Podcast forThe Podcast for
Tech Enthusiasts

Welcome to a new episode of Tobioth Zwei. As you hear, this is a premiere for us. It's the first podcast in English since we have a guest. Welcome, Karthik Rameshkumar, field CTO of GitHub. In this episode, we want to discuss AI coding, what is new on the horizon, and also what skills do we need in the future. So even if you are not a developer, this episode might be useful for you since we're talking about how to apply skills to your daily doings. So without further ado, welcome, Karthik.

Hey, thank you so much to the two Tobys for welcoming me on the podcast. I think the conversation to get on this podcast started a long time ago, but I'm so grateful to be here with all of you and be able to present whatever little I can share and the insights I can share with all of you. I'm looking forward to having a nice conversation.

Thanks. Thanks a lot for being our guest. Yeah.

Big pleasure. And for the audience, Karthik is sitting in front of us with a t-shirt from a German soccer team. Very, very good.

So I was just telling you that actually finding the German soccer team's t-shirt, this is special for the Tobbys, but finding the German soccer team's t-shirt is actually really difficult in Bangalore. I had to buy this one in Bangkok myself personally. So it usually goes on sale and sells out on day zero. So yeah, I'm a big fan of the German football team. Yeah. So let's see. World Cup coming soon?

Awesome.

Yeah, Karthik, what is a GitHub field CTO? What is your job, your new role, by the way?

It's a fantastic question because I think I'm trying to actively discover it as we are building towards it together. In a lot of ways, I think the GitHub field CTO is sort of a strategic stakeholder for all of the engineering leaders in the region for them to be able to have a singular point of contact on the GitHub side. I think the intent with this basically is to have someone that can have high-level conversations with stakeholders on the customer side, was the internal teams that we can then build a bridge and a liaison for a product features that we're trying to build, strategic conversations that we're trying to nurture around what direction a product should lead. And 3, to also feedback all of the signals back to our engineering team so that we can then refine our processes and build better software that our users will end up using a lot more. So I consider myself a custodian of our customers' experience on the platform. And my job is to have conversations with developers, have conversations with engineering leaders, have conversations with people that are part of the development ecosystem, testers, all the other people, right? then feed that information back so that the product team then is then able to take a holistic decision on prioritizing what features need to be delivered first to our customers. And in addition to that, I also help to sort of drive a little bit of understanding on where our thought leadership around the developer lifecycle basically comes for GitHub, right? I do a lot of talks. I do a lot of writing on LinkedIn. I do a lot of scrambling, sort of scribbling myself, sorry, to sort of write what I feel like my thoughts. And this sort of helps put the message out there in terms of what the development lifecycle looks like and to share a little bit of my two cents on where the world is going and what my observations are. And I think of late, the thing that I love about my role is that I think I sort of become like a best practices disseminator, right? So I get to talk to a lot of customers who are in the same boat. So Customers are very curious because sometimes when you're competitors, you don't get to talk to each other. But then at least with me, you can at least understand industry trends, right? Hey, if you're a bank, what are the banks thinking about? What direction do they want to go in? What does good look like is a fantastic role that I get to play. I think I'm excited about those roles together to sort of bring sort of insight into these boardroom discussions with leaders and help them build better software together with their teams, I think it's a very exciting role to have. But we're actively building it, and we're looking forward to all sorts of feedback on how to make that role better and more impactful for everyone.

It sounds very, very interesting, I would say. Cool. Yeah, let's speak later about that, what you speak with customers, because I think it's a tough time, because a lot of change, it's a fast piece of innovation. But one question, you mentioned region for the audience. You are from the Asia region, or what is your scope of where you're working with customers? This.

Is very interesting, because I think the first time I met Tobi's Opia is actually at an event where we were actually working with another customer. Shockingly, this customer is not an Asian customer, as in they don't have their headquarters in Asia. They're basically based out of Europe, somewhere else. So ironically, this podcast is happening because of one of the reasons why Asia is very unique, right? If you look at it, Asia has sort of become the developer center of the world, right? There's a lot of gravity that's shifted towards here because I think over time, one of the things that's been realized is that Asia has a ton of engineering talent, software engineering talent that passes out of its institutions and colleges. And this talent has now been able to then be employed in significantly impactful work across a long time. This started with all of the big GSIs, sort of a thing in India, and then a large part of a lot of the other Asian companies started off. But what's happened since 2020 2012, 2014 is that I think Asia sort of become the center, epicenter of software development in the world. So I jokingly say, right, so you guys can all go check it out. Actually, GitHub has this Octoverse metrics where we actually put the total number of developers in every continent and everything. Actually, if you go and look at Asia as a unit, it basically... smashes through the developer records on every other content whatsoever, right? Because it's just so many developers here and people that really want to try to get their hands dirty and more joining, right? Like I think one figure that I saw out of India is that I think every year we had about 300,000 software developers just from our tier one engineering colleges. I'm not even thinking about community colleges and all of those which add more developers. I'm just talking about people that graduate in computing sciences just every year is 300,000 people. That's a lot of people. that graduate in the core science. So I think that's the reason why, right? There's a ton of really smart talent coming out there, which means that a lot of the large European American organizations are setting up global capability centers in Bengaluru, in Pune, in Hyderabad, in Hanoi, in Vietnam. They're setting up centers in Singapore to drive conversations. They're doing it in Hong Kong, right? Australia, there's so many places where people are building these capability centers. I think what's happened is Microsoft is a fantastic example ourselves, where we are all gainfully employed. One of the things we understand is that Microsoft is a huge development team based out of Hyderabad. A large part of our development happens there in what we lovingly call the India Development Center, the IDC, right? So all of these are very interesting sort of views of where this epicenter has happened. So I'm uniquely positioned to kind of answer this culture because I work with customers on both spectrums, right? I understand how these products are built and what customer personas they're trying to look at. And then I understand sort of the builder persona out here who's sort of trying to fulfill those needs. And then I'm able to then bridge these two together in a way which is very unique to me. So I think this India perspective and the Asia perspective is very relevant to this conversation because a lot of the bleeding edge work is now being done out of here. If you see innovation centers, right, largely for a large number of European organizations, innovation centers are based out of India, based out of Singapore, based out of parts of Asia. It's spectacular to see the kind of thing. And all of this has then had bleeding effects for the economy ourselves, right? For digital payments to technology. That means that it's not just European companies that have had the explosion. It means that there are more organizations here who are building for the local market and therefore it starts a virtuous cycle of people building for other people. And it's a beautiful sort of ecosystem that's developed in itself. So I think that's the beauty of Asia. So if you look at Europe, I see a lot of, when I go to Europe and when I go to the Americas, there's a lot of innovation going on there. There's a lot of R&D. And then all of that R&D comes to India and other parts of Asia to get built, right? And sort of the bleeding edge and then, oh, what do we do differently? How do we move it forward? So I think that's a very unique perspective of this globalized economy, right? Where everybody's work feeds into someone else's work. And it's very exciting to see as we sort of go forward.

This also mirrors like our experience. So I have had like one project with colleagues of ours in Hyderabad, and Toby and I myself, we had a workshop together in Bangalore, and I was also recently in Bangalore. And the kind of hunger from everybody, you know, to innovate, to create something new that was really amazing. And when you walk through the streets, there are like posters on the streets saying like, Here you can learn about DevOps or you can learn about programming and stuff like that. And I haven't seen anything like that in Europe in one way or another. So that was really astonishing to see how much is actually coming out of these cities.

Yeah. And the energy, like Toby said, it's a different game, I would say. And you said we give it to you to produce, I think. There's already a change. Something like the idea comes from China or from India. And yeah, I would not say that we don't, can achieve something here in Europe or in Germany, but I think we need to adapt. We got a little bit lazy, I would say. I think now people are ****** on the podcast, but that's really my opinion. You should go there and you should feel that energy and it's like, wow, people, People behave different. And they have that energy and the smiley face and they want. The motivation is a different on a lot of people, I would say, not on everybody. Yeah, we were speaking about AI or we want to speak about AI. So Karthik, every day we wake up, every day something is new. How to manage the pace of fast innovation in AI area? What is your personal approach with that?

God, where do I get started? I think this is such a huge, this is such a huge part of our sort of where we have what our opportunity areas are as we sort of go forward. So there's a couple of ways to break it down, right? Like think about this. I think if we look at the pace of growth in this industry, I think There's two ways to look at this. Think about this. There's this one analogy that I read online that was very interesting where people said, you all think the AI way was really done a lot? Imagine this. If you were someone that lived in pre-World War I Europe, for example, right? Versus someone that came after World War I Europe, for example. And then let's assume that 10 years between here and there, you would have seen the growth of automobiles. You would have seen the ability for people to not have to do horse-drawn carts. You would have seen public lighting in most major cities in Europe. You would have seen commercial flight finally happening, right? The Wright brothers to commercial flight happening was like five, eight, nine years. So what people, some advocates of change and some people that really study this philosophy of how human scale change happens over time basically say that it's actually not very significant, that we have had other times in our past where we've gone very differently. Let's assume we put someone on a deserted island before World War I, when they came after World War II, they would see the whole world completely changed. So that's a huge mind shift because things became easier, things became more healthy. There was an entire pandemic. So many things happened in that time frame, right? So That's the beauty, right? So if we zoom out and look at it, then there, but today, because we are a large part of us are involved in that hype cycle, let's, I'm going to break it. This is a hype cycle, right? There's always a trough and a crest. There's so much going on that we all have to deal with, which is so shocking. Every day when you open Reddit is basically thread after thread after thread. Hacker news today is a whole different thing. Product hunt. was one of my favorites and every post and product is about an AI company that's now trying to anything that's to do with AI, for example, right? So whatever we used to do, there's an AI way of doing that right now differently and something new every day. Even in our own platform like GitHub, which I can speak about, right? Like we have change logs that I now subscribe to our own RSS feed. to be honest, because the pace at which the RSS feeds come out is easy. I get a notification from my mobile phone, and then I can see the change logs of what we're changing, you know, new models coming out, new things are different. So it's so fast and exciting for us that it's not. I think in one way, the AI wave has fed the AI wave in a lot of ways. So think about it. We are using AI to build these tools.

Yes, yes.

And then therefore we are building faster and then we are pushing more things and more impactful releases for our customers. I think it's wild to sort of see sort of how fast it goes. So I think for me, I think the intentionality is true. That's the first piece, right? You have to have an intention to gather more information, one. And two, I think today the more important job is filtering. critical information that you should consume versus what is information you can deflect and say, just know about a little bit and then go forward. So I think levels of knowledge is sort of what I've done. So I basically say, if I need to know this, I need to know all about this and I'll do deep research. If I need to know only what it does, I just need to know what it does. And then if it's really interesting, I'll get my hands dirty over the weekend. But if I were to get my hands dirty on every AI innovation that came in the past week, You would need an entire week to do that. just doesn't work.

Yeah, I agree. And I think what we want also from Toby and Toby give a message is, I think it's a hype, yes, but I think it's a change. When we think about software development lifecycle, and I think there's so much improvement, I think the... The biggest challenge is nobody knows how to make that in a professional way. What is the new software development life cycle with AI? Where does it work good? What are good patterns? Tools, like we said, is you wake up, there's a new tool. So the tool sets are not like in past, yes, you start a new project and you know what kind of tools you need for that and which roles. But I think, and the message what we want to give for the audience is there will be a change. I'm 100% sure. We don't know how and we don't know how much was hype and how it will be in the future. But when you are a listener and you are not working with AI and you are somewhere on the software development lifecycle, I would highly recommend you start with it. And like Karthik say, don't take everything serious what comes out. But yeah, start. That I think is a big message.

Definitely. I also like the historic comparison you were making. Like I was just, while you were speaking, thinking about when you say pre-World War One, for example, that was my grandfather's father's generation. So, you know, that's not that far away from me because, you know, my grandfather is dead now. But I actually, well, I think he died when I was 12 or something like that. However, I mean, I never met his father, of course, but I have photographs from him and stuff like that. So that's not like historically not that far away from me nowadays, and. even when the pace now accelerates with everything, it is also pretty clear that when you have like a long-term project of a year or one in a year and a half or so, the stack that you started with will be probably different than the stack you're ending up on at the end of the project because so much is happening in between, and you have to be really agile about it and don't like stick to the pattern that have always worked with you, but rather adapt to like new patterns and new software stacks and everything.

Yeah, I think it's a mind shift. Mindset shift, it's really something what is hard, because mindset shift is also always hard to manage, to achieve and to go through it. Yeah, let's discuss about software development lifecycle. I think we have the different phases and what I observe when you use AI, it's changing because maybe a role can make different things because of large language models. What is your thought about that?

I think we spoke about this a bit before, and I think me and Toby have discussed this when he was in Bangalore as well. The beauty about this was that one of the things that we noticed is that while there's a lot of this change going on about the software developer life cycle and how people are changing things around this, and there's so much innovation, and every new product is coming out there, every other week there's someone new launching, something new about some part of the software developer life cycle. I think the idea is that there's the underlying shift of skill sets that we need to talk about of what's sort of essential for success as we grow. I think I see a large amount of change that's happened already in the code generation space that's already sort of now plateaued, right? So I think we know for a fact that AI can write code. We know we've understood AI can write good code now. And I think with the more models that are coming up, we know that AI can write and validate the code it's written in significant ways. I think the innovation that's happening right now is on sort of agents for every single area of that part after the code's been written. Do we want to check the quality of the code that we wrote? Do we want to check the validity of the methods that were written to sort of carry out the functions of what the code's supposed to do? Are we writing code that conforms to organizational standards or international standards for security, for quality, for styling, for all of those things, right? So I think in a large part of this, I think the most important piece, and I think everybody's got to think about is that your skill sets of what you do are basically what AI models are trained on. So if you're a really good tester, an AI agent that specializes in testing basically then has the skills of what you do potentially well testing and sort of it then learns from the code that you wrote is where the public conscience for the AI LLMs are. And the more parameters we train these LLMs on, the more volume of capability that they have to sort of then learn it. So when we talk about trillion parameters and all those in models, that's what we mean is that they then are able to then sort of collect more information across different sources and then be able to take that information and give it as more insightful responses for you. So I see that in the rest of the phases of the software level lifecycle, there's a large shift happening on how we can implement it right now. I think code generation already there. I think very easy use cases. I think most organizations across the world have some form of AI in that part of the lifecycle. And as they go into sort of then after generating code, what do you do with the code that's been generated? Validation, testing. in a QA, maintenance, all those things are sort of where they're sort of going into next, right? And I think all of this then feeds into what they're trying to build as a product over sort of a longer period of time as an organization. So it can then contribute to your bottom line. So if you're saving more time, building more features, making customers happier, and at the end of the day, that drives positive revenue growth for all of them, right? So I think it's a very interesting sort of a place we're in right now, where we then What after coding is the million-dollar question that people are asking now?

Yes, and what I observe is in past it was that time of the truth is in the code, so you have people they was testing and they come to you as a developer and said... Why is it not working? So where did you check it? In the code. Because there was the truth. There was written what it really does. The documentation and requirements maybe was outdated, let's say like that, because of laziness or because of this amount of work what you need to put in to update that stuff all. But what I observed with AI, it's a button click. So even the tester can give that question to an AI agent and say, hey, why this code is behaving like that? I expect it differently and gets tester role explanation, even when he don't understand the code, it's transformed for him as an example, what is really, really cool. Yeah, that's, and I see a lot of improvement, like you said. There comes some issues because of telemetry. You see there is something in production agents starting and grabbing this context and research your code base and give you maybe some kind of a root cause analyzer. And maybe next step is to say, hey, here's some kind of a branch and then check it out. I think it's a fix. And how cool is that? You're sleeping, your software is not working, you're waking up and It's something there. Someone was already working on that. And yeah, it's a big shift, I think.

Sort of one thing that we discussed was very interesting is that human beings need eight or 10 hours of rest.

Yeah.

We work eight or 10 hours and then we have the rest of the day to ourselves. But the beauty of the AI agents that we have to realize is that they're able to just constantly work 16, 18, 20, 24 hours a day, if you want. There's no limit to them. And the beauty of it is, Every software engineer now with agents basically can then say, I'm going to spin up 20 agents that do 15 different things for me. And then I can then concentrate on this one task that I think is subliminally important for this use case and just focus on that and say, OK, I'm going to focus on this system architecture piece while you build the UI, you build the test, and you build the framework around it. And then you then validate that I've written everything correctly. And then four agents go about doing what they're doing in parallel. So I think we started seeing this towards the end of last year where agents started exploding. I think what we're seeing now is how can we make these agents context-rich? And again, this is just in three months. By last year, I just mean December. November, December, when the whole agent thing is only one quarter in, the conversations around how do we wrangle these agents? Is it skills? Is it MCP? Now there's a bigger conversation around those things. What are the right places to give custom instructions for our agents, whichever agent you're using? So all those pieces are very important. So I think if I look at sort of where we will shift, I'm very excited for the future because that means that as human beings and human developers, we focus more on really meaningful pieces of work that drive change, like impactful change, rather than actually sitting and building a UI, which is like a solved problem, like how many login screens do you want to design?

Yeah, exactly. Exactly. I thought about that. I have a, um, I follow one guy about the stock market and he write really manual articles. And then I, I ask myself why I give that guy 20 euro per month. It can do a large language model. And then I read that article and I thought it can be done by a large language model, but I need to know what I put into the prompt to to get that result. So even his experience and his thinking about that world, I don't have it. So I would not be able to make that prompt and to get that prompt. point of view in some kind of a large language model generated article. And I think that is something what made me think about software. So like you said, log-in forms, I think average can now everybody. But to make it more advanced, you still need to think. You still need to guide that thing and say, hey, I don't want to have that standard log-in form. I want to have that. nice one why ever what is what means nice yeah but I think that is now something like you said people can more concentrate about user experience about how I make my software better and not about how to achieve the the average I would say.

Kind of leveling the playing field for everybody I think yeah and if you want to be be or create a high quality product this is like the the human work you can put on top yeah absolutely yeah.

Which then sort of naturally segues us to this million-dollar question, right? Which I think is the most often asked question that I get is, okay, how about my job?

Yes. Yeah.

If only I had a penny for every time someone asked me that job is the question. No, no. I think it's very interesting. I think this is something that came up in many different places, including the European Parliament, if I'm not wrong. I think this was brought up as a very active conversation where they spoke about it. I was very intrigued by the whole debate where they spoke about this. The way I look at it, I think the best way to put it, though it might sound like marketing speed, but it's not. But I think AI is a force multiplier. Whether you're in software development or not, doesn't matter. Because if you look at my example, is that I don't write as much code as I used to, to be honest. But I use a lot more documentation. I use a lot more of the office tools, for example, from ours table that I use on a day-to-day basis myself. Because I'm a little bit more distant from writing code on a data basis. I do review PRs and all that, but I don't write code, per se. And what I've realized largely is that there's a lot of AI to be gained in, regardless of what work you do, where you are. And there's a lot of fear-mongering going on. And I want to be very open about the fact that today's fear-mongering, people are creating a bias towards this fear, right? And I think it sounds good on a newspaper headline. If you want to be sensational, right? Like, oh yeah, it's coming for your jobs. It's fantastic for a tabloid newspaper. It's good for page three, right? But the reality of it is I think, yeah, AI still cannot do a lot of things that only humans can. Like the example of what you gave, right? the gentleman that wrote the finance newsletter, for example, that you pay for access to. Journalism is a fantastic example of where no chance in hell, you will not be able to have an investigative journalist AI. AI can maybe help the investigative journalist, but then cannot do that. But even people like me, for example, in management, when we're writing user spec documents or feedback documents, or when we're writing what sort of products to prioritize, or when I'm writing a BRD or a PRD for sort of a demo or something that I want to showcase how cool something is, but when I'm building a talk track, this is really cool for me, right? When I say, hey, is my talk track making sense? Will someone understand? Give me feedback about it. So it's a lot of using of AI. Helps me sort of then sound out my thoughts, because in a lot of ways, we used to work, even though you had teams, you had to work in isolation, right? It was until someone gave you feedback on what worked or what didn't work, you were largely isolated to your thoughts on your things. You didn't have a chance to sort of say what it was. But now you can have a mirror, someone standing right in front of you that says, hey, I have this thought, I have this idea, I want to make it work and how do I do it? And then you can then say, hey, it doesn't sound good and use it as a sort of a sounding board or a validation board. So I think a lot of people have to gain from AI. There's not many people that are at risk from AI. I'm not going to say that there's no one at risk from AI because I know that's not true. There is some risk from AI in some areas of jobs that can be actually replaced with AI, there will be, like data entry and OCR, Very easy use cases, image recognition. These are things that are part of development. I think we should be cognizant about those things. But a large part of what it takes to be human is still going to be human. And I think what we spoke about in the last, when we were speaking about right now, the skill set shift is critical for us to sort of have the right skills to be more impactful in whatever work we do, whether we're accountant or whether we're a software developer or whether we're office worker, it doesn't matter. It's just about Being open to change, understanding how this tool works, taking it one at a time. Don't speed through it. It's okay. You don't need to know all the innovation that week or that month or that year, but choose one innovation that drives significant impact for you and your work stream and just keep using it, keep better at that tool. I think that's the best way to do it.

Yeah, definitely. And the one thing I thought about the other day is I was writing an iOS application and I thought, well, it's not just me hacking in the code, which used to be my job like 15 years ago. I feel more like a kind of product manager or something like that, where I have the AI as a sparing partner and I'm putting in ideas and I'm getting feedback and I'm creating code and then I'm testing this stuff and it feels more or less, yeah, as I said, like a kind of product manager, when you're working with AI alongside.

Yes. And you can try out stuff faster. I think that's a good point. So making a short proof of concept. In past, you had this discussion with business and they had a wish and you as a developer said, ah, it will not work. And because there was a lot of effort, you never could show them that what they want to achieve. But now maybe it's one or two prompts. Yeah. And then show them. Something like what we called in past some mock-up application, and I think that is a nice thing, but the other story is, you say it feels like a product manager and a product owner, whatever, but... You need to digest what comes out. I think that is something that is the tough part of it. I see a lot of people, they use that things and they say always, Yes, yes, yes, it's the new next button behavior of people. You, and that is my biggest challenge. How can I follow this AI, whatever it does, whatever ideas comes up? How can I digest that? How can I be still a pilot? Because in the end, I need to understand what comes out there. And I think that's the, for me at least, the toughest part of it, to graft that knowledge.

I think we have a sort of a... We speak about this in our prep, where we're speaking. And that's the interesting thing. So there was this large analogy that I brought up. And I think I'd love to bring that up again for the audience as well, I think. It's just the way we used to-- we don't Bing something or Google something. It's a very foundational part of the human experience. We are humans. We are dumb. We don't know things. We Google things. So we always search on Bing for things. It's very simple. Very simple, rudimentary human behavior. Everybody knows how to do it. I don't think there's any, there's very few people that are on the spectrum who don't understand what it is, right? But the fact of the matter is today, because of the ability of us to have so many tools available that can get us that information, we just stop doing certain things in a certain way. And I was giving this example, right? So for example, let's say I just wanted to search for a certain interesting fact about, how does a certain engine type work, for example, and say, how does this, how does Mazda's rotary engine work, right? Typically, when you ask Google about the Mazda rotary engine, it will just give you a bunch of 10 links where you go to Mazda's website where they explain the rotary engine, or you would go to, you know, a physics website that told you about how the physics of it worked, or you would go to a car review website where someone wrote a car review about how the car that has the rotary engine works, for example. But today we either go open up a app, right? It could be ChatGPT or Perplexity or whatever, or Bing and search and say, okay, how does it work? And then it just gives you an entire output, like sort of the idea of what you, exactly what you ask and it gives you just that much as an answer. So it kills curiosity is my, is sort of the way I would look at it, right? It kills the ability for you to then say, How do I go about and do it? So I was talking to the team and I said, one of the ways I do it right now and I feel that it works really well is that I actually added a custom prompt inside all of the AI engines I use, which basically says, whenever I ask you for something that looks like a search, always give me back three links or three pieces of information that would be nice for me to study. And give a link for instantiating where you got that information from. So in the master case, for example, you'd come back and say, this is not the first ever time the rotary engine was tried out. Do you know that in marine applications, there are boats and ships that actually have this rotary engine as part of their thing. Do you want to check more about that, for example, right? So it then builds that curiosity for you to then go and drive and find new things. This goes back to what we spoke about AI, right? AI is a tool at the end of the day. It's like a hammer or a power drill. You have to know how to use the tool to be able to extract the output from the tool. And you have to have the right set of instructions for you to be able to use that tool. It's like a workman's boot. If you are in a factory and you don't have the right sized boot, you're always going to feel like it's either too tight and it's hurting your leg, or it's too loose and it keeps slipping off. Because you have not customized it for what works for you. Right now, I think a lot of people are using AI and finding that it is not as helpful or it's for instant gratification or instant help. But the more you customize it, the more you talk to it on what you need, I feel like the better it gets. So I think that's the foundation, not only is it for other AI, but the same thing in software development as well, right? As you customize it for what works for you, it gives insanely wonderful results.

Yeah, exactly. And I think the example you gave with the prompt about peripheral learning and getting different facts also besides these really specific query that you're giving it is actually brilliant, you know, and because all that context information, you're not getting that right now. And I would probably adapt to that and try it out myself, you know, as a next step. This is a great, great learning.

I think it's, you brought it up in a conversation before that recording, but I thought about that and I find that analogy, like someone makes a trip, so let's go to, I don't know, Africa. And he planned everything and you say, Toby, you want to join? And you say, yes, I join. So, and you make that trip, but it feels different. What I want to say is someone planned it, you jump in and you, Enjoy it or you plan it. And you think about that, what you want to see, you make some kind of research. I think that's my energy. So when you ask someone like a ChatGPT and you say, hey, what is a good way to visit Africa? You get an answer, sure. But I think you should think about what is important for you, how to, what you need to put into the prompt and how you need to challenge the answer. and think about that. That's my way of thinking. Yeah, definitely. So jumping in and be a guest is maybe not the right way in using of AI. Yeah. You've got to be in the loop.

You've got to be at the driver's seat. You've got to be the person that's orchestrating all of this happening, right? I think we've fixed it in software engineering, right? Because we've always had this process in the past, you know, where any code change that is written by anyone goes through this process called a pull request, for example, right? So for the audience that doesn't write code, a pull request is basically a unit of work for software engineers who, let's say you changed a few files of code and then you submit a change request. A pull request is basically a change request where then someone that's a more senior person can then review the code and recommend the changes that they feel would be more impactful in the code base for them to then be able to implement in their future. So In software development, there's already a gate. There's always already a gateway for you to be able to then go ahead and do that. So what you see is that a lot of large part of the agents are now part of that gateway and the humans are in the loop because they are the ones that are approving the pull requests. While someone can review a pull request, the approval of the code that's been written is always human. Someone's always in the loop and someone's always checking the code to ensure that it's valid and sort of then passes all the requirements that they have. So I think We've had horror stories in the past. A good example is how, I don't want to name this person, I don't want to name the tool, but then a tool came out in the newspapers a couple of weeks ago where this tool basically went out, deleted an entire production instance. And this software giant basically then had multiple hours of an outage because an entire set of agents were running on a set of permissions that they had granted for them that they should ideally not have had. in the first place. They were running on YOLO mode, for example, right? So it's like proper YOLO. And then that's shocking for me, right? Like I've seen, you know, Molt, so we all saw the announcements of, you know, Moltbook and Cloud Bot being acquired, for example. We saw what happened in the Cloud Bot apocalypse and all those things. So there's a lot of interesting lessons for a lot of people to learn, right? I think There's a lot of change. We have so much change every day. And we have to be in that mindset that, hey, we have to change the way we operate from in sort of different ways. But it's an exciting time to be there, but it's also a time where you have to be very cautious about what steps you take, what permissions you craft, what levels at which there, and ensure that at every point in time, there's a human that's saying, in your analogy to be like, if you want someone to go to Africa, Of course, if the AI LLM basically said, you know, put your head inside a lion, lion's face. That's probably not a good idea. Yeah, maybe. It makes an amazing Instagram photograph, but...

Unique experiences.

It's a unique experience, put your head inside a lion's face. I'm like, nope, I'm fine.

Exactly. But I think like that is like the deterministic versus non-deterministic. It's like, it's like this is a brilliant example, you know, for that. And when you're mentioning the production outage and stuff like that. So you also need to be prepared like for non-deterministic output, although it's a computer, you know, we are always used to like input validation and a certain output, you know, that is quite deterministic. And this is something we have to Yeah, have in our minds that AI might not always be the right tool for certain things and choose your tools wisely about what kind of output you want to have.

So this is a pet peeve of mine. Why do you need to use AI for everything? You don't need to necessarily do that. One of the most recent examples that I came across is very interesting. In a past life, I used to be a security researcher myself. I used to do a little bit of cybersecurity. And that space is full of tools that do that, right? There's so many leaders in that space that operate and have built amazing tool sets that have done a fabulous job of limiting risk. I'm not going to say that all the software that's written in the world is now safe before AI. It's not, but there's tons of progress on how to do stuff more safe, more securely for all the organizations out there. But ironically, what I realize is that all that comes out And then we have such a large, strong set of tools, including GitHub's own tool, right? For example, in code security. And then you basically see that there's a non-deterministic tool that's probabilistic that comes out. And then all of the industry then has an entire cycle says, Oh my God, that's the next best thing. So basically the translation of that is you have okay with an LLM that has 80% chance of getting something right. And you're okay with that running security scans and checking your stuff. But you're not okay with using a tool that has 100% chance of probably getting it right. Not maybe 100%, I mean 90%, 95% of getting it right, right? You're okay with taking a probabilistic tool, but you don't want to take a look at a deterministic tool that knows what it's doing and just has the right output, right? It's very interesting. It's like having this engine in your car that might work 90% of the time. Would you rather have that? Or you say, no, it's okay. It's got 98% reliability, but this works 90% of the time, but goes at 700 kilometers per hour. Or you can have a car that goes 300 kilometers per hour on the autobahn and then still, you know, it still works 98% of the time with reliability. That's the million dollar question, right? Yeah, I just don't get it.

Yeah, and I think that's the point. And even the example of what you brought up with this deletion of production database, I think, That is also a shift from a developer. In past, developers said, hey, I'm on my local machine. I'm in a special local environment. I can do whatever I want. But when you use AI, AI is very, very searchable. Let's say grab stuff somewhere. And when you don't think about where it got that information, what kind of information is, let's say you have somewhere an environment variable, some Why ever a productive database connection string? And you give him some questions and he asks you, can I use that tool? Can I execute that command? And you don't make that, I'm a pilot, I'm in charge, and now I want to understand what this thing is trying to achieve. And you say, yes, YOLO mode, then you are in ****. When it comes to customer or developers, they sometimes complain. They say it's not reliable. I don't like that. And I think that's that's not supposed to be the last. Yeah. Yes. First of all, when it would be reliable and deterministic, then we would not sit in here and speaking because the idea is it's generative and it's some kind of creative. And that gives you also a superpower, this challenging, getting new ideas, new thoughts when it would not be non-deterministic, it would also only give you that what you know. So it's not what you want. But I think that's the biggest challenge now for developer or for anybody who use that tool. It's when you need that kind of creativity with this uncertainty and you need to judge it. And when you do not need that. So for example, I had one case where a customer wants to create an agent to delete users and add users. And then I said, why you want to have that with AI? It's an important thing and it should not be- You mean you need an API call? Exactly. So why to do that? And that is, I think, We now are laughing about that, but I think a lot of people have that challenge because they have that mindset about it's something like smart. It's something like it works like smart and it's like a human and it's better than me, whatever is the thoughts and the fear, but it's not like that. And I always think about when is it really good to use this undeterministic stuff and when it's not good. Or you make a refactoring in your big, huge code base. Why not to use a traditional refactoring tool where you know, okay, I made a good renaming and everywhere it's now this renaming happened. With AI, I would not say it's not possible, but it's another game, I would say.

Like I went to this ATM last week. It was the most interesting, insane example. It just drove me nuts. So I went to this ATM. I don't want to say which bank, but then we went to this ATM and this Indian bank, and then there's a bank in India, and then They basically had an ATM that said, talk to a new conversational AI engine to withdraw your money. Oh, yeah. Wait, what? It's a ATM. It's 4 buttons. You press your title account, checking account, savings account. You press the amount of money you want. You put in your security code, and then the money comes out. How complicated is that? You don't need an AI agent for that. And just for kicks, I tried the AI agent. And because this ATM has to have a mobile connection, and then there's an LLM that's then sending a response back out to the ATM agent, it got stuck. Like, in the middle of the transaction, it got stuck, for example, right? Like, why? You don't need AI in an ATM. It's a four-step process. It's not complicated.

Yes.

Yeah, and now say your security pin loud and clear so the agent can listen to it. That's great. This is like having like a button on a double or nothing, you know, and make it a little dilificate.

Yeah, I like that. To be serious, we're living in a comfort world, luxury world, because at Microsoft, and I think at GitHub is the same. we can use LLMs as much as we want, let's say like that. I think there is some limit, but I know nobody will reach that. And it's a game changer when you don't have to think about how much token do I spend, how much tokens do I have today, how much premium requests, whatever. And it's nice that you can try out and to learn when is this the right tool, when is it not. But Still, even when I could use it for everything, I don't do it. And I think that's the biggest learning from my side. So, and that's the tough part. Where is it good and where is it not? And when you have something like an increase of productivity and you feel good and not like more chaos, yes. So let's say you make one prompt, you get an output and then you sit more time there to make that finishing and read through it and adapt it, then you would do it by your own. And then that is not something what should be what you want. Yeah.

So there's this learning philosophy called first principle thinking, right? Like where you basically say, you've got to ask an infinite number of whys to get something right, to get to the root of a problem, for example, right? You keep asking why. Like, why do you need-- so for example, if someone says problem, like the ATM use case, for example, if someone went and pitched it to a board and said, hey, I want to have a generative AI-powered voice assistant in an ATM, the first question the board should have asked was, why? And if they had asked the question, say, why, and then the why, and then maybe there is an accessibility use case that I'm not seeing, for example. Maybe there's someone that cannot see, for example, that for them, it's a better use case, right? But that's the why. You get to the core of it saying, okay, it's an accessibility feature that can be accessed by someone. But then if it's a button on a screen, how does the person that wants the accessibility feature access it, for example? So it's a critical flaw in the why chain. You have not asked the right amount of whys to get to the why of which matters here. In this case, it's a why saying, okay, this person came in. And then this person is, let's assume that they have visual impairment. they still cannot see the button to click and open the agent in this case, right? So I think one tip I would say is one of the things that I use, especially with my team, because like Toby said, we're in the middle of this AI wave ourselves. We have all the AI tools at our disposal. I keep asking my team, why? Like, why do we need to do it a certain way? Why can we not do this without AI? Is there a certain API that exists today that can do this for us? Why can we not use an existing tool or a code base that already exists for using that? Why do we need to generate something? Why do we need yet another login page, for example, right? That's the why framework for me.

Or raising these questions, why could we not achieve something in past? Because it's logic has some limits from computer. There was always some edge cases in the past where you said, Nice feature, but we cannot do it because it's too much effort or hard to program. So maybe then that the things and corners where you should think about using AI or I thought about that. We have a formalized world here in the Internet. So you want to have a pizza. It's not like you would do it in the store. It's like, okay, what is your first name? What is your last name? So nobody asks you that in the store. So I think a cool thing could be that you make it more the input of the data more like humans. So I want to have a pizza, I'm Toby, I'm living in a city and so on. And behind the scenes, this information, unstructured information will be grabbed and put into some kind of a structure and mitigated. And I'm not bothered with some kind of a formula and I need to click and then you are on the end and you click the button and it says, oh, you missed something and the data are lost and stuff like that. I think that's, for example, we can improve and I like that. But Don't put it everywhere, I would say. Yep.

Especially not in ATMs.

What I'm excited for is that thing, right? Where you have an agent that can talk to an agent and then sort of work with you. Imagine this, in the same use case that you're ordering the pizza, I think where we're talking about A2A frameworks and all that right now is that we're basically looking at a way to say, okay, let's assume that Toby has his own Toby personal ordering agent or something like a Jarvis from Iron Man's assistant Jarvis, right? Like for example, like you have an AI assistant that knows everything. So you open up your phone and say, Jarvis, please order a pizza for me. So in this case, Jarvis then knows everything about you. Jarvis knows you only like pepperoni pizza. Jarvis knows Toby likes pineapple on his pepperoni pizza. I'm just joking. I'm not sure he doesn't like pineapple on his pepperoni pizza. But like it knows your personal preferences. It knows the pizza store that you like to order from. It knows the phone number and then it places the request. And on the other end is another agent from the store, because again, why would you need a human to take an order? And then these agents talk to each other, and this does not have to be in human interaction. I think that's the next bleeding edge, where if agents can talk to each other in a more efficient communication format that is not human language, then it's much more efficient for them to then pass information to one another that's contextual in this use case. And then once they've passed information of what sort of pizza for where, where the addresses and all that, that agent disconnect in a couple of seconds because that's all it would take for two agents to talk to each other in say something like a JSON format with A2A, for example. And then the information is passed and it's done. So it's wild if you think about the possibilities of something like that. You save so much time for the person that's ordering, you save so much time for the pizzeria that does not have to have someone manning the phone line because That person can then be actually making the pizzas or attending to customers who are at the store, for example, right.

So yeah, it's very interesting. Yeah, and they can focus on the real business value, making good pizza and not sitting on the phone and taking calls and yeah, exactly. Did you see that video? We can put that in the show notes. There was some example. It was two agents and they was calling each other. And in the beginning they speak human, yeah. And then they say, hey, I'm an agent, da, da, da. Then the other one, yes, hi, I'm here. I'm also an agent. And after that, they started immediately to switch the language. And to improve that, what you say, it should be more easy. And maybe human language is not the best way for machines to communicate. So they switched and say, hey, I can speak that and that. We have an improvement and awesome to see. I think a lot of people are scary, but don't be scary, I think. Don't be scary.

Even if you look at GitHub, right? I think that's where the frontier is for us, right? Like if you look at, GitHub already has agents, like we have agents that do different types of things for you, we have security agents, we have coding agents, we have agents that can do a lot of things for you. I think a large part of our innovation right now is one in governance and control. Like how can you control these agents and how can you govern these agents? What agents and what LLMs are you using? What scopes do they have access to? That's a lot of the work that we're doing right now. We call it the agent control plane. And I think we have this offering called the agent HQ, where we basically try to bring different agents together into the one sort of a place. So you could have a cloud agent and you could have an open AI agent and you could have a, you know, you could have a, In the future, you could have, say, a Gemini agent. Whatever third-party agent wants to come onto the place, right? And you'll be able to then use an agent of your choice with a product like AgentSQ, where you can then choose the right agent for each task. Because what we've realized is just like being in a supermarket, right? Every LLM has something that is really good at doing, and then you can pass the LLM with just doing that. And for example, sometimes I've heard that Grok is really good with unit testing. I didn't hear about that until two weeks ago. Some people told me that, try Grok for unit testing. That's really cool. Like if you had a Grok agent that just did testing, for example, right? And so all these things are really interesting. So you can then mix and match things that you want. And then I think the next frontier for us is just that piece where how can we make agents talk to each other more effectively? How can they pass information between another end? Retain the context, because at the end of the day, ours is a huge platform, right? Like if you have hundreds of thousands of projects on GitHub as a large enterprise today, for example, how do you transfer and retain context between agents that are working for two different developers, but two different developers working on the same piece of code? That's a million-dollar problem to solve. That's a multi-billion-dollar problem to solve, right? How do you know who's doing what at what time, and how do you think the right intervention for who needs to sort of come in? That's the next horizon for GitHub, right? When we look at simultaneous use of agents and how developers are writing as they're writing code and they're using agents to implement changes in code, how can we communicate with each other sort of effectively? I think that's sort of the next in the horizon for us as we look at it as well. It's exciting times to come, to be honest, I think, as we... I'm very excited. I'm excited for the fact that When I wrote my first line of code, when I was in my first year of college, right, my first ever line of code was in C programming, right? And I was scared to start. I was scared because I did not know what it was. I did not understand what it was. I could not understand the characters in front of my eyes. Of course, as I became a more proficient developer, things start making sense. If you do something oftentimes enough in your life, you become good at it, right? Practice makes a human being perfect. But I think the beauty of what we have in our disposal right now is you can get rid of that fear. You can become a better singer with AI today. You can become a better author with AI today, write better things, right? You can become a better manager with AI today, ask it questions on how do I react to some, you know, one of my direct reports, how do I react to a better manager? You can be a better human being today, you can be more curious today, you can be what you want. with something that was just not possible five years ago, right? And have that ability to then drive those skill sets, which I think is fundamentally paradigm changing. And that excites me, like the expansion of human possibility, right? Of all of us being able to then do these new things is very exciting for me to then see what we will do collectively. And as long as we got our head on the right places and we do the right things for each other, we care about each other, and we keep building the way the human race has always looked out for each other, then I think the future is bright for all of us. I think we'll collectively build a sort of a world that looks forward to sort of then new innovations and new horizons for all of us together, right? Significantly saving human time, putting our brains together for efforts and things that we really need to solve so that our energy is spent in the right places. I'm excited for that.

Yeah, definitely. And thank you also for bringing up the human element of the whole chain. You know, I think this is something we in each and every technical discussion we leave out for often enough, you know, and bringing that basically back into our heads is really, really important. Yeah.

Yeah. And I think human also is a good word. And you mentioned some different models, different tools. Yeah. So and I think all Or what is a challenge for me? Someone comes and say, Hey, the model is better, but it's not like in the past that you have some kind of a deterministic feature matrix, and now you can compare, Okay, it has this, and you make a... black and white decision. Let's say it like that. Because when someone says that to you, you should think about, okay, what was the context? What was his prompt? What kind of data was accessible? Custom instruction and whatever. And that makes it so complicated. So, and when people come to me and say, ah, it's better. So then I challenge them and say, why? What did you do? And even with models, it's, I think, a tough part even for us who can use whatever we want to decide which model is now the right one. And it's my, my energy is like it's some kind of some employees. Yeah. So every model is an employee and you need to get a feeling how, how good is this employee? What, where are the strengths and where are not the strengths? And sometimes you even have to ask yourself, how can I give a better input? How can I better get better instructions for this guy that he can achieve in his style of working, whatever it is. And I think that's the biggest challenge for people who was not in that role of instructing someone. So more taking instructions and then digest this and do something. And it's about humans.

We saw this in GitHub, to be honest. At one point, I think our catalog had 21 models. And I think you can still find this on our documentation website, but I think we have an article that basically says, if you are using which model to use for what use case, I think it's still there on our doc site. You can still go check it out. But we realized through our AB testing for a lot of developers, that developers were just confused, like what to use. One of the interesting things that I think we've done, and I think a lot of the industry is now following suit with that, and I've seen that change, is what we call intent detection. We basically then try to understand, if you're trying to do a certain type of work, and we try to then understand what is the intent behind your work. Are you trying to write unit tests? Are you trying to write code? And then we then give people a random new model called auto, where you can basically just have choice of whatever LLM you want. we chose the LLM that we felt is best for you. We went out for AB tests and then trust me, people did not choose it. So today we actually incentivize that on our platform. But if you choose auto, we give you a 10% discount on premium request usage on GitHub Copilot. If you actually have auto as a default model and we're incentivizing users to use that more because we feel your experience with something like that is better because we're then letting An LLM, an agent, actually sits in between, which we call an intent detection agent, and then identifies what's the intent of the code base and then the prompt that you're giving, and then diverts you to the right model that will then do the task for you. And then whichever model you use, you'll get a 10% discount on that.

Yeah, that's brilliant.

Yeah, but this is, I think, the biggest challenge now. What model is good, what is not good, what is the model for the right use case for my project, technology stacks. It's a tough question. And yeah, it would be nice to have something, like you said, an AI, what has.

Some-- And then tell you which AI to use for this AI work, yes.

Exactly.

It's like having an AI supervisor that supervisors AI agents that are supervising other AI agents, which is actually real. It's a real use case. It's happening today.

And another point, I think, what to add, and also about mindset, I think a lot of process in past are made like they are because of humans. And one example, I see now people moving wikis in markdown into the repository for the reason it's easier to grab for AI systems. that knowledge. But let's assume two years back, you're sitting in an interview and someone asks you how you share information in the teams, how you have some kind of a knowledge platform. How you do that? And you will sit there and you say, it's a text file in the repository. You will not get a job. And I think that is now also something what is changing because we want to have, we have that large language models, we have easier, an easier life at processing. big, huge text of language. And it's a change. And even made me surprised. I saw that and I thought, it's genius, but hey, come on. Two years back, I would say, no, don't do that. Use something. What you see is what you get, editor or whatever. So to make that more easier to edit and adapt and to read, but it's not necessary anymore. Crazy times. And interesting, I would say. Yeah.

I just want to wrap that up. I know we've had a ton of conversation, but I think one place where I want to change, I think I'm going to be the guest that asks you guys the questions on your podcast. I want to have that recognition of being the guest that started this new trend. But I'm going to ask this question of you. So I think I love asking this question. Whenever I have a panel with CXOs or whenever I'm doing this, ask this question. I love asking this question. One is I love to ask, tell me one thing of everything that we discussed, one thing that keeps you awake at night, something that you're not necessarily afraid of, but something that gives you thoughts, that gives you thoughts of anxiety or thoughts of things that you don't know necessarily how something will happen, right? Not necessarily a fear, but you're uncertain about something. And tell me one thing that you're excited for, something that excites you and like, oh, this is fantastic, what's your exciting thing, to both of you, yes.

Yeah. So I can probably get started. So one thing, well, it's not necessarily keeping me up at night because I still have a good sleep, but I'm thinking a lot of right now is for younger generations. So I have two children, they're aged 11 and nine right now, and I'm also volunteering at a school in Frankfurt, Germany here. And these students are grade seven, you know, and whenever it comes the discussion up with AI and what they're doing, I think AI impacts these younger generations a lot more. First of all, they get used to using AI on a daily basis. So these young people, they're quite confident in using ChatGPT, for example. But on the other hand, it impacts their choice, for example, what they want to become later in their life. So we have had a conversation with somebody who said, well, you know, I want to become a software engineer, and I'm not sure if this is actually the right choice. And I mean, based on our discussion, Right now, I think it's still the right choice to go into software engineering and stuff like that, but these students, they still have, I don't know, six to seven years in school, and you don't know what is happening in the next six to seven years, and same for my children, so... I mean, you cannot beat knowledge, so it still makes a lot of sense to go to university and study a certain subject, in my opinion, But giving them good advice, what the world is going to look like in six to seven years or something like that is quite difficult nowadays. So this is something. which I'm thinking a lot, in terms of newer generations, in terms of programming languages, in terms of how we're doing stuff in the upcoming future. And there's a lot of where I'm spending a lot of thoughts on.

Very nice. And what keeps you excited? Yes.

Yeah, I mean, it's basically the same, you know. We were discussing like things opening up every day and new things coming up. And, you know, whenever I'm getting a chat from Tobi, have you seen the latest model, have you seen this, and have you seen that, you know. Yeah. You know, I'm trying things out myself like every day just to keep up with the pace, you know? And so I think the amount of AI knowledge is also not distributed equally within our society. So we have people who are at the forefront, you know, who are far more advanced than I am. I'm somewhere in the middle, you know, I guess I have a good overview, at least in terms of developer topics on AI and stuff like that, you know, but there are also people in my personal environment who said, well, can we switch off that whole AI thing? I'm not comfortable using that. So AI is still keeping me excited and being really having a good conversation about AI and good thoughts and when to use it, when not to use it. And so this is also something which keeps me excited with new possibilities, as well as deciding deliberately that for a certain problem, I'm not going to use AI today.

Okay, good points, Tohui. Yeah, what keeps me awake? I ordered a MacBook Pro first time in my life. The question is when it gets shipped. No, to take that more serious. Oh, God. I think what keeps me awake, I think it's this energy. So, two years back, I think it felt like everything is settled in our industry. We know where to go and how it works, and a bit bored, I would say. don't take it personal. Yeah. So no, nobody will listen to this, but it was my feeling. And now it's, it's this new, you feel like a child. You feel like, uh, everything is changing. Everything is possible. You see people starting to code who you never expect that they can code. Yeah. Because of the role, because of the, the time, what they have to do something like that. Um, and that makes me, yeah. Yeah, excited, and it's not taking my sleep, but it makes me a lot of thinking. How is this changing our industry, the world? And I liked the times, I think, because... One thing, what is serious, I think in past, I was a software developer. And when I was developing software, you have this imagination about what you want to achieve in your brain. And it's only in your brain. And it's very, very tough to translate that to product owner, to a customer. So you have always this kind of transformation of these two worlds. It's domain driven design. I think Eric Evan was writing about that and how tough it is, yeah, to bring these two worlds together and to understand where are the gaps and what I observe and what I think with this technology, I think this two worlds come closer to each other and maybe it's easier to bring in other people to give them a feeling what is possible, what is not possible to try it out. And I hope that it makes our world better in in the sense of better products. So not an ATM, like you said, but something really like, okay, now, because technology is always just a human sit in front of a machine. So you have that interface, let's say like that. And I hope that this kind of interface, it's designed by the limit of what was possible. And I hope that now we see really cool things like, Don't make me think when I want to order a pizza online or on an app. So let's do it. Yeah, and that is also my excitement, I think. It feels like a child, yeah. So I'm an old guy, but it feels like a child and playing around, sitting in a playground and doing fancy stuff. And I like it. I enjoy it, really. Yeah, that's my thought about that. What is about you, Karthik, to give the question back?

I think I shared a little bit before, right? I think when I was talking about what GitHub was working on, I think I'm really excited for human capability. I think that exactly what you're saying. I think we're amplifying people's skills so much. I'm so excited to see what everyone will do. I'm really genuinely excited. Like every time a developer walks up to me and shows me something cool that they tried out with Copilot, I get so happy. I feel so happy that Hey, this is so cool. Have you tried this out? That's genuinely so exciting for me. And I love talking to developers just for that one reason. What keeps me up at night is, I think I share a lot of what we spoke about. I think what keeps me up at night is, will we be able to, as a community, like me and of course the two told me is a part of this community that's trying to share and disseminate this knowledge with all of them and trying to make people use AI in a more effective way, right? Because as I said, there's a lot of fear-mongering going on out there. There's a lot of misconceptions about how to use AI going on out there. When I sleep, one thing that keeps me up is about my community, about my solutions engineers, about my field teams, about the people that work through this, about our DevRel teams that are working across the clock trying to share this information, saying, how can you use it more effectively? How can you use it to the right use cases? How can you build the right things? There's no right or wrong for anyone. And of course, it's always gray. Everybody has their own use cases of why they do some things. But I think it's there. I just hope that there's enough people that can share their thoughts, share their learnings, and educate as many people as we can. Because I think we've left a lot of people behind in the past with technology. And I'm really sad about that. Like when the mobile wave came forward, we left a lot of people behind. when we went to apps and smartphones, we left a lot of people behind. And what I don't want to do is leave people behind in this way. I think everybody deserves a seat at the table. Everybody deserves a chance to, you know, have that insightful ability to then be accelerated by AI and do something interesting. And I hope that all of us together will be able to do that for the better of humanity. That's my only thing that keeps me awake.

Thank you. Wow, thank you for your time.

Yeah, that's it for this episode of Toby Olds Wei Kartik. Thank you very much for joining us today. This episode was great fun. So for the listener, if you have any feedbacks, let us know via e-mail. Until next time at Toby Olds Wei. Thank you very much.

Bye.

Bye.