Episode 4: Designing Intelligence: GitHub Next, AI, and the Human Side of Code

Welcome to Friends in Code Spaces. This is a show about celebrating the magic that happens when we run into each other IRL. I have had the pleasure of seeing Idan Gazit talk in person. He is a Senior Director of Research at GitHub Next. I saw him talk in person and I saw the entire audience go, and then he thoughtfully brought them down. I am so happy to have you on the show today, Idan. I'm so happy to be here. Thank you so much for having me. Yeah, absolutely. So if you had to describe your code space, what code space are you in? Like my code space currently is like I'm in the AI code space. I mean, oh, wow. Currently? I guess currently, currently, I mean, like last week I spent some time messing around with also, yeah, obviously AI things. I come from sort of like a hybrid front-end, back-end design background. So it's like whatever, I just like making products. Awesome. And lately I've spent some time, I think the thing that I'm really excited about lately is small language models. It's like we're getting to a place where we can actually start running that stuff in browsers. And like, yeah, I need to download three gigabytes of model weights, but then I don't need the cloud. I can just do some inference locally. And that wasn't possible a year ago. So that's really exciting. So I guess now that's what I'm doing, but like ask me next week, it'll be, I don't know, whatever I'm doing next week. I love it. I love it. I think that there's also like, there's this misconception that people have that people stick in a code space and also that they know all the code spaces. So like, so we don't, we don't. And I want to get to that. And I think that like when we get in real life and we talk, I would love to, to, to dive more about that. And we never have time, right? Like, like we met, we met at a loud conference. I'm like, what'd you say? So I'm so glad that we get, get a chance to sit here and do this. And so I have, we have the show friends in code spaces. It actually doesn't have a theme song and we kicked it off a little bit earlier. If you don't, if you don't mind doing it. That's it. That's it. So now we've got the theme song. So we're going to reuse that. We'll, we'll get that going. I get royalties, right? Absolutely. Absolutely. Your country accent. We could probably work on that a little bit. Maybe it's better. Just bleep the whole part out. Okay. So we've got some games that we're going to play. So, so when friends get together, they play games and we are going to play this first game that we have is called context window. So you know how AI works. Maybe the listeners at home might not know. Actually, we don't know. I don't know how AI works. If somebody else can tell me, it'd be good. So there's training data and there's training data cutoffs. And like, sometimes it doesn't know and sometimes it hallucinates. And so what I did was I wrote a prompt and I wanted the prompt to say, I want you to, you have the chance to interview Edan Gazit from GitHub next. And what do you know about him? And I want you to be truthful about it. And then I want you to ask questions that would help you with your further learning in the future. So, so we're going to, we're going to fill the context window because unfortunately, it knows a little bit about you, but it really doesn't know much about, it knows about GitHub. A lot knows a lot about GitHub, of course. Right. So, so one of its training data sets, probably. But it is interested. It's interested in, in learning about, about you and about different things that you did. So these questions, so first thing I had to do was I said, generate some topics. So I did a nice chain of thought. I said, so with the information that you have, I want you to do this. And then there's, you had an exciting launch that obviously is not in the training window. So I added it to the training window a little bit, a little bit about it. So, and then I had it ask some questions, some, some topic areas. So this game is, you get to choose, you get to choose what you put in the context window. Okay. So can you see that? Is that easy enough to see? Okay. Do you want me to read them off? Yeah. Yeah. So yeah. Yeah. Okay. So AI and machine learning applications, co-pilot workspace. We'll definitely want to talk about that one. Human skills. Yeah. So I said, I said, what do you want to learn from Edon? And it said human skills. So I have human skills, I guess more, more so than the AI does. So congrats. Oh, cool. Okay. I don't know if the AI is trying to like butter me up or something. It's like, you have skills. I'm like, yeah, I do. I think I do. Innovation in software development, research, software development trends. I mean, what do we choose? Yeah. I have, I have lots of, I have lots of unfounded opinions about all of these. So I don't know. No, take it. It's your choice. Oh God. This is the game. It's this is the game. Maybe it's like software development trends. Cause I think that that's what I ended up thinking a lot about. Yeah. Awesome. So I'm going to do this. This is probably, this is in beta, but this is probably what would have happened as we were going to come in here and you, you get to choose. I think let's see what Antony sounds like. Okay. Okay. So Antony is going to ask you a question here. As we move forward, what do you think will be the key challenges facing software developers and how is GitHub next working to address these challenges and support the next generation of developers? That's actually a really good question. Wow. I mean, I don't think the challenge facing software developers has really changed over the years that the tools and the methodologies have come up, but we're all still grappling maybe even more so today with just the amount of complexity, the amount of abstraction, right? Like, you know, once upon a time computers were the size of rooms and only people with like a math degree could operate them because you needed to go and manually flip switches to like input, you know, here's the, what I want to stick in the register. And now I've got so many layers between me and the machine and I don't understand them all and I don't even need to. And so the history of software development has been this steady march of progress towards more and more abstraction. Like we went from assembler to higher level languages that were still close to the metal, like C, right? And so we got comfortable with a piece of software existing between what I say and what actually gets executed, right? Because it's taking my high level language and it's converting it to binary for that platform. Great. And then as time moved on, you know, now we have memory managed languages. I no longer need to like manage my own pointers and do math and things like that. Exactly. No more alloc and no more, oops, I freed something I didn't alloc or oops, I double malloced something like, you know, that's bad. So there's this march of progress that, you know, it happens in leaps and spurts, but it's always towards, monotonically towards this higher level of abstraction. And so I always ask myself, like, how am I going to come alongside a developer and help them to reason about their code base? And as much as that is about writing code, because at the end of the day, we're being paid to sit down and write code, even if the writing is hitting the delete key a lot, which is the best form of writing code. For sure. It's the hard part of writing code is not the typing. It's all of the work that happened leading up to the typing. Like, how do I identify where work happens or where this system exists and what I need to touch if I want to change it? Like, that's a perennial problem. It was a problem 30 years ago. It's a problem today. It'll be a problem 100 years from now, because we're just changing the scale of the problems that we're willing to attack. Like, previously we were building these little sheds, and now we're trying to build houses, and next year we'll build skyscrapers. We need power tools to help us with the parts that power tools can help us with. But the work of building is always going to be the work of building. So how do we come alongside that? How do we address the challenge of synthesizing understanding in people's heads? I think that's a forever challenge. Awesome. That was an incredible answer. With an analogy. I really like that analogy of us moving and building more and that we've been doing it. We've been building. And you've been building. Let's keep going. That was excellent. So that was the one from that that we had there. Do you have another pick here? I think it's your turn. I am interested in what it thinks about human skills, actually. Okay. Let's find out what's hiding behind door number two, Bob. So there's a couple here. This first one is actually, I think, pretty good based on the talk that I've seen you do. So it's, as someone who is heavily involved in AI and machine learning research, which you are, what are your thoughts on the impact of automation on human skills and software development, which we just talked about a little bit? And how do you think GitHub Next is preparing developers for this shift? I mean, at the risk of sort of repeating what I said a minute ago, but in slightly different words, I think it's about conserving the human attention for the things that only the humans can do. Like the models are good and they're getting better, right? To the point where maybe like in some future time, like not now and not next year, but I don't know, in five years, who knows the stuff moves so fast. But I don't, we're not looking to, and neither do I believe that we're really ever going to erase the human steerer from the picture. If anything, I think it's the other way around. We're going to have more developers because all these people like doctors and lawyers and journalists and biologists, whatever, like even a little bit of software in their lives can make a huge impact. Like a school teacher wants to automate something, right? And right now, like they're stuck outside of this, this like this futuristic, you know, utopia with their faces up against the glass trying to say, like, give me some of that. Like I've been, I've been using this Excel spreadsheet since 1993. And while it's good because it's what I can do, certainly there's something better, right? And so I think as we, as we move forward, we want to help developers by conserving their attention and shifting them to like higher order architectural concerns that they can leave behind the drudgery and focus on really the interesting parts of the work. While at the same time, opening the door, like we have this generational opportunity to bring more people in, people who don't have the advantages I had going to like, you know, college and studying, you know, computer science and this and that and whatever, like, but somebody wants a little app that's going to help them do whatever they do. How do we enable them to do that? Like, you know, and so it's like the same sort of technology can help us with both of these struggles. And I think, I think what you're, what you're talking about, there is a little bit of a human skill that you had right there. You had a little bit of empathy, I think that is, is not something that this guy knows about too well, right? Well, I mean, it can fake it. Yeah, but, but I, but I appreciate that. And I like that you, you bring that into your work a lot. I think that, that I see, I see your talks and I see you, you, you deeply thinking about humanity. So I appreciate that. Thank you. Thank you. Yeah. Let's see. So I would like to move into the next game. We, there's more questions here, but I think we're going to get to them through this next game. So this next game, you've, have you seen the show Hot Ones? No, I haven't. I just heard about the Conan episode, but I haven't watched it. Everybody said, you got to watch this. And I haven't. So it's like about eating hot stuff. Okay. Right. So we have one called Cold Ones. Oh God. So you get your choice. Do you want blue or, or a Coke? I'm going to go with Coke. Okay. Is this something from your, your past looking at this? Have you, have you had one of these before? I haven't had a, like a, like a big gulp since I was like a kid. Yeah. That's exactly what we're trying to do here. So cheers to that. Cheers to kids. All right, here we go. Go for it. I'm ready. Go in. Hit me. Do it. I'm not succeeding. Oh, there it is. Does that bring you back? It does. Yeah. It's like this, like immediate nostalgia. I thought, I love that about, about Slurpees that. I used to, I used to be tolerant of things that are this sweet. And now I have like, like a Coca-Cola and I'm like, oh my goodness, how do people like ingest this much sugar and not go into like shock? It's so good though. My kids would be all over this. It's a shame they're not here. They would demolish this in three seconds. We can, we can, you can take it with you. I'll take it as a doggy bag. Yeah, for sure. So did that, that brought you back to being, being a kid? Yes. And I want to get back to the place where you first entered your first code space, where was your first one? And it was, it doesn't have to be around Slurpee time, but I just, I'm getting you to think backwards here a little bit. Your first code space. My first code space. Well, I mean, the first, first was, I remember like logo on an Apple II, Apple IIc, IIb, like, you know, as I'm old, back then dinosaurs roamed the earth. I remember that. I was with them. Like that was, I remember that I was, I was transfixed and my parents for one of my birthdays, they bought me an Apple II, it was an Apple IIgs, which at the time was just like, whoa, it can do color. And I learned how to write basic on that. And I remember just being enthralled that I could, that I can make this machine dance for me. Like, you know, it's like, and all it did was like, you know, print, like, you know, like, like fart, go to 10, print again, like, you know, cause I was 10 years old or nine years old or whatever, but like, but that's, but that's what I, but, but I could make it do that. If I wanted to, I could have like an infinite, like, you know, like potty humor machine. And, and, and yeah. And from there, you know, I, I, I picked up a book on C and I was just like, oh wow, this is really hard. And then I picked up a book on Pascal and I was like, oh, this is still really hard. And then at some point like Java came around and I was just like, I think that was the first time that I actually started writing real code, but it all started with basic and before that logo. Yeah. Awesome. Awesome. And I think that a lot of people that are thinking to get in there, one of my, one of my code spaces is like teaching beginners how to code, right? So I like, I, I love that space a lot. And the, the first thing that you wrote, what was the first thing that you wrote in Java? So you, you, you brought yourself, by the way, let's do one more. Let's do another one. Here we go. Okay. Ooh. Yeah. I want you to remember that system out.println. Yeah. Yeah. Void public static void main, you know, think back to whoo. What was it that you wrote there that you were like, you know what, maybe this is for me. Wow. I don't even remember like really the early stuff. I remember, I remember doing development on the, on the Mac. This was like, maybe later this is like middle or high school. And it was, it was code warrior. Do you remember that? That was like the IDE at the end of the day. Like it had like, you know, this sort of like, it came in this box that had like the sort of like yellow and black construction, like, you know, motif. And, and I remember like getting like a window to draw and then doing like graphics drawing on it. And like my first program was like, I had to draw a hexagon. So I had to do like math to like, you know, figure out like, okay, it's, it's kind of like the logo thing of move to line two and then like changing the angles. But like, I wanted to be able to like, arbitrarily scale the thing. And for that, I actually needed to do more math. And then that was, that was the start of, you know, sort of my, at the time, my love affair with graphics programming, which persisted through high school and then college. And then I thought I was going to go do game development. And then I realized that working in game dev is a, is a rough existence. And I was just like, no, actually I want to, I want to do some other kinds of software where there's less deadlines and more ability to do random other things. So. Awesome. And, and how did, how did you get into the, the, the research side of things? How did that, what was that transition? By accident, largely. I think, I mean, I ended up on the research side, this wasn't an actual, like, you know, intentional path. It's, I had sort of. I think they really are, right? Yeah. It's like, I'd done a lot of different things. I'm, everybody remembers like Jack of all trades and nobody remembers the back half of the idiom, which is and master of none. So I, I, I'm a developer, but stack me up next to really good developers. I'm not at the top of that list. I'm a designer, but put me next to like real designers. And it's just like, come on, man, who are you fooling? But because I can sort of glue those two sorts of functions together, I've always ended up in these sorts of roles that where, where that is something valuable and not all jobs accommodate like people who want to like straddle a couple of lanes. And not all jobs value people who are hybrids because of that and master of nothing. They're like, oh no, we're only looking for the best developers. And I'm like, that's not me. My value lies elsewhere. And so through that, I ended up, I was doing that at Heroku or not really research, but just sort of like a front end back end, the sort of the human interfaces to our data products and Jason Warner, who was the SVP of Eng at Heroku. And then he moved over to be SVP of Eng at GitHub. And then at some point he wanted to focus more on sort of the forward looking sort of like, where does, where do we go next? Not like, how do I make the trains run on time in this engineering org, which is 1,500 people. And so I was talking with him when I left Heroku and he was just like, yeah, I'm starting this thing as the office of the CTO, what became next. And our job is to look a couple steps ahead and try things that might seem ridiculous, but could alter the trajectory of the business. Awesome. Awesome. And I was just like, okay, does GitHub really care about this? He was like, I don't know, come find out and we'll try and we'll see. And I was just like, okay, yeah, that sounds like fun. And so that's how I ended up in this research thing, not exactly by accident, but because we sort of put together a team of weirdly shaped hybrids to pursue things that seem ridiculous. What if we had AI that could actually help us write software? That was not obvious at the outset. Yeah, for sure. And so I think it really, I'm really bullish on teams of oddly shaped people that sort of have more than one skill working together. And so, yeah, it says research, but really it's prototyping. Love it. If you had some advice to give yourself when you were drinking that Slurpee back in the day, time travel, and say where you're coming from right now, what would that be? Make more. It's just 100% the best teacher that I've ever had. And I've had some phenomenal actual teachers teaching me. But the best teacher I've ever had is the sweat I invested by myself in front of my keyboard making things. And it doesn't matter. And you're always asking yourself, whatever I'm making, it doesn't matter. Nobody cares. It doesn't matter if nobody cares. It matters if you're exercising that muscle and building up those proverbial 10,000 hours. I never look back on that and say, oh, that was time that wasn't well spent. Awesome. That was a wonderful answer. Wonderful answer. All right. Let's cheers to that. And then we're going to put these away so you don't need to have any more. All right. We're going to switch into our next game. And producer Peter is going to come with a little iPad because we're going to play a game. You handled Slurpee pretty well. I'm very impressed by that. You know me. I can handle my Slurpee. Can we get these out? Oh, yeah. Remove these for sure. I don't need any more. We are going to do a little game here. So I'm going to give you a prompt. And I'm also going to give a LMM a prompt. And it's the same prompt. And so that prompt is you are incredibly thoughtful, which we've already talked about. You are an incredibly thoughtful human. And this is an incredibly thoughtful AI now that I told it to be so. So you're going to answer a question in five words or less. And I want you to make sure that it's the best representation of your idea, which is probably more for it. But you probably would do that already. All right. So I'm going to load the question up here. And we need to scroll down here to it. What does the future of coding look like with advancements in AI? In five words or less. You're going to scratch that on the board there. And then you're going to get to choose. Are you going to play? Is like when we cut the video, is there going to be like the Jeopardy music? I think so. Should I try to just do it now to make you feel that? What's the question again? That's why they play that music, right? Yeah. What does the future of coding look like with advancements in AI? Five words or less. Oh, no. I ran out of room. That's always happens. But it's four words. I came in one word under budget. Nice. Let's see it. Share it up here to this one here. This is not representative of my handwriting. It's finger writing. It's finger writing. It is more by more people. More by more people. That's beautiful. That's really good. And let's see how we do with long with three. And we got to talk about both of them. Code augmentation, not replacement, which I think is along the same lines. It's grammatically shaky, but sure, I'll take it. It's true. We are absolutely like, you know, I agree with the LLM in so much as it's, you know, right now, especially in the wake of, you know, you see like, you know, startups launch and they're like, ah, we're going to replace the need for software developers. I'm like, who are you kidding? Like Power Tools didn't replace the need for people building things. No, it led to us having more building happening because the Power Tools made it possible to build more. So I agree with the LLM. All right. Awesome. Great. So we're going to let's get to the next one here. Oh, no, I just, I just totally destroyed my next answer. Yeah. You walked right into that one. So we're going to ask this, what should you tell developers who are concerned about being replaced by AI in five words or less? Which I agree, I like what you just said there. Five words or less. Do, do, do, do, do, do, do, do. I came in one word under budget again. You are on top of this game. That's right. It's don't worry. I don't. That's great. I just I don't believe that it's that it's really going to replace developers. I really think that it's going to simply move existing developers up the stack. When was the last time you wrote Assembler? Yeah, I never have. Exactly. Like, you know, and yet once upon a time and still today, like, you know, if you're working in something that's like super performance, like compute bound, and you really want that inner loop to be as optimal as possible, then, yeah, you're going to bring in somebody who knows Assembler and they're going to get right down and they're going to optimize that tight inner loop. So there will be fewer of those. But way more of the rest, like, you know, so, you know, like, yeah, there will be some segments of software development that become, you know, esoteric or exotic, you know. But it's more than made up for by the huge amount of other software development that's unlocked and the kind of people that are going to be able to, like, step up to it. And I love your philosophy so much. I love when you talk like that. Focus on value additive tasks. That's kind of like, good luck, don't get fired. Yeah, I like yours way better. Yeah, I don't like the LLM here. The LLM is basically being like, I'm coming for you. Awesome. So let's get we got one more. One more here. And this is probably for everybody who might be interested in anything that you guys are putting out. How can I get on the shortlist to get access to the killer new tech products that are coming out of GitHub next? That's not on there. Do I do I get to, like, promote stuff? Yeah. Okay, hold on. Yeah. How do I get on that? What's the what's the move? And then I think I think this one, I'm going to let the LLM go first, because I really do want to know, like, what do you think? What do you think it says? What do you think? What's the? I mean, the LLM is going to say be famous, because it's true. Because famous people get it. Yeah, that's true. Like, you know, if like, you have 400,000 followers on Twitter. Here you go. Then, yeah, you better believe that, you know, if you're going to reach out to people and be like, I would like to beta test your thing and write about it. Yeah, look at that. Well, see, there you go. Yeah, I didn't. I didn't that but at least for a man, am I still supposed to do the thing? Okay, yeah. Like, in the case of GitHub and get up next. Yeah, specifically. Yeah. There is. And you're making you're making URLs in hands. That's amazing. I fit it. Success. It is gh.io slash next dash discord. And that's our discord. And basically, it's a community of folks who are interested in playing with things that are, you know, half baked, because they're all prototypes, and they might be buggy. No, they don't might be their buggy. And, but it ends up sort of attracting the community of people who are like, really willing to put stuff through their paces. And so we try to prioritize folks that join the discord and that are there. Also, when we're fielding these sort of these early products, these technical previews, we're doing it for a purpose. We're not just doing it, you know, for fun. We're doing it because we need people to use it. And then we need to hear their feedback. And we need to hear where it's broken or where it's not working well. And so we prioritize the people that are already like there and like, waiting to engage with us, as opposed to like, you know, while you're out there on the Internet, maybe you have some feedback, but how are we gonna know it? Awesome. So they're gonna jump in there. They're gonna ask, and I know the question they're gonna ask, because I've been seeing these demos. And what's that latest demo that's been going around? Can we talk just a little bit about, you're gonna give a demo tonight. Can you talk to me about what that demo is gonna look like? Do a spoiler. Do a spoiler. No one's gonna hear this. Spoiler. Yeah. Nobody? It's just between me and you. Yeah, just between me, because we're gonna have to cut this up later. The answer is 42. Oh, wow. Exactly. I've blown your mind. No. So the thing that we launched last week, I mean, last week, relative to when we're recording this, was Copilot Workspace. Singular, not plural. Copilot Workspace. Yes, exactly. Everybody who says it in the plural, like, you know, somebody will show up at your doorstep and slap you on the wrist. Not really. This is like spooning back the tide. I've already seen like a million tweets of like, Copilot Workspaces. I'm like, no. It's true. But Copilot Workspace is our first sort of stab at what happens after chat. Like right now, a lot of the AI stuff is all centered on various flavors of chat. Because the models are good at chat, right? That's what they were kind of trained to do. A lot of them. And reaching for more specific experiences is hard. Like really hard. And, but I don't think that chat is the be-all and end-all. It is, it has a place in the pantheon of user experiences. But how do we do something that's a little bit more actionable? Like you have, you have OG Copilot, which is suggesting like a line of code or maybe like a function, right? And then you have chat, which is good for sort of talking through ideas or sort of like reasoning about a space and sort of planning. But not really, it's not as directly actionable. Like I have a whole conversation with the LLM about a topic. I've now learned about state management in whatever. And I'd like to now apply that learning to like the file I have open. You know, it's on you to effectively like copy paste things out of chat. Yeah. So like I want bigger suggestions than like one line. And I want more actionable than chat. Copilot Workspace is basically a tool to help you go from like task to code with these moments of steerability in the middle. Because if you just try to one -shot it, you know, like pull the slot machine handle and say, I would like you to implement this feature. The chances that today's models like correctly understand all of what you mean when you say, I want this feature are low. Right, right. They're not likely to like, I mean, sometimes you get lucky, but it feels like a slot machine. So how do we create these moments of steerability where the human can be like, here's an edge case you didn't consider. And here's, so like we have this flow of like task to spec where, you know, the model first describes like, how does it work today? And where is it doing that work? And how is it that it understands what you want it to work like in the future? And once you've confirmed both its picture of like what exists today and what is going to exist in the future, then you're like, okay, like, you know, make a plan. And then it actually builds a very specific plan of like, in order to go from current to proposed, I have to touch this file and do these things and this file and do these things and this file. And then again, you have a moment where like, no, you forgot to like create the database migration or whatever. You can again, steer at that point and tweak it. And only once you click implement, then it actually generates the code. But by the time you get to that state, you've transmitted so much information about how to do it correctly, that it's more likely it's not a slot machine anymore. And on some level, by doing it, you're also preparing yourself for the review phase. Because if you just said like, do this thing for me, and then it spat out a pull request's worth of code, then you're sitting there being like, man, now I need to review all this stuff, which is not a good feeling. But here, it's like, by engaging with that spec to plan to code process, by the time you see the code, you have a pretty good idea in your head of like, what you're expecting to see, and it's right or it's wrong. But even if it's wrong, it's easy for you to spot like, where it's wrong and tweak it, adjust, whatever. So I think this is sort of our first attempt to pierce through that suggestion ceiling on like, you know, little suggestions. And I think we're going to see a lot more things that are focused on the cognitive work of coding that is not just generate me a line of code. And the demos that I've seen of it are incredible. And I'm excited to see you do more and more of those demos. Is there any place that you want to plug where you're going to be talking about this? Or any place where we should go to take a look at this? I don't know, the GitHub Next Twitter, for sure. Like, you know, we try to reshare also, like, you know, as people, as we're flagging people in, they're writing about their own experiences, and bad, you know, people are like, well, it worked. And like, it was great. But it didn't figure this part correctly. And it hallucinated there. Or like, we had one user who wanted to localize an entire repository worth of. And so the prompt was like, you know, like, translate this, like all the markdown documents into another language. And like, we didn't think about this case where it's not, it's not that it was supposed to mutate those files in place, what he was really asking for, but he didn't say it this way, was create new copies of every one of these files, and then translate them to whatever. But, you know, natural language, it's like, you know, if I say to you, like, translate this into that, you'll understand, make a copy. Yeah. But the computer didn't. Yeah. Like, so those kinds of things are like, you know, well, how are we going to figure that only by letting people use it? Awesome. So stuff like that. Awesome. And pulling in that research and research. And designing what's next. Awesome. Thank you so much for doing all that work. Thanks for all your thoughtful, again, about, you're so thoughtful about humanity. I really, I appreciate every talk that you gave about that. So thank you. I'm really gratified to hear that. That's the message that's getting through. Yeah. Awesome. Thanks for being on the show. Thank you so much.