Inside Roblox’s AI-Driven Employee Experience

🎧 Listen Now:

In today’s episode, we speak with Arvind KC, Chief People and Systems Officer at Roblox, about the AI-driven employee experience and how merging HR and IT is transforming their workforce.

Arvind KC leads efforts to evolve processes, systems, and culture, ensuring that Roblox’s teams thrive and reach their highest potential.

KC has over 25 years of executive experience, focusing on how technology drives exceptional organizational performance at incredible companies like Facebook, Palantir—where he was CIO and CHRO—and Google, where he served as VP of Engineering.

Here’s what you’ll learn in this episode:

• Why integrating HR and IT is crucial for a seamless AI-enhanced work environment.

• How to adopt AI tools where they can have the most immediate and impactful results.

• The importance of setting realistic expectations for AI’s productivity boost.

• How AI could reshape organizational structures and what you need to prepare for.

• Why starting your AI journey now—and continuing—is critical for long-term success.

Key Insights from Arvind KC

Here are the actionable key takeaways from the conversation:

1. Integrate HR and IT for a Seamless AI Experience:

While it’s definitely not the norm, merging these traditionally separate functions under one umbrella could dramatically enhance your organization’s ability to empower talent.

KC emphasizes that this integration is not just about efficiency; it’s about creating an environment where people can truly thrive.

2. Adopt AI Tools Strategically Where They Provide Immediate Value

‍AI adoption should focus on areas where it can provide clear and immediate benefits, such as in engineering and customer support.

For AI to be effective, it must be seamlessly integrated into existing workflows rather than imposing entirely new systems.

KC pointed out that successful AI deployment is about finding models that are fit for purpose, like GitHub Copilot in engineering or chatbots in customer support, and don’t require teaching employees new workflow

3. Set Realistic Expectations for AI’s Impact‍

In software engineering AI may make people up to 40% more productive, but KC cautioned to not think that’ll apply to all roles.At the same, he said that, "We’re overestimating the hype cycle driven by LLMs, but we cannot underestimate what AI can do in the future."While the hype around AI’s short-term impact is often exaggerated, the long-term possibilities are vast. It’s crucial to start small, scale gradually, and be patient as the technology matures.

4. Prepare for Changes in Organizational Structure

As AI continues to enhance productivity, it may lead to significant shifts in how teams are managed and how organizations are structured.

KC envisions a future where AI enables a different scale of management, possibly reducing the need for traditional managerial roles.
Jarvis-like AI collaborators could further supercharge individual contributors, meaning the org chart of the future could look very different than the one of today.

5. Start and Continue Your AI Journey

No matter your industry, the time to start with AI is now.

Experimentation and continuous learning are crucial as AI is not a trend to be ignored. The companies that begin and maintain their AI journey will be the ones that truly transform over time.
KC made it clear: "For all companies, there are only two things they should do: start and continue. This is not a trend you can ignore."

Learn More and Lead with AI

Subscribe now to Lead with AI and be the first to hear my conversation with another AI leader who's transforming their organization.

Want to level up your AI skills today?

• Join Lead with AI, a course and community for executives looking to embed AI in their daily work. Our next cohort kicks off on September 2. 
• Take our self-study courses on 10X Better Prompting, Building Your First AI Employee (Free), and 50% More Productive with AI.
• Check out our AI Top 100 and our picks for the best AI Websites.

Want to get your whole team involved? Check out our offerings for customized AI courses for leadership teams.

If you have any other questions or feedback, or would like to be considered for the podcast, just send me an email: daan@flexos.work. 

🔔 Available on:

Transcript:

Daan: You are both leading people and systems at Roblox. What does that role entail, both at the high level and maybe what does your day-to-day look like? 

Arvind: If you think of any company, there are three things that a company does. First, they build the product. The second is that they build a business. And third, they build a company. The simplest version of my role is that I'm responsible for the third, which is building the company. But the way we made it real was to take all of the functions that cause us to attract amazing talent, enable them to thrive, and bring them together in one apparel, specifically like classical IT, HR, InfoSec, facilities, and real estate. All of these contribute to creating conditions for how people work with each other and how people thrive in an organization. And that's my highest order limit: to create a really thriving, high-impact workforce.

Daan: I'm sure when you go to conferences and you speak to peers, that is not something that most people do. So what are some common questions that you get when people hear that you manage the whole tech side as well?

Arvind: Yeah, actually, it's interesting. Like, most people don't do that. You're right. Most people aren't surprised by that because they just see what usually happens. And my background is in engineering. If you're an engineer, you're not interested in some of other people's tasks. And if you are really good at thinking about how people and organizations thrive, then you have pursued that as a career, not engineering.

So people have difficulty straddling these two. The first time I straddled these two was at Palantir, when I was both their CIO and, for some time, their CHRO. And I think people appreciate and see that, yeah, this makes sense that you're responsible for the overall experience. So they get it, but I don't think many people do it still.

Daan: Clearly, I think it's two worlds that seem to be very separate, but especially to our topic of today, they are probably actually closer than ever, right? When we're talking about AI, for example, so what does your day-to-day look like? What are some of the initiatives that you work on? 

Arvind: Yeah, so it waxes and wanes between things that are very strategic, things that are like tactical and immediate crises of the moment. Ultimately, at any level of leadership, your job comes down to asking yourself, what is the most valuable thing you can be doing? And in case you're not doing that, what's preventing you from doing it—spending time on it and building systems that cost you money—is spending more of your time on things that are incredibly valuable.

So that's the principle of what I spend time on. So if I look at my day and say, am I working on the right strategic items? If the answer is yes, then I keep doing that. If the answer is no, then I know that there is something broken in the machine that I've built that I need to fix to avoid spending time on strategic stuff. And the strategic stuff varies across systems and across people.

Daan: Yeah, surely. You joined Roblox when the AI craze was already going on. It was at a point where, finally, most people started actually hearing about it. So at that point in time, was there already an AI strategy?

Were there already systems that Roblox was looking at? Obviously, it's a very product-minded company as well. So, for the internal teams, what did that look like at that point in time? 

Arvind: So Roblox is 80% engineering, so it's a very technical company, and we've always had a focus on ML and AI.

I think what you mean by AI becoming hot is that general AI became hot, right? And then people understood that AI could be real. In the product, we do a whole bunch of things. And ultimately, we are a platform for creators. So Gen AI is very applicable to unleashing creators at scale. So we have tons of efforts around that.

But the spirit of your question was like internal work. And I break it into a few categories. There are natural places like Code Assist, which work really well. And then adjacencies, like we can enable Gemini and other things for people and products where they are. There are also things like support, where it's been a very good use case that we have deployed. LLMs are good at reading something and then helping people in a simpler way with that.

And they don't have to read the whole thing. Where we have shied away from, which I think is implicit in your question, is where it has gone into the domain of decision-making. I think that the regulation around that is still not super clear for us, and I think humans are really important for the final decision-making.

And a simple example could be, should we hire someone? And there we have been tended to be human first and like really tried to keep AI as a copilot as opposed to the main pilot.

Daan: Super interesting. So there are at least two departments where it does play a pretty significant role. So one, it sounds like in engineering, which in this case is 80% of the company. And then there is also customer support. So maybe you can give people a bit of a look into, like, how Roblox operates, and especially in those two teams, how Gen AI was approached, how it was introduced, and what are some differences? I'm pretty interested. Are there some differences between trying to bring that into an engineering team versus maybe a customer service team? 

Arvind: I will answer the last question first, which is that there is not actually a significant difference. Because you look for three things to be true, one is, do you have a model that is fit for purpose? And second, can it be deployed where work is happening, as opposed to asking people to go somewhere else? And the third is, can it function as a copilot versus a main pilot?

If you think about the technical audience, and it says, you think of something like GitHub Copilot, you had a model that was fit for purpose because it was trained on a whole bunch of code and programs. It was deployed in the IDE where work was happening, and people really saw it as an assist as opposed to a replacement.

And the same as we do with customer service, we have a model trained on the right knowledge base. It is deployed as a chatbot because that's where people are interacting with our customer support.

And then it has the ability to say, Okay, this is beyond my capability; stop to a human, and that way it is seen as an assist, as opposed to someone who takes away your job.

So I think those patterns are very similar to how we've deployed them.

Daan: Was there a difference, though? Because that makes a lot of sense, there is a really good use case for it. And like you said, it's happening in the flow of work. So you're not asking people to go to a new platform to do something that's very foreign to them. It's happening where they already do the work. Was there any difference in how people were looking at it when it was just introduced in terms of maybe war customer service, people more frightful that eventually it would automate their entire work away, or were engineers more like that? Was there any difference in how people perceived it? 

Arvind: I did not see that significantly at Roblox, but I have, as I talked to many of my friends and colleagues, noticed that the technical audiences tend to have a deeper understanding of what the product can do, and hence I've not bought into the hype that this is a replacement.

Whereas the non-technical audiences are still trying to figure it out, is this a replacement for what I do? So there is some angst and, hence, a higher burden of AI change management that is associated with deploying these. But I think you could very easily explain and give a reason that helped people understand that this is taking away the grunt work of your job, allowing you to do higher value-added work.

Daan: Do you also think that because Roblox is a pretty technical organization, people just understand more intuitively that no matter what demo you see or what you're hearing about, it's not really realistic that, at least in the short term, Roblox is actually going to take your job away? Like, people just get it. 

Arvind: I think, like in every organization, you need a certain amount of work for people to understand and grasp it. But at Roblox, I have encountered excitement and curiosity as opposed to concern.

Daan: What are any specific things you put into place in terms of, like, how you introduced, for example, GitHub Copilot to the teams, or what platform are you using for the customer service side?

Arvind: Our customer service is a combination of internal bots that we have built, and we've also worked with Sierra. And, in terms of the differences between Copilot and Sierra, I think with technical audiences, which maybe goes to your previous question, things tend to be more widely adopted if they're useful.

And whereas with a little bit of non-technical organizations, you have to nudge to make a particular thing happen, largely the nudges are not extensive. It's minimal, and as people see the value of it, they adopt it quite well. Like in our HR organization, we have a bot that uses the leading LLMs. That is our primary support mechanism.

And the team has been extremely happy to adopt that. They can see that the moment they turn it on, it's the same question that is asked a million times and gets answered by the bot. So people just feel like, oh, my God, I don't have to do that. Somebody else is doing that.

Daan: Yeah, it's hard to say no to that. So that makes a lot of sense. And I saw on Twitter or X that you've been experimenting with some cloud agents. Obviously, especially within your very technical company, maybe people are also looking into what's going to happen after manually prompting something, getting some help in the flow of work, or maybe programming some GPTs. Then eventually, do you see a future where fully autonomous agents will also be a part of the organization?

And, like, how do you see the collaboration between the people in the company and those agents?

Arvind: I got really excited by agents initially when I saw them, not just the cloud agents, like what you could do with LangChain and then LlamaIndex. Then, I'm more cautious now than I was before, and I think the initial idea was to give a rough task to an intelligent coworker who's the machine, not a human, and then let them go and figure out how to do it and come back to you for some minimal checks. And then they keep doing the work that is there.

The actual implementation of agentic workflow just doesn't seem to be close to that vision. I was on the waitlist for Devon, but I never got it, and I don't think it's still open. I tried Open Devon, and that was not a fun experience at all.

So, I think part of the challenge is that agentic workflows, in some sense, rely on LLMs to have things that they don't intrinsically have. Like it cannot reason and plan. It doesn't have memory. And you need to, like, figure out a way to solve all of these. So there'll need to be some architecture changes before that becomes real. And so today it is still to me much more of a more efficient task completer, as opposed to some of them, I can give a project to a project just being a series of tasks that you go and figure out.

And so I still think LLMs are operating in task mode and not at a higher level, which is a project or a problem mode.

Daan: Do you think that we're going to get there? And if so, within how many years, because OpenAI just released that sort of like the five stages towards, like a fully autonomous organization? How many years do you think we're away from anything near that? 

Arvind: Again, keeping it gritty and real, I'm skeptical of that. By the way, I'm not an AI researcher, but my readings are more aligned with how Yann LeCun and Mattar are thinking about it. But I think that the transformer architecture has some intrinsic limitations, and that's not going to get us anywhere close to AGI. And a different approach is needed, which many researchers are working on.

So to your question on how many years it'll take us to be more autonomous as an organization, I'm optimistic, and I think that you can see that in five years or so, but I don't think it is around the corner, and I don't think it is going to happen through LLMs.

Daan: So how is it going to happen? 

Arvind: The alternate architectures that Meta is exploring, for instance, or Google is exploring. Yann LeCun has a good paper on it. To me, those are more promising than what LLMs are capable of. Then, in some ways, I think like this, one of the challenges is that there's a hype around LLMs because you've taken something that is very confident and articulate and associated that with competence.

That's a common mistake that we make. And hence, we think it can do much more than it actually can. And I think the ARC is a really good example of a test that LLMs fail continuously, because it's a simple puzzle that a six-year-old can do but cannot be done by LLMs because it's not from memory. They haven't seen it before. And so as we think about how we can truly enable reasoning and planning and then augment LLM in that architecture, I think that'll work.

Daan: It sounds like there's definitely a limitation because you're talking to a non-techie, and most of our community will be a non-tech audience.

There's like a limitation. So there is a lot of efficiency to be gained from AI. And for a lot of people, they're still very early in that journey of even learning how a ChatGPT works and maybe getting the most out of it before they really get all the benefits of AI. 

But it sounds like we are going to maybe ramp that up over the next couple of years. And at some point, there is going to be a plateau in terms of how much we can get out of it. 

Arvind: I don't know if there'll be a plateau anytime soon. I think the key point I'm making is that we tend to overestimate a trend in the short run and underestimate it in the long run.

So I think overestimating the hype cycle we are in is driven by LLMs. But I think we cannot underestimate what AI can do in the future. So, if you think about 10, 20, or even 50 years out, AI is going to be very much embedded in things and be part of your organization, but that's not going to happen right now.

In the current architecture, we still have a few technological breakthroughs that are necessary before we approach anything, namely, AGI.

Daan: I will ask one more question about the future, and then I'll go back to current-day reality. So the future question is: where do you see it eventually going? Let's say Yann LeCun and others solve this problem where the AI can do way more than the LLMs can do today. What does an organization look like at that point in time?

Arvind: I still think humans are a very integral part of an organization, but what you're going to see is a bunch of augmented intelligence. And for me, maybe the best model to think about is Tony Stark with Jason from the Iron Man series or the Iron Person series. I think that's going to be more of a reality. What happens to many people is that you can have a very intelligent system, but you have to figure out the user interface.

But you can have a very diligent companion that knows more about you, that can brainstorm with you, that can guide you, and that can help you do a bunch of current things. And how does it come as a form factor? Is it just like a thing on your phone? Is there a different form factor that people use? Maybe you wear glasses while you're doing things. I think all of those would be helpful, but I see the future where it's a very personalized assistant, like JARVIS is to Tony Stark. I definitely think that's in the realm of possibilities.

Daan: Much more than a ChatGPT memory can do right now, and just remembering where I live and what kind of job I do.

Arvind: We've all tried those things, but this is much, much more context; it could range from, if I'm maybe wearing some glasses, looking at a particular food. It knows me and my data in a very secure way to say, actually, that you should not choose that food. It's not good for you. You have not slept enough today. You have not exercised enough. Why don't you go here and get this juice? I would put those glasses off immediately.

Daan: That sounds terrible. I don't want to have that advice. Let me enjoy my ice cream. Come on. 

Arvind: Okay. Yeah. That's where the glasses would know you and say, hey, you like ice cream. Just go have it.

Daan: Exactly. Go straight. Take a left. There's the ice cream shop. Okay. So we recently saw this. I still don't know if it was a PR stunt or if they actually dropped the ball, but this whole idea of lettuce saying that you're now going to have AI employees in your organization.

So your point of view is more like, you're actually still going to have the same org chart as always, which is humans, but all those humans are going to have their own personalized AI. That's going to make them better. 

Arvind: I think the org chart will undergo some changes. I don't want to posit that the org chart will be the same.

I think the amount of work you can get done per person will increase. So that'll push towards a given organization having a smaller number of people for the same level of revenue output. Of course, that doesn't mean that the total work in society will decrease. There will be more organizations coming in.

I think the other thing that will happen is that the organizational structure could change significantly. I don't think you need it; you've classically had a 1:7 ratio of, say, a manager to a person, because that's the capability of a human.

But if you say that every person gets an automated agent as a coach and mentor, in addition to the human manager, you can start scaling things very differently. And I think that the skill sets that are required for people to be successful in organizations will undergo an evolution.

Once Excel and computers came in, you needed people who could build those tools. So similarly, you're going to have people who can read and like to work with AI in this augmented workforce. So organizations will change, but I don't think... Always, like in any technological change, jobs get eliminated, but work does not get eliminated. So like the sum total of work, there will be the same for humans. But the shape of that will change.

Daan: The shape of the work will change. So eventually, will everyone be able to use AI and have an equal opportunity to use AI? 

Some people obviously are way ahead, and I'm really curious. As you're talking to your peers, I'm sure a lot of people are going to bother you with a lot of questions, like, how does this work? How does this work? What tool should I use? What platform should I use? 

There seems to be a lot of companies struggling to just upscale people to get them to use AI tools, and we just had this whole conversation about how this one pharma company was reported to have dropped Copilot because after six months of trying it, they didn't really get anything out of it.

Then everyone replied and said, oh, yeah, that's because there wasn't good change management. There wasn't good training (see here for our recommendations on the best generative AI courses.) The AI is just not smart enough and intuitive enough for people to take it up in their workflow. Like you said, if a coder gets Code Assist, they're just going to use it. It makes sense. Like, why not? For other people, maybe the use cases aren't really clear, or people just don't know how to open the tool or where to do it. 

What kind of problems do you see in companies trying to get people onto AI? 

Arvind: Change management is always a big problem.

Most technology transformations fail because you're trying to change people, change processes, and change technology at the same time. So you go to reduce the variables of change.

One thing that you can do is start small by embedding this in things that people already do, like co-piloting on GitHub, which gives you a very good footprint. If people are already using Slack or a browser, can you bring AI to those tools? And through a browser, can you do a browser extension and bring it there as opposed to asking them to do something different? I think that's one aspect of the change.

The second aspect of the change is that you want to be real about the expected benefits. So if you pick software engineering, some of the estimates are... That's, by the way, the best use case for LLMs.

Some of the estimates are that it's about a 35%–40% improvement in productivity. So you're not going to get, for the best use case, 35%–40%. You're not going to get a lot more for the non-best-use case. So don't think of this as a thing that is going to completely revolutionize your workforce in a one-year timeframe; it is going to revolutionize your workforce in a 10-year timeframe.

So if you know something is going to change what you do in 10 years, the only thing you can do is be very consistent about it. Be small, be consistent, always learn, keep deploying it in multiple places where you can, and don't expect transformation results overnight; you will have transformation results in 10 years.

Daan: Do you think it makes sense for every company right now to do that experimentation and get into AI, or would you say, from your particular perspective, certain companies don't worry about it; maybe in two years, the software will be better, it will be more integrated, it will be easier for people to adopt, just wait it out?

Do you think everyone should get on it? Or is there still a bit of waiting as well? 

Arvind: No, I think this is long. And I'm not thinking about AI in general or Gen AI. I think AI is a trend that is here to stay. I feel like for all companies, there are only two things that they should do. They should begin. They should continue.

Daan: Okay. Definitely start, and definitely continuously learn. 

Arvind: It's not a trend you can ignore. Its equivalent to saying that in 2000, do you think a company should be on the internet or not? And you'd say, Should they wait for Web 2.0 to come? Because Web 1.0 is like a bunch of janky technology. And you're like, no, not really. You're going to get better with Web 2.0 if you start your journey. There's no right time. The right time is now. I think the classical thing is that the right time to plant a tree was yesterday. The second-best time is now.

Daan: Is it now? Okay. Then, how about on an individual level? And we can take this from really senior leadership like you to maybe mid-management to individual contributors. What kind of AI skills should people pick up? 

Is it again? Does it make sense for everyone to start going and learning prompt engineering? (Check out our prompt generator.) I know you had a post about how, for something like computer science, you need a lot of people's patience because not everything is going to work out immediately. And sometimes you have to break through pretty difficult problems, given that people are already very busy and feel like they don't have time to do their current job.

Do you still think everyone should pick up at least some AI skills? And if so, where should I start? 

Arvind: By everyone, I'm going to qualify it as everyone who's interested in the corporate sector and, let's say, knowledge workers. I have friends who want to be in national _____, and I think they should do that.

But if we take the context of knowledge workers, the short answer is yes. Let's talk about how. I find enormous use cases that help me on a day-to-day basis.

For instance, in Gemini, you can give it a YouTube video link and say, Transscribe this for me in less than a paragraph, and it gives you a very good summary. You can upload your video in the AI studio that Google has, and again, I'm using Google as an example, but you can do it in multiple other places. And then ask it to say, and I've uploaded videos of me playing tennis or swimming and said, tell me how I can improve my technique. And actually, it does a pretty good job of that.

I've been able to create storybooks for my kids. You can just have Gen AI draw diagrams and write stories about the values you want to teach. So, there are tons of use cases that you can use and have fun with. So, I would encourage people to always get in with that. So that's the knowledge worker who's not a technologist as the core thing.

If you're a technologist, and especially if you're in the space of ML and AI, you need to think of this as a 10-year learning journey, not like a one-year, 24-hour learning journey in AI. And that doesn't mean that you have to wait two years to build your models. You can take the approach that companies like Fast AI have, where you start building and training models in the first hour of your lessons.

And then you just keep unraveling the layers of an onion. Sometimes you'll go down the path of model building. Sometimes you'll go down the path of mathematics. Sometimes you'll go down the path of building applications, but think of it as a 10-year learning journey.

Daan: It sounds like a long journey, but it makes sense because these are pretty fun and meaty subjects. 

Arvind: Talking about a 50-year trend, just think about how far ahead you will be in 10 years for a 50-year trend.

Daan: Yes, it's been a long journey already, and surely there's a lot more to come.

We're at the end of our time. So I just have a couple of quick-fire questions. And number one is going to be your favorite AI tool. 

Arvind: Very tough. I thought about this, and it's a tough one. I like Suno quite a bit for the music it generates. That was a lot of fun for me.

But the most used one is one I want to make between, like Claude, OpenAI, and Gemini, and especially like the AI studio Gemini, because I think the multimodal capability is better than what I see in other places. Meta is great for images, by the way.

Daan: Are you using all of them side by side? 

Arvind: Yeah, I've subscribed to all of them because I want to understand how they work.

Daan: Okay, hopefully with the company card. Then, consider one unique use case for AI in your daily routine. So I think we already heard one in terms of getting Gemini to critique your sport, but what else do you use it for? 

Arvind: I usually find it very valuable to have someone explain things to me in simpler terms. My previous pathway used to be that if I understood something, I'd Google it. And then the 10 blue links were very onerous for me because I had to click each link and try to summarize and synthesize it. Whereas now, if I want to understand a concept, I'll just say, Explain this concept to me. And if I don't get it, I'll be like, okay, make it like a five-year-old. Give me a few examples. And I just keep going deeper and deeper, like now trying an alternate explanation.

That really helps me learn things faster, and it can range from, as we've had, scenarios where, if it's a family medical thing, you share an image and then you say, Explain this to me, which comes up with a bunch of medical gobbledygook, and then you're like, Okay, I don't understand any of it; tell me better.

Daan: That's where you're getting towards that JARVIS idea of, okay, it's really tailored to me, and you can use analogies that I get and from my context, so that's where it starts to become more personal already. 

Arvind: AI cannot be about you; I know you. It has to be that you know me so that you can have a better conversation.

Daan: Yes. Obviously, the more data it gets, the better it will become. 

Arvind: There's a scary, creepy part of it, but as a technologist, I think that can be solved.

Daan: Yes. I think that's where it always gets a bit worrying when people say, Oh, I got this new AI plugin, and it's recording all my meetings. I was like, Oh, where's that data going? That sounds scary. 

Arvind: Maybe blockchain finally has a valid use case, like being decentralized stuff somewhere and not in a single entity.

Daan: Right, running locally, or anything like that.

Then comes the final question. So what should business leaders do? What's the one thing that they should do tomorrow if they want to get themselves and their organizations ahead with AI? 

Arvind: Can I give you two?

Daan: Yeah, of course.

Arvind: I think the first is that whatever project you think has some traction, just get started on it. Can be scrappy. Don't worry about it. Just get started on it because the lessons you'll learn from it are much more than trying to wait for the stars to align.

My second thing would be to find a way to spend some time with the students in the university who are working on ML and AI, because there's a sheer excitement and knowledge that they bring. That most executives are out of touch with. I love connecting with students at Stanford, Berkeley, or Santa Clara University. And every time I speak to any of them, I walk out. Oh, wow, that's great.

Daan: That's so cool. Okay. Number one: get started on your AI experiments. And then, number two, find a couple of ML students. Sounds fantastic. 

Arvind: And hang out with them. 

Daan: And hang out with them and get excited. Okay. KC, I'm excited. Thanks so much for being on today. 

Arvind: Thank you, Daan. I enjoyed the conversation as well.