineer would would. Um, all right. So, this is big and this is something that um is is probably much longer than a a five minute answer or the time that I'd have to to really get into it here. Um the first thing though that I I that I am going to say again is that the um that the core fundamentals again don't change. So the way that I'm going to build out that um that repository is the exact same way that I'm going to build that out without AI. So I'm going to make sure that I've got, you know, my documentation. and I'm going to make sure that it's discoverable and navigable and all of that good stuff. Where the AI is going to change my approach is in how I'm going to introduce all the extraneous bits of context, all the other things or what I like to call the AI infrastructure that Copilot's going to need in order to be successful. So, um, there's a I'm sorry by by I'm always monitoring chat. Um, and I I never have a pure straight linear chain of thought, so you're just going to have to go with me bouncing around a little bit. Um, Amy asked the question, what's my go-to AI model for web pages? This is my own personal opinion. Um, I use Claude for basically everything. That's Christopher Harrison opinion, not GitHub opinion. Um, in any event, let's go back and talk a little bit about AI infrastructure. So when we talk about AI making mistakes, yes, there's going to be instances of hallucination where um AI like does something that's factually wrong or it makes something up that that doesn't appear. Um, and that absolutely happens of course, but the most common situation where I'm getting back a wrong response from AI, it's not because the code of the response is invalid, but rather it doesn't fit in where I'm trying to work. And so, the analogy that I always like to go to, and anybody who's heard me talk has heard me give this um, a million times, which is that my partner and I woke up the other day and she says, "Hey, you want to go to brunch?" Absolutely fantastic. Who doesn't love brunch? And I recommended a spot nearby. And then she says, "You know, we've been there an awful lot. I'm kind of craving something new." And then I recommended a different spot. And then she said, "You know, come to think of it, I'm really kind of craving waffles." And all of that, of course, is a very normal human conversation back and forth is that a question was given and then based on the context at that moment, a a response was was then provided. go to brunch. Sure, here's a spot. And then a bit more context was provided. So effectively what was happening is she got the response back and it wasn't what she was expecting and then she provided more context. And that's exactly how we want to approach working with AI as well is make sure that it has the context. make sure that it has the understanding of what you're trying to build, how you're trying to build it, and why you're trying to build it that way. And so this is where things like instruction files and agent skills and custom agents and MCP servers all come into play. And with the exception of MCP servers, those other three, those instruction files, um, custom agents, and, um, agent skills, they're all primarily just markdown files. And so those then just become assets inside my repository. And I'm going to treat them the exact same way that I'm going to, you know, set up a workflow around them. I'm going to set up permissions around them, that I'm going to control who's allowed to access them. I'm going to or uh, who's allowed to modify them. I'm going to to set up a workflow around making sure that we have some validation that before we introduce changes that those do in fact represent the changes that we want to see for our project and then incorporate them um that way. So that's really it is a approach it the same way that you would and then b when it comes to all of that infrastructure make sure that you've built that out make sure that you've described hey this is what it is that uh that I'm doing and then um treat those the exact same way that I would treat other assets inside of the um uh inside the uh the repository. So yeah, hope that at least kind of points in a um uh in a great uh in a great um uh direction. So what's the typical um clean up making GitHub um look pretty? Is there um simple method or reference? I'm not I would love a little bit more context on on that question. Um, but I would say sort of like two things is the way that I'm going to approach making the code look pretty is I'm of course going to do a manual scan through and read and so forth. Um, but I would be remiss if I didn't highlight um if I didn't highlight copilot code review um that I can turn that on on a repository when I make a pull request. compiler code review can then go in and and and give my code a review. And what's great about it is that it will actually read my instructions files. So assuming that I've got good instructions files that describe this is how I want my code and and everything else, copilot code review will also go in and read that and then give feedback based on what it sees there. And again, stop me if you've heard this before. AI doesn't change the fundamentals. So I'm still going to use tools like prettier. I'm still going to use llinters. Like all of those tools exist and they're wonderful. So I'm not going to swap them out just because I've got some shiny new object in the form of AI. If they still work, I'm going to keep using them. And so I'm still going to use all of those core tools. Same that um uh that I would um uh previously. So um uh yeah. Um tamed craziness, thank you for the uh for the compliment on the uh on the nail polish. Um, this is always my uh my spring color. Um, so up here in in Seattle, spring is kind of a big thing because we actually get to um see this this yellow object that exists up up in the sky. I think some people call it the sun. Um, it's it's kind of cool and we actually get to see that again. And so, yeah, I always like to be able to uh to celebrate uh celebrate spring. So, yeah, thank you for that. Um, let's see. just kind of going through um which I wear where my ear Oh, these are just um my um my uh my iPods. Um are there any um Claude resources for for students? Dylan asks, so I can't speak to Claude, but what I can speak to is this. Let me go um real quick. Uh search this up. Uh GitHub um Copilot CLI for beginners. Um, and now it's just playing in my headphones here. I don't know if it played out there. Um, but in any event, let me whoosh. Let's go like that. And yeah, um, if you just do a Google search, uh, there is a, um, full video series uh that I did with the one and only, the amazing uh, Kayla Cinnamon uh, where we go through Copilot CLI for beginners. Uh I would also be uh remiss if I didn't mention uh Cadisha's fantastic course again more um more music there. Um she did a wonderful uh copilot uh for beginners as well. So if you're just looking for core go uh co-pilot absolutely check out Cadia. If you're looking specific for copilot CLI then again with the music uh check that out. I have no idea whether or not the music is actually being um shared out or not, but uh but there it is. So, um yeah, cool. Let's see. Well, that I was at 1 point. There we go. Sorry. Again, the the the adults left and provided very little guidance. So, I've got buttons over here. I'm not entirely certain what they all do, but we're we're we're figuring it out as we um as we go here. So, um yeah. So, let's see. Um Alli asked a question. Is it possible to create job opportunities for individuals who rely on artificial intelligence in a professional manner? And I'm gonna Ally, if you don't mind, I'm gonna I'm gonna uplevel this question, which is the kind of the higher level of like, hey, um, what is this going to mean for developers? Because this is another big question that I get asked quite frequently. And I always want to highlight the fact that um well a I'm speaking for myself but also b that like this is the way that I currently see the world and I will say that for me copilot is an accelerator that allows me to build upon all of my experience as a developer. So when I'm getting in and I'm creating code through or with the help of C-Pilot, everything that I'm doing, the the prompts that I'm creating, the view that I'm taking on the code itself, the different redirects that I'm going to give to Copilot, the architecture that I'm going to have it build out on, etc. All of that is 100% based on my experience that I I would not be able to create what I'm able to create at the level that I'm able to create it if not for the experience that I have as a developer. So when it comes to like job opportunities and and using AI and so forth and so on, 100% that I would not be the developer that I am with Copilot if I wasn't the developer that I am without C-Pilot. So um yeah, that's sort of my um my my my my opinion there. Um let's see. Um Ryan adds in um build systems single source um and then treat GitHub as the backbone even when it's not. Yeah, absolutely. I think that's I think that's a fantastic u fantastic approach. Um I'm going to throw this up here. Um, so I'm obsessed with telling my AI to make my site SEO, Lighthouse 100 Plus, um, PWA. Understand that they have their own use cases, but what are you guys using to make sites functional? I have a resource, uh, for this. Um, let me do this. Let me fire up. Um, there we go. And I make sound effects as I go on about my uh my business here. I don't know if anybody else makes sound effects while they write code and debug and and so forth, but you're getting to hear that. So, awesome copilot. So, awesome-copilot.github.com um is an absolutely wonderful resource. So on here, this is um a communitydriven site uh that is maintained by um uh by by Aaron and who's wonderful. And what you're going to find on here is a full collection of different agents. You'll find a full collection of different um instructions and you'll find a full collection of different skills all set and ready for you to go. And so if I go in and I do a search here and I do a search for, let's say, SEO, what you're going to notice is that there's actually a custom agent that's already been created for helping you generate um SEO. And so I can use this as a starting point into going out and building or doing my improvement. You'll also notice that there is a button here to do an installation inside of uh VS Code. So, I can just go ahead and hit that little thing. Uh go ahead and just download it, copy it, and then bring that into your project manually. But one of the other big things that I like to do is this is let me um just going to be a little bit bad there. Put that into um YOLO mode there. This is probably going to have me do an update, but that's going to be okay. Um, cool. And let me do this is take a look at my project and at awesome copilotgithub.com and let me know if you see any custom agents agent skills or instructions uh files that would help you generate better code or perform as a better AI assistant. Okay, there we go. So, um All right. Uh don't you can ignore the the little Lost um thing that I've got there. That's uh irrelevant to what it is that I'm doing. So, this is one of my favorite things to do is if you're anything like me, and I know I am, I will be in the middle of something and then I'll have this great inspiration or I'll realize, oh, I really want to get in and um and do something. So, let me go out and just like real quick as an aside, do this. And one of the biggest places where I find that is when I'm working inside of Copilot that I'll be getting in. I'll I'll be doing whatever it is that I'm doing and then I'll realize, oh, I've added on this new feature or I've taken a dependency on this new framework. So, let me now go ahead and figure out ways that I can get Copilot to work better with that. Again, context is is key. And so, awesome. Copilot is awesome. I mean, it's it's it's right there in in in the name, but there's so much that's in there, and a lot of times I'm not entirely certain of like, okay, well, what's in there that's really going to help inside my project? And this is I I think sort of the great little like tip. If if I was going to give you any one tip to take away from today's session slash babbling, it would be to ask copilot. Just when in doubt, ask copilot. So if you don't know whether or not copilot's able to do something, ask copilot. If you don't know how best to approach something with the help of co-pilot, ask copilot. That copilot knows itself pretty well. And you can also like if you want to make sure that it's grounded in the docs, specifically tell it, hey, I want to make sure that you're able to do this, make sure that you go look at the docs on GitHub or like do those types of things. And it's really then going to help it out. And this is a perfect example of that of, hey, I might need some additional context that could help C-pilot out. So, I'm just going to tell Copilot, hey, take a look at my project and take a look at um awesome co-pilot. Let me know if you see any custom agents, agent skills, instruction files. They're going to help you out. And so, now you'll notice that it went in and did a um search and it came back with a whole bunch of items here. So, it checked the project layout. it realized, hey, you're using um Tailwind, so this is a direct match. Let's maybe go ahead and grab that instruction. Um your website is public facing, so maybe bring in the accessibility um instructions. And it has a couple of others in there uh to bring in. As far as custom agents goes, uh you'll notice that it's got that accessibility agent that we uh talked about before, the front-end performance inspector agent, maybe the playright tester, and so forth. And then I could go in and just tell it, hey, yeah, go ahead and uh and grab these. So, let's maybe grab um uh grab the playright um generate um test and the Oh, what's another one that we could go ahead and grab? Maybe the uh ally or the accessibility agent um agent and uh let's also grab uh the Tailwind one uh Tailwind instructions and put them in the right place. Perfect. Okay, so I go ahead and just tell Copilot to do that. We'll give that just a second here. As we all know, moving your mouse makes it uh go faster. So, we'll go ahead and download those uh those files. And give this just a second. We're getting there. We're getting there. We're getting there. There we go. Perfect. And now I can go in and um another little thing if you haven't already seen inside of Copilot CLI, if you hit exclamation point or bang, uh that will put you in a shell mode. And so now I can go ahead and execute a shell command. So I can open this up inside of uh VS Code. And then now if I open up my GitHub folder and now I go into my instructions. Now it's got that uh that tailwind. And so now I can go in and uh and take a look at uh at that review that make sure that that's going to behave the way that uh that I expect it to. I've also got this skill right here on generating tests. So now I could go in and uh and take a look at this. Might actually want to expand that out uh a little bit. Make that a bit more uh powerful than uh than it currently is. And then you'll also notice on myops um on my agents. There we go. There is that uh accessibility agent. So might want to go in and kind of play around back around with the the tools that are up there. And then I can see everything else uh that uh that's in there. and I can go in and do my review and uh and so forth. So, yeah. So, awesome. Copilot. Um great place to go in and uh and find uh whatever it is that uh that you might want. So, cool. All right, let's go ahead and stop sharing. Uh I probably have a billion different messages now. Um I do. Let's see. Um, please send your connections folks uh out of invites. Um, uh, no. Um, yeah. let's see. Yeah, Cloud cannot run locally. That is that is true. One great advantage to Copilot is that uh you do have a pretty good choice of models. So like baked right in you have access to um to claude you have access to open AAI um if you're inside of VS Code you even have access to uh to Gemini uh and uh both in copilot CLI and inside of VS Code uh you have bring your own keys so if you want to use your own model then you have that uh that capability as uh as well there. okay, other questions. Lived in Vancouver for years. Welcome to the sun. Yeah, absolutely. Absolutely. Yeah. Like I I adore Seattle. I really do. Um, like there's a shade of green up here that you just don't get in in in a lot of other places. And of course, it's only that shade of green because of the amount of rain that we wind up getting. We're not really, it's not so much the amount of rain because if you like go look at the measurements and so forth, Seattle isn't um I don't even think we're in the top 10 for like the total accumulation of rain. It's the number of days and it's just that gray mist that will just grind at you. So, come about March, come about April, you're you're you're you're ready to uh to see the sun again. So, yeah. Okay, cool. Glad that it um uh that it didn't um uh that didn't play music. So, okay, cool. Um other questions, please throw them in there. This is this is it. Like I said, y'all are driving the show, so I have I have no agenda, hidden or otherwise. So, please, by all means, ask ask questions if you've got them. I'll I'll I'll make up answers. Talk about my collection of um um Funko Pops back here. uh my uh my cal uh it's always the the the the mirrored um camera here is throwing me off, but uh but yeah, I've got my Cal Raleigh and then I've got um uh the two from uh from Ted Lasso. So, but uh in any event yeah, Victoria and and Seattle kind of get the uh the Gulfream. Yeah, it does generally kind of like miss uh miss Vancouver. So lived in Seattle uh in June and only saw the sun for uh and saw the saw only sun for 38 straight days. Yeah, that is so this is something that that is genuinely underrated about Seattle and the Pacific Northwest is that there is no better place on earth than July and August in um in Seattle. Now granted, um, climate change is having a bit of an impact. Um, we do get hotter temps than we once did, and sadly when forest fires crop up, um, the smoke can become a bit of a challenge. But that being said, um, July and August typically are wonderful because the highs will be in the uh the 80s, maybe like low 90s. Um, and that's about it. And it's sunny, it's bright, we don't get rain during that time. It's wonderful. So July, August, Seattle is absolutely spectacular. Um, so um, so yeah, let's see. Um, Copilot has a new feature. Could I demo it? Agents window in stable work in an agents first way across all your projects in the new agents window. Is that are you talking about on um on github.com? Is that what you're talking about? I think you know what I want to do? Give me two seconds. Talk amongst yourselves. I'll give you a topic. The industrial revolution was neither industrial nor revolution. Discuss. Um okay. Let me let me do this. Let me um Okay, this is a little just side project um that I have. And I think is this the No, hold on. I'm getting there. I'm getting there. Um, I think this is where I want to be. Yeah. Okay. Um, hold on one second. I need to kind of play around. Okay, I apologize. I'm needing to like clean up a couple of things on my GitHub here before I start sharing my screen. So bear with me for just a moment. Um there is that and let me do that and okay I think this is Yeah. Perfect. Beautiful. Beautiful. All right. I love it actually. Yeah. This is going to be wonderful. All right. Let's go ahead and Oh, that's been on the screen forever. That's all right, though. Let's go ahead and hide that. This is my second monitor over here, so that's why I'm looking off to the side. I want to add to the stage. So, I think this is what you're talking about. Um, so what I did here is I went to uh github.com/copilot aents and this is my agent screen. So, this will show me all of the agents that I have worked with basically across every single project. Um, and what's cool about this is this could absolutely be for agent jobs that I have assigned to copilot coding agent. So, if I have gone in and I've um assigned a job here by just like choosing copilot, hitting send, and letting it run off in the background, and now it's going to be doing its thing up in the cloud. That works just fine, and it will show up uh globally inside of here. But one other really cool thing that you can do in Copilot CLI is you can um set up remote sessions. And so with remote sessions, I can actually have copilot CLI then become accessible to me to me wherever it is that I have access to copilot. And so I can go slash remote on. And then what I'd be able to do, and this is actually that little request that uh we had made um just a few minutes ago, is you'll notice that I can actually now see this right here inside of my browser. So, I'm able to access this and I can see that session right here inside of um inside of just github.com even though I started that here on the client. And I can even go in and go remote on which would allow me uh whoops remote on. There we go. And so now you'll notice that remote steering is enabled. And so now I can even It is always amazing to me how Chrome picks picks the exact wrong place to put the tab. Put it over here rather than putting it over here where I was last. Anyway, um so let's see. Now that I have remote on, are you going to let me steer? You are. Yeah. So now I can actually steer it from um from here. So if I remember right that playright generate skill um that playright um generate skill looks a little uh light. Uh can we add some additional uh capabilities uh to it to enhance its ability to create uh playright tests. There we go. Chug chug chugga. Um, is slash remote available on all subscription tiers? Uh, yes. Yep. I think it remote might be in like a considered in a public preview or whatever phraseology it is. Um, but um but yeah, so you'll notice that I sent the message here. So that the the playright scale looks a little light. You'll notice that it's doing its thinking. It's uh you know going to do its thing. So it's doing all copilot things. So, I want you to just take a mental picture of this right here and notice all the text there. And now, let me sneak back over here. And now, what I want you to notice is there's that text right here inside of my locally running instance of Copilot CLI. So, this is allowing me full remote control over my local Copilot CLI session. And again, it's secure. It's scoped to just me. So, I'm only the only one that's able to to access it. And what I really like about it is I can even then um fire up. Let me actually just do the thing. I don't have to just talk about it. I can do the thing. Let me um let me open up GitHub. And let me go to copilot. And yeah, here is we'll see how well this shows up. There we go. Yeah, there is actually my session then right there on my phone. And sort of like as a a perfect real world example of this is I was working on something um yesterday and I uh went to the gym for lunch and so like real quick before I went in and did my workout. I sent Co-Pilot a real quick message like literally like through my phone like hey can you go ahead and take a look at this for me? And so I went in I did my workout came back out and it had built the thing out. I reviewed what it had done and then gave it yet another command and then on my way home wound up getting a flat tire and I wound up getting stuck there for about an hour or so and so like all right well I can still be productive while I'm sitting here I got nothing else to do and so legit I'm now just on my phone typing away and sending co-pilot commands all through that so that remote capability is absolutely spectacular so now I can both like see what my session is and even be able to um steer that from wherever it is that I have access to co-pilot. So I can do that on on the on the web page and I can also do that on my phone which is uh which is wonderful. And um tamed craziness um asked the question, let me just throw that up there. Does remote keep model continuity um the same models doing the work? Absolutely. Because it's actually using my copilot CLI harness to uh to do its thing. So, it's going to be the same models, have the same um context, and everything else. So, I can send a command over here, send a command over here, and work back and forth, and it's all going to have that same context. It's all going to have that that same model, etc. Um, speaking of sharing things out, one of the other real big things that I um really like is slashshare gist. So when you're creating code with copilot the prompts the back and forth etc are potentially artifacts and you know there was a question that was asked earlier about jobs for developers and using AI and things like that that one of the skills that developers now need to learn is how best to work with and interact with AI. And those are skills that also then need to be grown and developed inside of my team because the projects that you're working on and the approaches that you're going to take, you're going to start to learn new tips and tricks and so forth of how it is that you did different things inside of Copilot. And it's great to be able to share those out. And so one of the big things that I always recommend for all organizations, for all teams, is keep your training and keep those lines of dialogue open at all times. And so that can be done through like a dedicated Slack or Teams channel where you allow people to like share their little things that they've discovered uh when using AI when when using Copilot. Um, but also do this through like lunch and learns or if you do like an internal conference, then like have a session where you bring in three, four developers that like shares something up that was really cool and really unblocked them through using copilot. And a great way to like build upon that is this little shared gist. So just assembly forward/shair just hit enter and then I poof just like that I now have Why do you keep throwing things over there? Um in any event so now here is a gist of that co-pilot session. So if I kind of keep scrolling on down it's going to show me all the messages that that co-pilot was sending. It's going to show me all the things that I sent up. It's going to show me all the different commands and tools and so forth and I can drill in and I can see all of that right there. Um, this I I think there there's a whole bunch of like little switches and so forth that I just adore inside of Copilot CLI. I think the one that does not get near enough attention is that forward slash uh share uh that being able to just like real quick, hey, this is what I did. Share that out in a gist or share that out in a markdown file is is spectacular. So yeah, I um I really like it. So Brian asked the question, what's up? Um honestly, I don't know, but we're here. We're having fun. We're answering questions. We're poking around at things. So, um, yeah. Um, Dylan asked a question. Am I the only person in the world who uses Edge? If I'm on Windows, um, I use Edge. Uh, but on a Mac, I use Chrome. I don't know. I'm sure there's some rationale behind that. Um, that makes logical sense, but yeah, I use Chrome on uh when I'm on my Mac and I use Edge when I'm on uh on Windows. So yeah, but I would also say this is, you know, when it comes to to browsers, I this is completely tangential, but like I I only care about one thing. Is it does it does it browse? Does it does it render HTML? all the other features of like, you know, tab grouping and and and and and injecting AI and in into a browser and like all the other like little things that the browser can I I don't care. None of those are my use case. I I I just want my browser to be a browser and that's it. And and maybe I need to go yell at the kids about being on my lawn. Um in any event, okay. Um cool. I'm going to change my um my dialogue here because I or my screen here because I realize that I am on the screen. Okay, cool. All right. Um let's see. Um other questions that people might have. I got about 10 minutes left before they're going to give me the the the hook here. Um how about on on Linux? I I don't have a Linux desktop. Um yeah, I'm I I I have my work Mac and I've got my PC which I use for gaming and that is uh that is basically it. I I tried Linux. Um again, I'm going to be dating myself. There was a a a dro of Linux that I really liked um called Caldera and it was using the um KDE desktop which I thought was pretty slick. Um, but my absolute favorite part of it, this was, you know, back in the good old days of like um needing a a CD or DVD. Yes, I am old enough to remember floppies as well. Um, but um you would then pop in uh the the CD uh or the DVD and let it start installing off of that and copying files would like literally then take it like a good three hours or something like that to go ahead and and do its thing. And what I loved about the installer is it would ask you the various questions about like what features and locations and what you want your root password to be and all those different things. And while it was doing that, it would already be copying files locally onto your hard drive. And invariably, there would be some gap in time from the moment where it asked you all the questions till it was finally actually ready to use. And depending on the version that you had of Caldera, it would pop up a different game. So it would like give you a Pac-Man clone or it would give you like a Tetris clone. And so you could then just like sit there and like play Pac-Man or play Tetris while you were waiting for it to install. And I just always loved that. I just thought that was like a really cute way to like, hey, you can be sitting here for a while. Here's here's something to do. So, um yeah, in any event, um yeah, Amy does highlight, yeah, some some websites don't work properly in certain browsers. I generally find um great success with uh with Chrome and and Edge that those are the the two again on on Mac and uh and Windows. So, but um but yeah, other other questions. I got nine minutes left here. Ask questions, other thoughts, comments, questions. Don't forget to hydrate. Very important. You know that water bottle that you filled up and you put on your desk, it's sitting next to you and is full. Make sure you drink out of that. That's specifically for one person who I think is still watching. He knows who he is. Other questions? We got nine minutes here. I'm gonna go look at the um update. Okay. Um do I have a demo share? I This is sort of Yeah. Uh I This is kind of what I've been doing. Um it's just this I I like I I I was not lying up top is I I 100% like had no agenda. So I mean I could go ahead. I mean I've I've kind of done a little bit of a demo here. Um, what else could I do that I really like to um to show off? I did remote. I did share. Um, oh, um, I could do this. Let's Let's do this. That's not where I wanted you to be. Boop. Um, and he'll spin toys. All right, let's do this. Um um CD valard I I see your question. I'm going to send co-pilot on a task here and I'll come back to I I see your question though. Um copilot um let's do this and let's do this is okay. I'm going to say plan. So, I talked earlier about context, and context really again is key. And one great way to help Copilot understand what you're trying to do, how you're trying to do it is to go through a plan stage that yes, Copilot will typically do that automatically if it realizes that you're going to be doing something complex and like it it should plan. But one of my core tenants whenever working with AI uh is don't be passive aggressive with it. And what I mean when I say that is if there's a piece of information that copilot's going to benefit from tell co-pilot. So like can it find a particular file on all on its own? Yes. But if you tell co-pilot about it then it's guaranteed to be able to to to find it. If you want it to do something a particular way, will copilot figure out how to do that? Yeah, there's a pretty good chance it will. But if you tell Copilot, then it absolutely will. So don't be passive aggressive. And so along those lines, like absolutely right out write out a plan. And one of the big things I really like to do is plan like let's say um add um filtering uh to this website by publisher and category. Let's create uh a plan and uh rubber duck it. So um rubber ducking the uh the namesake of this stream right here is about taking a um rubber duck. I actually have a a little tiger um rubber duck. Is that focusing? I think it is. Um I have a little tiger rubber duck right here. Um, so, um, um, uh, your your your whole thing with rubber duck is when you're like troubleshooting, you just go ahead and talk to the rubber duck and it will give you then, uh, responses back. Um, responses back. It's really just a way to, um, uh, to talk through a problem. And a lot of times, just saying it out loud will uh, will really help. And so, um, what rubber ducking in this instance is going to do is I'm using Claude here is it's actually going to fire up a GPT model once it builds out the plan and then run that through the GPT model to get a second set of opinions on it from another model. Um, which then really kind of helps make sure that we're building out something the right way and and do it pretty strong. So yeah. Okay. So let's get to this question. Um sorry to ping about it again. No no no apology necessary. Curious my thoughts about um starting on AI today. I'm front-end engineer and want to automate more design dev glue work and don't really have any idea where to start teaching myself beyond agents and skills. This is a fantastic um question here. Um, and I'm going to go on full screen here. Um, right after I just highlight, you'll notice right here, there's that rubber duck call right there. And so now it's going to critique the filter, the feature plan, and then it will uh incorporate any of its findings into into that plan. Okay, give me just a second here to kind of just rearrange a couple of things. Um, there we go. Perfect. Okay, let's see. So, um, all right. So, let's bring this back up. So, curious about thoughts on starting with AI front end engineer. Um, needs to automate more design dev um um glue work. Don't really myself beside beyond agents and skills. I highlighted earlier um the couple of courses on Copilot for beginners and I would definitely recommend seeking those out if you just fire up Google real quick. GitHub copilot for beginners. There's a couple of different YouTube series that are out there. Uh there's also a great learning hub on awesome-copilot.github.com. Let me share that out. Um awesome copilotgithub.com. There's a learning hub that's there as well, which will have a whole bunch of resources to get in and and start exploring from there. Um, here's here's my biggest tips for getting started in working with an AI coding assistant. Number one, and I know I said this about five times already, but it bears repeating. I'm going to say it one more time is always remember that context is key. That if you see that co-pilot isn't giving you what you're expecting for one reason or another, it's not following a pattern that you expected. Maybe if you're a React developer, it gave you a class component rather than a function component. um it didn't follow some guideline or or otherwise or even in instances where it's it's hallucinating, it's because it didn't understand the full environment in which it's working. It didn't fully understand your request. It didn't understand the context. Again, going back to that very simple example, you want to go to brunch, absolutely recommend a spot based on the context that I had. Perfectly valid response. And then more context was shared. And so that's a very big thing then that you can do with copilot is help it understand that give that more context. The second thing is to keep picking right up on that analogy is notice that back and forth that all of these are trained on LLMs large language models and the whole design of an LLM is to be interactive is to have that back and forth. So if I see co-pilot not doing a thing that I expect, I want to try and figure out why and I want to help point C-pilot in the right direction there. So just because you weren't able to single shot something doesn't mean that it was a failure. It just means that there was something that wasn't quite right in the prompt or in your instructions files or otherwise. That's okay. Just rephrase, reframe, add more context and help point C-pilot in the right direction. continuing to build on that. Ask co-pilot questions that if you want to know like what's the best way to build something or what's the best way to frame something or what's the best way to approach something, don't be afraid to just ask Copilot those types of questions and it will really be able to help you out with that. And so like the the through line in all of that really is all about really kind of that back and forth context of making sure that Copilot understands the requests that I'm making and what it is that I'm trying to do and how it is that I'm trying to do it. but also at the same time is me getting more experience in how Copilot's going to behave and how it's going to react to that environment and and the environment in in which it's it's it's working. And to go one step further on this and I'll I'll kind of finish on um um on on on this. Um Brian says, "I've been on camera long enough I can go back to work now." Um and and so I'll close on this which is again to say to experiment and to play that these are tools and you're you're never going to be an expert at a tool the moment that you start playing around with it that this this takes time to get in and and to explore. So absolutely play with it, work with it, um continue to to iterate and to build and if you're part of a team, ask your team members, share experiences with one another, and so that way you'll all just kind of continue to uh to build from there. So yeah, this was rubber duck Thursday. Um I don't know whether or not the adults will ever trust me back here again, but in the meantime, um I had an absolute blast. Uh, I hope all of you had a good time as well. You can find me on uh on LinkedIn. Um, if you just do a search for Geek Trainer or do a search for Christopher Harrison, uh, you should be able to uh to find me on there. Absolutely. Come come find me there. Come chat. And, uh, thanks for taking a little bit of time out of your day to come hang out and babble about Copilot. So, yeah. Thanks a lot, everybody. Bye-bye.

Get daily recaps from
GitHub

AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.