SPEAKER_01: 0:07

Hello, welcome to Microsoft Community Insights Podcast where we share insights from community experts to stay up to date in Microsoft. Today we we would we will dive into GitHub GitHub co-pilot agent. So could you please introduce yourself, please?

SPEAKER_00: 0:21

Yeah, sure. Thanks for having me. I'm Johan. I've been a developer for nearly 30 years now. And I do a lot of work on the.NET space. Uh, I work for a Dutch company, Burkler, uh, where I am a principal consultant. And yeah, I do a lot of things uh around.NET, but I do start to get more and more involved in using AI in different companies, and where even different companies ask, Well, can you please train us in how to apply AI in our development process? Because yeah, simply booting up Copilot and simply start adding some prompts to it. Well, it does work, but it's not that well performant. So uh yeah, I I really like like what I do. I like what the industry is moving towards, and I know that's a quite a scary thing for a lot of developers who think, well, okay, um will AI replace me? Well, as I always tell during trainings as well and during all kinds of conferences that I speak at, I don't believe that AI will replace us, but you will be replaced by a developer who knows how to manage and work efficiently with AI if you don't pay attention to your skill set.

SPEAKER_01: 1:52

Yeah, so I think uh what Johan said is that you need to learn AI in order to keep up to date with the industry, because if you don't know AI and someone else knows AI, you they will be above you. So you'll still have to learn how to use it. Because even things like Git Co-Pilot, it still works with.NET, so you still have to make sure it's best, use it the best of your ability. So think of it like a like assistant or in another best friend.

SPEAKER_00: 2:25

Oh, definitely consider it the best comparisons I can give is compare it to a junior developer joining your team. Yeah, a junior developer can do wonders for your team, but a he or she still needs a bit of steering and a bit of coaching to get the job really done and the way that you really want it. Because it's pretty difficult, and there are so many different variables to consider when building a software system. Uh, simply allowing AI to generate it for you, okay, it will come up with some kind of solution, but will it be the solution that you are really looking for? And not only just now, but will it actually work in the future as well?

SPEAKER_01: 3:15

Yeah, and so that's why it's important to have human in the loop, whether it's to check the code to check whether it's correct, the one that you intended to. So it's very crucial. So last few months, so I think last year there was GitHub universe released, and there were so many changes happen on the universe with lots of like introduced with custom code palettes and stuff. Before we dive into like GitHub code, custom code palette. Do you want to just uh tell us how do you use GitHub Code palette at your workplace on day-to-day?

SPEAKER_00: 3:51

Okay, how do I use it? It it really depends a bit uh to what extent I can use it. What I do see is that the adoption, at least in the Netherlands where I live, um in companies is well changing a bit. Uh, some companies are quite reluctant to actually allow developers to use AI, and then I hope to convince them that it really will help you in your workflow and will improve your performance. And then I don't can't use it that much. For other companies that I work for, it's just fine, and I can use whatever I want. And I must say, the biggest advantages and the biggest knowledge I get next to my regular job, I also do some work for charities because they simply don't have the money to pay for a consultant, and they do have some kind of software system that they would love to have built, but that's just me, and I only have limited hours. So, in those projects, I really let AI help me and even let it steer a lot of the development because then I can make good progress in the limited hours that I have.

SPEAKER_01: 5:11

Okay, that's amazing. So, is there any particular features or aspect of GitCopile agent mode that you like? Or you use the most?

SPEAKER_00: 5:21

What I like the most, at least what was released recently, is the combination between the plan and the agent mode. Agent mode does a lot, but if you really want it to research your code base and really wants to come up with a good plan before it starts actually changing code, I love the combination of starting with a plan, making sure that the plan is correct, and yes, that is your responsibility as an engineer who is actually entering the prompt, and then from the plan, start in agent mode and start really generating the code that you really want. And I must say, if I look back at about a year ago, I was not that confident about the code that got generated at first. But if I now look at the code that is produced, well, I really like it most of the time. And yes, you there are still some bugs in it, and yes, you still have to steer it and update it a bit, but it's quite decent code, and that's what I really like love about it as well. The actual quality of the code generation is so much improved over the last months that I would say it's now really usable. Where it was once okay, nice to have for generating a test, uh a unit test, but now you can really work with it, and you can really work with it on larger uh topics. But yes, you still need to steer it, and yes, you still need to think about how will I divide everything, because simply putting in a project plan and hitting the run button that still doesn't work.

SPEAKER_01: 7:12

Yeah, it's it's similar to the phrase garbage in, garbage out, depends what you put in the prompt. You get something bad return because you still have to like adjust the prompt how you want it, adjust the custom like agent how you need it to be, whether it's a sales agent, you just as long as you adjust it, you'll get what you need.

SPEAKER_00: 7:34

Yep, and that's what good is as well about the change changes that are coming now to GitHub, or are actually part of GitHub, uh, that you can create those custom agents where you can actually configure them what you want them to be able to do. And otherwise, you would simply have the regular agent have to do everything, and yes, you can uh hook up some hook up some MCP servers and hope that it works, but you do need to have some a bit more steering and possibilities to actually adjust, and I must say, in the newer versions, that really is a big advantage. Using it uh on a new project. There's also videos about with uh from the team how to apply it from uh an existing project because let's be honest, uh, a lot of us don't work on Greenfield software anyway. Uh so you do have your existing software, and they even have some guidance how you can apply it in your um existing software. What I did, because you will see that the video paused, uh I did pause it on purpose. There is one downside on SpecKit. Uh, if you of course look at the premium requests that you have, it will use a lot of premium requests. So that therefore I stopped it at a certain point when I thought, well, okay, uh you came quite a somewhere that you can actually see what kind of interface that thing can do and build automatically uh without burning all my uh premium requests because they will only get renewed by in February 1st. So I do have to have some premium requests left to do my production work. Um but this really is quite a nice thing to shoot to to keep in mind. This is this is preview. I can't tell you what will happen, it's still a Microsoft product, it might get cancelled, uh it might be evolving into uh release. I'm not quite sure, but it looks really really promising. So I if you are really into pipe coding and having an AI do some of your work, do keep a close eye on SpecKit and how it evolves in the future.

SPEAKER_01: 10:14

Yep, and you can do that by just I think just keep an eye on the change logs on the GitHub itself and see what's been changed.

SPEAKER_00: 10:25

Definitely, that's that's what I do for a lot of uh projects. Uh I like that Microsoft is really into open source and that most of the projects that I use on a day-to-day basis are fully open source. Uh, and I always keep a close eye on the issues list to see what is changing, but even if you don't want to uh double check the repo from time to time and see the teams are quite good at keeping uh everybody updated on the new progress that they have made. So simply keep an eye on the uh the repo and see how it evolves. And I think this is very, very powerful in the near future.

SPEAKER_01: 11:10

Yeah, and once Backit's been evolved, you can do a lot of things, even to production, you can create production like pro applications with it as well, and integrate it into your CICD development lifecycle.

SPEAKER_00: 11:24

Yep. So it this really could help you a lot with at least part of the applications that you have to write, and then you will get the situation where people were talking about years and years and years ago when I was um well, I won't say when I first joined the industry, but there have always been developments where you can actually specify what the user wants and have systems being generated. Well, I think if I now look at the status of AI and look at things like SpecKit, that can come through to at least for some portion of the applications. There will always be parts of applications that you will have to work on your on your own to make sure that they comply with everything that you want. But a lot of simple applications, things like speckit, can really do wonders.

SPEAKER_01: 12:26

Yeah, that's amazing. So uh is there any last words that people want to you need to for people that don't know about spec key or GitHub uh copilot that you want to tell them?

SPEAKER_00: 12:39

Um the best advice I would give you simply start experimenting with it. Uh you can get a free uh version of GitHub Copilot, and no, you can't do everything that I showed you with it because you do need a premium version license for that. But please be kind to yourself and please start investigating in how you can apply AI in my day-to-day job. Because just repeat myself, you will not be replaced by AI entirely, but you will be replaced by a developer who knows how to use AI in the future if you don't keep up to date. So that would be my call to action. Start working with it, start experimenting with it so that you will get some knowledge, and just keep a close eye as well on Microsoft Learn. There are some very good modules there where you can learn how to apply AI and learn how to work with uh GitHub Copilot, and even uh, although that are Microsoft MVP, uh also sometimes from time to time look at what the others are doing because, of course, Microsoft with Copilot is in close competition with all the other uh AI uh applications. So do keep an eye on what's happening there because this is not going away, this is will be part of the development experience and development workflow. So do get to know how to use it so that you will not be replaced by a developer that did take time to learn how to use AI.

SPEAKER_01: 14:29

Yeah, that's amazing as well. So thanks a lot for joining this episode, John. So I hope everyone learned something about GetCoPala Customer Agent and how useful it is, and learn about some like uh expect kit as well. So you can just try it out in your own time and has it evolve. The GitHub product will be evolved, and you can do amazing things in production, and you can also use it as a prototype as well to demonstrate something like for like hackathon as well. So it's very good. Okay, no Rice, and see you tuned to the next episode. Thank you. Bye. You're welcome, bye bye.