Stephen: This is your host Stephen Sargeant the Around The Coin podcast? For two and a half years, I sat behind the scenes producing one of the best compliance podcasts in the world. The Public Key podcast by Chainalysis, and we now have the former host, Ian Andrews, who's been in the blockchain space, the tech space.

He's probably one of the most technical marketers that aren't working in product marketing side for some of the biggest companies, but he is now the chief Revenue Officer. Of Grok. He is in AI and he is gonna tell us all about ai. He's gonna tell us how me and you can use it, what we should do to start our learning journey.

What's happening in this space? What does inference mean? What does training mean? What are the chips doing? Who's Nvidia? Are they a competitor of Grok? What's the difference between this grok and Elon's Croach? He is gonna spill the tea, as they say, on everything that's happening in the ai. We very rarely do an only AI post 'cause frankly, most companies probably just wanna shill whatever product or rapper they've created.

But Ian is so thoughtful about the way he approaches new spaces, and he is going to give us the goods in this episode. If there's any episode that you wanna learn about AI and he even tells you who to learn from, this is gonna be the one right here. Go listen to it and go, you know, check out Ian on LinkedIn and Twitter because he's putting out some pretty good content and we're hoping he comes out with his own podcast on AI

really soon.

Here's the episode. I know you're gonna love it.

Stephen: This is an Around The Coin feature episode. 'cause we got Ian Andrews, who's now the CRO of Grok, but me and Ian used to work together. I was behind the scenes when Ian produced over, I think, 150 episodes of the Chainalysis Public Key podcast. So if you wanna see an excellent host, have some of the fa fascinating conversations.

You guys gotta to check out the early episodes of Theis Public Key podcast. Ian, we now have you in the hot chair. To talk about everything AI, 'cause now you're the CRO of Roc. So maybe just give a little background of who you are from, maybe your VMware days all the way to now.

Ian: Well, I don't know how much time we have. I, I'll, I'll do, I'll do the quick run through. So I've spent you know, now 25 years in early stage technology companies. Some that might be memorable early in my career, worked at places like Opsware and Endeca More recently, you know, spent nine years at a company called Pivotal that we took public in 20 18 and later sold to VMware for about two and a half billion dollars.

I spent four years at Chainalysis. Leading the marketing organization and just in December joined gr, which is one of the hottest companies in the artificial intelligence space, and it's been an absolute blast.

Stephen: The hot. I thought you were gonna say what you spent, you know, when you say you, you exit the company and you're part of that for 2 billion. I thought you were gonna say, and I spent it on the Lambo or something cool like that. But tell me, in the world of tech, like that's like pretty much three winners, right?

Like you go to, you know, before VMware acquires. The company then Chainalysis and now grok, which is like nobody now, I wouldn't say nobody really knows about it. And then it is like all of a sudden I'm hearing Groq, Groq, Groq, , you know, you know, battling Nvidia, like, it's like obviously one of those up and comers, that's a serious contender in this space.

If not taking a, know, significant market share. How are you maybe for those entrepreneurs that are like, Hey, I want to go back in and work with companies, or, you know, for people that are in payments and tech, how are you picking these tech winners? Is there any, you know, way you approach looking at these jobs or you're just like dart dartboard and going for it?

Ian: There's a little more I think calculus behind it than just throwing darts at a board. You know, for me I've always been tech first. It's, it, it is hard for me to get excited about. I. Working at a company if I'm not fascinated by the technology, you know, joining Chainalysis, it was a space I knew very little about.

But it seemed like they had really cracked the this concept of looking at data on the blockchain and making it useful to humans. I got really excited about that, that drove my decision to join the company. With rock, you know, I spent maybe the better part of 2024 looking at ai. You know, I was, I was making up projects at work actually to, to use AI to actually try and apply it in our jobs to make selling more efficient, to make our marketing team more productive.

And at some point I realized, Hey, I need to be in the space and started talking to a, a number of different companies. I. To really, you know, see where the opportunity was. And what became clear to me with Grok is they had built something incredibly special with a very clear value proposition, which was in order to run AI models, you needed to be fast, you needed to be high quality, and you needed to be affordable.

And grok is all three you know, so 10 x faster than an Nvidia, a third, the price, a fifth of the power consumption. It's a winning story. And, and the decision to go here was also driven not just by that technology, but by the market trend, right? Me, it became really kind of obvious that the next decade is all, all about AI moving into every piece of technology, right?

This is gonna change how we live, how we work. And so you, you jump on the, the market trend with a great company. It gives you the opportunity to have an outside success which is what we're aiming for here at.

Stephen: I think you're right. There was kind of these phases, like I'm, you know, going into my 44th year, it's like the first kind of 10 years was like all about the new internet and then there was like social media on the internet. Then it was like, Hey, blockchain for every company. And now it's like AI is kind of that baby where it's like.

We need to implement it somehow. And I remember having those discussions like, Hey, what are you talking, learning about? And you would always bring up like, yeah, I'm dabbling in ai.

Ian: Good.

Stephen: I'm curious with, you know, so much VC money in the AI space.

Ian: Yeah.

Stephen: did you look at, like, I'm sure a lot of companies like, hey, we can do this and we

Ian: Yeah.

Stephen: funding, but you know, everything has that AI type of wrapper versus like, Hey, this is going to be the picks and shovels of ai.

Did you

Ian: Yeah.

Stephen: lot of companies that had that funding that were, Hey, we need a marketing or a revenue person to really drive this. How did you avoid those shiny objects? I would say in your career.

Ian: Well, I talk, I talk to a lot of companies that I think fit the profile of what you're describing, which is raised a lot of capital, had built an interesting product that maybe was seeing some growth. But for me there was, there was a lot of skepticism about who ends up as an enduring winner here versus a flash in the pan.

Stephen: Right. I.

Ian: It turns out, you know, in the age of ai, it's incredibly easy to get a prototype product to market, and there's a large cohort of users who are very excited to try anything new. And you know, the question is, is do those new users stick around beyond, you know, the first or second monthly bill? And and can you continue growing after that initial pop of excitement?

And do you have any differentiation? And I think I talked to a lot of companies that couldn't really answer those second and third questions, right? They got out of the gate quickly and then they're like, oh man. Like growth is slowing down. Users are churning out. We're not quite sure how to retain folks or be competitive.

'cause I think there's two things that you've gotta have to be a successful company, right? You need a great product that's well aligned to the, the market demand, but you also need to have like a distribution strategy. And, and in ai it's kind of unclear right now, will the market produce a lot of startup disruptors or will the big incumbents who already have distribution advantage win?

To give you a specific example, like, you know, Google is building great models that are very affordable. They're running them on Google Cloud, so you and I can go use Gemini, but they're also building great products that just kind of plug into the overall Google Workspace environment. If you've used Google Slides lately, you probably have seen, like Gemini will now create slides for you in the product.

So am I gonna go use a third party product that builds slides? Like that's a hard decision to get a large number of users to move to that new application if it's already in the product you use. And Google can offer a good enough experience and probably capture a substantial portion of the market.

'cause it's just there, it's a feature rather than a, a standalone company. So for me, making that decision was really trying to qualify who's got the potential to be a standalone, enduring company. Versus a feature in somebody else's product. And when I looked at Grok that, that became, you know, very clear to me that there was, there was opportunity for us to build something that will be here for, for years and decades to come.

Stephen: That's so funny 'cause when I was designing the questions I was, I'm using Google

Ian: Yeah.

Stephen: like there was this little thing that said, Hey, Gemini wants to know if you want to shorten this question a

Ian: Yeah.

Stephen: like, Hey, you know, I'll click on it. Like I'm not gonna go Google, get Gemini to use it or download.

In on, you know, on browser. I'm like, Hey, they did a pretty good job and I'll tell you which question it is when it comes up, but

Ian: That's right.

Stephen: that's interesting. I wouldn't go out to find Gemini probably, but now they've implemented in the things I do every single day

Ian: exactly.

Stephen: wouldn't I just click on it and try

Ian: Yeah, there's, there's millions of users using Google Workspace every day, and so those millions of users now you have like as a startup, offering an equivalent product, such a higher bar to go win them as a customer.

Stephen: And if someone from Google is listening, if they could find a way to take these large video files that I have and be like, Hey, do you still need this for a client? Or can we delete this and, and free up a little bit more space for you? I know, I know it comes up when I'm at like 90%, but you know, if they could just do that on a regular basis. curious, Ian, why did you go into like general marketing and,

Ian: Yeah.

Stephen: chief revenue officer, you're super technical. I would think people that with your, you know, skillset they might go into like more on the product marketing side

Ian: Yeah.

Stephen: know, more on, why did you go into like, Hey, we're gonna lead the company and I'm gonna do this overall marketing.

Because I think you're

Ian: Yeah.

Stephen: of the most technical marketers that I've seen and I've talked to a lot of marketers

Ian: Yeah. Well, I, I always think about my interest and maybe depth in technology as being my secret weapon, right? It, I, in every role that I've had from when I was first starting out as a, an inside sales rep, you know, a few days out of college the thing that was very clear to me was. Understanding the products that I was representing allowed me to be much more valuable and, and build much better relationships with customers and prospects.

And so from, from the earliest days of my career, I leaned into that. Everywhere that I've worked, you know, I've spent a lot of time with our engineering teams and our product leadership to really understand. Both our technology and how it, you know, fits in the market. And as I've kind of moved up the career ladder, I would say I, you know, even though I've had marketing and, and sales leadership roles, I spend a lot of time collaborating with the product and engineering team and helping to shape the, the product roadmap.

And I think that's kind of set me apart. It's given me many of the opportunities that I've had in my career. As a result of that, not sitting back and saying, Hey, I'm, I'm just here as the guy who gets leads. I don't really understand what the product does. And you'd be surprised how many people in marketing or sales, you know, take that approach.

You know, I see sales rep, bad sales reps who show up to meetings and they introduce everybody and they say, okay, I'm the guy that, you know, brought the donuts and the coffee. I'm gonna turn it over to my sales engineer and he's gonna explain what the product does. Like I, I just don't see that as being like a valuable role.

And so I think for people listening who are kind of trying to plot their career strategy, if you are, you know, technically adept, like lean into that doesn't necessarily mean you have to be a technologist as your primary job, but like we live in a world of technology. Like know what you're talking about, invest the time, get smart.

Stephen: And I think that's what we don't see a lot. And in all industries, it's like the marketers know how to market, but they don't know exactly the fe, they're telling you what the features are, but they don't really know how those features solve the problems for the customers.

Ian: Totally.

Stephen: never, as you said, they've never gone deep enough to understand that.

And I think that's a disconnect that most people, everyone's had when they've talked to a sales rep, it's like, Hey, can you do this? And the sales rep's like Yee. Yeah, let me get back to you. Let me talk to the product person.

Ian: Yeah.

Stephen: them on our next call. And it's

Ian: Yeah. Yeah.

Stephen: have somewhat of an understanding of like, hey we, it's possible, but it's some nuance to it. Take me back now, 'cause I remember when we started the Chainalysis public key podcast, what I thought made it so successful was you said, Hey, you know what? I wanna take people on the same journey I had when I started out in blockchain.

Ian: Yeah.

Stephen: know, the first couple years I'm learning. Let's do this for ai.

Take casino. You've only been there a few months,

Ian: Yeah,

Stephen: take us through that journey.

Ian: yeah.

Stephen: you learned? What's a big myth in the industry?

Ian: Yeah.

Stephen: walk us through some of that ear, those early stage learnings that you're like, oh, this clicked for me, or This makes a lot more sense. Or from the outside,

Ian: Yeah,

Stephen: seems like this, but it's really like that.

Ian: well, I think there's kind of two camps. If you, if you look at the broad world of one AI's gonna replace all the humans and no one's gonna have jobs that's like the, the idealistic sort of ultimate productivity standpoint. The, the other camp is, oh, the AI's going to, you know, build robots who are gonna kill all the humans or enslave us, or something silly like that.

Neither of those stories are true. But if you read just, you know, headlines in, in mainstream media, you would get to that conclusion. That one or of, of those two things is the future. I think the reality is we're about to see an explosion of productivity. Like the humans are gonna get an order of magnitude more capable.

And, and this is in every discipline. So if you're a salesperson, if you're a marketer, if you're a product engineer, if you are a software developer, if you're a product manager, like pick a role and AI is going today can make you. You know, three times, five times more effective. And I think over the next 18 to 24 months, you see that move to like 10 x to a hundred x more productive.

Stephen: Wow.

Ian: And so, you know, my encouragement would be don't sit back and try and read about this. Like go use it. And, and you can do it in really small ways. Like, one of my favorite products is superhuman. It's an email client. I've been using it since Superhuman first launched. They're, they're funded by my good friend Ed Sim.

So I got like the early beta invite when the product first came out. And initially their pitch was really simple. It was like incredibly fast email. So it sits on top of Gmail or Outlook goes really fast. It's got a great mobile client and it's a user experience I can't get away from. It's a product I pay for outta my own personal pocket, you know, a significant amount of money every month.

They've recently introduced ai and at first it was like, oh yeah, we'll, we'll help you, like, refine your email or you know, maybe craft an auto response for you, which were not features. That for me, drove a lot of productivity. You know, oftentimes the, the tone and voice wasn't quite right, but just recently they introduced a feature that creates automatic follow up emails.

So it looks at your message that you've sent to somebody and it will, you know, if it goes by without a response for a day, it will automatically populate back at the top of your inbox, a pre-draft email that follows up on the previous message. And this is so awesome to me. It is one of the coolest features that is incredibly time-saving.

I know I'm gonna have to track like, you know, did this person get back to me?

Stephen: Right.

Ian: know, do I have a note of like, you know, follow up with this person if I don't hear from them in a day? And then I gotta sit down and write the email. It's like all I have to do is hit send and

Stephen: That makes a lot of sense. I think as you know, anyone working in any function that seems like, oh, Ian, that's not a huge deal, but when you're like, Hey, I'm a business. I was supposed to get back to this person. Oh, it's only been a couple days, and then you go back to your Gmail, it's like 17 days ago, you sent them an email.

You're like, whoa, whoa. How did two and a half weeks go by

Ian: totally.

Stephen: following up with them?

Ian: Totally. And you know, for people like ourselves, you know, I'm on Slack and Signal and Telegram and iMessage and email. You know, like just keeping track of all the conversations is difficult. So having the app actually take control of that and go, oh, clearly this is a conversation that needs following up on, and the thread's been dropped, and it is like proactively prompting me is an incredible productivity boost.

But there's no, nothing technical about it at all. In fact, it's the reverse, like it doesn't require me to know or understand AI in the least. But it, it's incredibly helpful. I'll give you another example, which is for anybody who's, you know, in a, in a sales role, I. You, you probably do customer proposals all the time, you know, and often you're responding to requirements the customer gave you.

So, you know, maybe a formal RFP or maybe just a basic one that's like, Hey Stephen, I need a proposal that needs like these 10 things. So the other day we had one of these pretty comprehensive, you know, like four pages of requirements covering, you know, technical details all the way to like economics and, you know, details about our support offering.

SLAs that we offer, we put together a, a package like comprehensive proposal. A bunch of us reviewed it, you know, so a lot of time went into this and we said, Hey, before we submit it, let's just ask one of these LLMs if we've done a good job. And so we uploaded the the RFP document, the original requirements, and we uploaded our response and said, Hey.

You know, what do you think? Would you improve anything? Like really simple prompt. There was no like crazy prompt engineering going on here. And the output was like, well, you know, you're, you said this in response to question like two, and then you seem to contradict yourself down here in question 15,

Stephen: interesting.

Ian: really answer question eight.

Like you, you sort of missed the response there. And you know, question 22 felt like your answer is a little confusing and you should consider revising it in the following way.

Stephen: I love that.

Ian: Right. And again, like you don't have to be a technologist to do this. We literally uploaded two files that we had already created and, and got a response back, you know, almost instantaneously that was incredibly thoughtful.

That had, you know, things that we had missed as humans reviewing it. And so, you know, probably saved us you know, a day's worth of extra review. It would be like the way to quantify that and helped us produce a much better response for the client.

Stephen: That's amazing. Can you walk us through what Grok is? So it's funny 'cause I sent you a message, I think it was a few months ago. I'm like, oh my God, they're talking about your company on the podcast.

Ian: Yeah.

Stephen: have to go listen to this episode. And you're like, no, that's not grok that I work for. It's like a, you know, grok ai, which is like Elon's thing.

Ian: Yeah, yeah, yeah. Totally.

Stephen: Context is,

Ian: Yep.

Stephen: what's GR that you work for and gr that

Ian: Yeah. Yeah. Yeah. So Elon Musk's company is called XAI and XAI has a product their their LLM, which is called Grok with a K. That's definitely not my company. My company is Grok with a Q. We were founded by a gentleman named Jonathan Ross. Jonathan is famous for his work at Google where he built Google's AI chip, which is called the Tensor Processing Unit, or T-P-U-T-P-U today runs the artificial intelligence and machine learning services behind all of the big well-known Google applications.

So it backends everything from like Waymo's self-driving cars to Google search to Google Translate to a ton of other services. So basically if you've ever done anything with Google, you've kind of touched the TPU. And Jonathan, while he was at Google, had this realization that while Google was clearly far out in front of the market the rest of the world would ultimately catch up in terms of their needs around machine learning.

And so he left Google in 2016 to found rock. And build for the world, the technology that would enable them to run run, you know, the algorithms and, you know, machine learning tools that would power their businesses. Now, like a lot of visionary founders, Jonathan got the direction right, but the timing totally wrong.

Stephen: That's gonna, that was like nine years ago. We haven't been talking about ai. It's, what was he doing for six? The first six.

Ian: Struggling to sell products. You know, and he, he'll, he'll be very open about this, right? So the team, you know, he assembled a world class team who built the grok chip. So at the core of our technology stack, we actually manufacture, design, and manufacture our own semiconductors. We call that an LPU.

And at first there really was no workload that would warrant a customer. Switching from, you know, whatever their default kind of CPU or GPU product line that they were using to something like rock, right? An unproven unknown startup showing up with this crazy new chip design. It wasn't something that a lot of people were ready to jump into.

And, and so they struggled commercially until really the last two years. And in that time, there were two big unlocks. First chat GPT launched. You know, that was a little over two years ago and that was the moment where I think the world woke up, woke up and said, whoa, AI is real and it's gonna change everything about how we live and work.

And then as people started digging into it, they were like, this is going to command a totally different compute architecture. Like we can't just deploy this on some random EC2 instance, and it's gonna power our business. Obviously Nvidia has profited enormously from this, right? You, you probably saw their recent quarterly earnings, 40 billion in revenue, 70% profit margins.

They're, they're making a bank. I think they're now the second or third most valuable company in the world. But people looked at that and said, well, but there's gotta be alternatives, right? You know, it's, it's hard to get Nvidia chips. It's hard to. It. They're expensive to acquire, expensive to operate, what else is available in the market?

And that led this moment of people getting very interested in grok. And so that really kicked up in 2023. But I would say we still had a product packaging mismatch because AI is really a developer led transformation, right? People are building applications that use ai. When you show up to a software developer and say, Hey, I can give you faster, cheaper super high quality inference, people are like, yes, I love that.

Please, please do. And then you say, here are my chips. They're like, dude, it's 2025. What am I gonna do with a bunch of chips? Like, that's not my gimme an API in the cloud.

Stephen: Right.

Ian: And so our second big unlock was at the beginning of 2024 when we launched Rock Cloud. Because now you didn't have to take my word for it.

That grok is fast and that grok is high quality and that you know the prices are incredibly reasonable. You could go try it. So right now, anybody that's interested can go to console.grok.com and create an account in about 30 seconds. And you can be using GR and you can experience the speed for yourself.

You don't have to trust me, you don't have to. You know, hope that some slide I showed you in a meeting is true. Just go set it up and if you've built something on open ai. We implement their API stack. So if you've already got an app running, you can take it and point it at rock's API in about a minute.

You know, the, the most minor code change you could imagine, just repointing to a different, different endpoint. And, and try us out. And that experience has been incredible.

Stephen: So it was like you're saying, Hey, here's a chip, and they're like, well, now I have to figure out what to do with a chip to

Ian: Yeah.

Stephen: faster. Now with Grow Crowd, you're like, Hey, just we'll use the chips

Ian: That's right.

Stephen: You just jump on and use the technology. Is

Ian: right. That's it. We took all the complexity out of it. So instead of you, you know, buying chips from us or buying, you know, servers full of chips or racks full of servers, and then we will, you know, ship 'em to you in your data center and like we took all that complexity out of it. So we've built a global network incredibly scaled.

Highly redundant. Always available and, and you can, you can try it out right now. And because of that, we've seen over a million people sign up for Gro Cloud in the last 13 months. We've got 175,000, you know, monthly active users. We've got thousands of people now on our paying tier. It's it's been incredible growth.

You know, we're, we're now serving tens of billions of tokens every single day out of Rock Cloud.

Stephen: That's crazy. You know what's interesting is I was watching the clip of a podcast with the CEO Jonathan. He's basically saying, Hey, we're not. We're doing a favor for Nvidia. We're not really competing with them. We're taking the thing that they don't really wanna do and we're doing that the best.

Ian: Yeah,

Stephen: like competing, like who has the best chip?

Ian: yeah.

Stephen: of explain that concept of like, is Nvidia only chips and you are chips and like access to the technology?

Ian: Yeah,

Stephen: kind of the difference between GR and nvidia?

Ian: I think maybe stepping back a little bit, the, the big there's two big workloads that you're gonna do with GPUs or with grok. You know, one is training where you're actually creating a model. And, and if you maybe read in the news about Elon building this mega cluster down in Memphis, Tennessee, and doing it, you know, in the record time.

That's a cluster for training. And training is a, it is a fairly complex process. You need to assemble a ton of information you know, complex pipelines, like really very technically complex. And there's, there's probably, you know, tens of companies in the world who are doing training at scale. You can think about, you know, the big model providers, open ai, philanthropic you know, the big cloud providers.

Like Google obviously does a lot in the AI space, specifically with DeepMind. XAI obviously doing that as well. Companies like Misra, it's a relatively short list, you know, in the, in, you know, compared to all the companies in the world. But those firms are spending hundreds of millions, probably billions of dollars on training infrastructure.

They're building incredibly large clusters of Nvidia GPUs now. Grok doesn't do that at all. That's not our business. We don't do training at all. We only do what's the other side of the market, which is called inference. And inference is kinda one of these fancy, like inside the technology bubble words.

When I got here, I was like, why does everyone keep saying inference? I had to Google it. Easier way to think.

Stephen: it and I still didn't understand. I Google before this session,

Ian: Yeah.

Stephen: what it meant.

Ian: E, easier way to think about it. It's the part where you run the model. So after you've trained the model. You're gonna run it so users can actually interact with it. So if you've ever gone to chat GPT and typed in a prompt, that's inference, right? Nothing more or less than that. It's a

Stephen: is like the amount of data that CHA PT is ingesting, so that when you do that, you know, run that search,

Ian: Yep,

Stephen: able to spit out the answer. Okay?

Ian: that's right. And so training is kind of a, a, generally like a one time process. So for each version of the model that exists. It's been through, you know, a, a process that, you know, took anywhere from like three to 10 months. But it, it's one off. And at the end of that training process, you have a model and then you're gonna run that model.

For every user that touches it or interacts it is, is an inference request. And so if you just kind of think about this from a market sizing and opportunity landscape, you know, there's, let's call it 50 companies that are doing serious at scale training. And then you've got, basically every company in the world is gonna do inference, every application, every piece of software will incorporate AI over the next couple years.

And every human on the planet will be constantly triggering or interacting with these models to produce some output, right? Like you and I are, we're recording this podcast right now. You're gonna take the output of this podcast and you'll, you'll feed it through some sort of software that will produce a transcript that's inference.

Right. You'll then create a bunch of social media clips that automatically edit the videos and create some tweets and some LinkedIn posts. You'll feed that through an LLM to either write it the first time or maybe clean up what you draft. That's inference. So now multiply that times, you know, 8 billion people on the planet, and inference will be the much larger workload.

I mean, even Jensen Wang and Nvidia said it's a billion times bigger than the training opportunity.

Stephen: That's where the speed comes in from gra. 'cause you need like with, if it's gonna be that much bigger speed is gonna be the

Ian: That's right. That's right.

Stephen: for an hour and 50 minutes to get one answer. You wanna type something in chat, GPT or perplexity and they need to, we need that answer yesterday.

Right?

Ian: That's totally right. And, and if you think about it, you know, training is an offline exercise. Like there's no humans interacting with it. Obviously the. The researchers who are, are conducting the training run, they care a little bit about speed, but there's no interactivity to that process. But when you start to think about end users touching an application, like you want it quick, but let me give you another, another leg up in this, right?

Everyone's talking about agents and agents. You know, I I had the CEO of a company called Crew AI at my sales kickoff last week. His name's Joe Mora, incredible leader, has built a fantastic company. And at some point I said, Joe, like, how big are you guys? You know, talk to my team about how, how we should work together.

And he goes, well, you know, we now employ 150 agents and 30 humans,

Stephen: Wow.

Ian: right? So he talks about the agents, you know, in the, in the same context as the people in the company because they've got a level of autonomy and authority to actually go take action and do work. Inside the business. And if you just think about that ratio at Crew ai, you know, there's kind of five agents for every, every human, right?

So now instead of running a team of people, you're running a team of the, you know, semi-autonomous software processes. So you're the marketing leader. You know, you might have somebody that's working on you know, monitoring social media, and then you've got another agent that's working on, you know, content production and a third agent that's working on content.

Promotion and maybe somebody that's monitoring like your promotional ad spend, right? So, you know, we used to think about those as human roles. Now think about them as agent roles. And in reality it's probably not a single agent. It's actually like tens of agents, what the crew team calls a crew of agents who kind of interoperate together to solve like a domain specific problem.

And in that world, the speed starts to compound. It, it's not just like you or I looking at a screen and kind of ty typing back and forth. It's, it's suddenly like, you know, tens or hundreds or thousands of these autonomous processes that are interoperating with each other, and they can go as fast as the computers can go.

And so inference speed in that world becomes even more important than what you just said.

Stephen: The pushback I feel with AI is people like, you know, but what about the eq? it's, you know, similar to smart contracts, they have to do exactly what you feed into them and

Ian: Yeah. Yeah.

Stephen: much room for creativity.

Ian: Yeah.

Stephen: are your thoughts on that? Because like. You know, it's great to have that human intellect and that, you know, maybe empathy, but it's also a lot of the process.

You just said like five to one humans, those five humans aren't gonna be coming to you about vacation days or come down sick. So like what are your thoughts about like, hey, this is the exact reason why we probably need more AI agents, versus like, Hey, they're gonna take jobs of people that are gonna add in that extra

Ian: Yeah.

Stephen: gonna make that product special.

Ian: Yeah. I, I, I would contest the the premise of the question. I actually think that you can have an AI interaction that is incredibly empathetic sympathetic, feels, you know, human in a lot of ways. And, and I think we're gonna continue improving on that, on that nexus, right? Like one of the, the most popular use cases right now is improving call center experience, right?

And so it's not about, oh, let me, let me turn off the human as quickly as possible. Like just give them their information in kind of a unfriendly way and, it, that's, that's not the case at all. It, it's actually how can we provide a higher level of service, a better customer experience a more, a more useful greater customer satisfaction outcome from the process.

And so, so I, I'm gonna challenge you a little bit there. I, I think the, I think the the machines can do a good job there.

Stephen: To your point, the whole, you know, the worst part of customer service is like how quickly you actually get to a human. That's the problem. It is like, we wanna talk to a human,

Ian: yeah,

Stephen: want to go through 17 prompts to get there

Ian: yeah,

Stephen: be on hold for 45 minutes. 'cause we're mad by the time the human gets on the phone.

Ian: totally. And, and you know, half the time that that individual isn't empowered to make a decision. That you want them to make anyway, right? Like, hey, I need a refund, or, Hey, I, I had a problem with the product I bought from you. Or, Hey you know, the, the thing you're charging me for is not actually what I ordered.

Like all of those questions, you get that level one agent and they're like, well, I actually can't resolve that problem. And so then you sit on hold longer to get to another person who escalates it to another person, like, well, we'll get back to you in a month. Like that entire process is, is pretty well known.

Like we've all experienced it with one of our providers or vendors and, and I think we have the opportunity to completely change that landscape of experience.

Stephen: Yeah, that's huge. I'm curious though, like,

Ian: yeah.

Stephen: technical hat. I don't know if there's a physics, you know, portion of this. It seems like Elon broke a lot of physics when he went through that, what he did in the US and built that huge training model. What keeps you from being 200 X faster or 300 x faster?

Like what's the constraint with the speed? Like

Ian: Yeah.

Stephen: could you get, or

Ian: Yeah,

Stephen: the path on becoming faster? Because I'm

Ian: yeah,

Stephen: when someone says they're a hundred times faster. It's like, well, I'm sure your clients are also asking you like, Hey, what gets me the 150

Ian: yeah. Yeah.

Stephen: rest of my competitors?

Ian: Well, you know, we're, we're building semiconductors and, and so it is a physics problem, right? You're dealing in, in speed of light and, you know, density that you can pack transistors on A chip. Crock has a super exciting hardware roadmap. We're gonna be introducing you know, both new chips and repackaging of, increasing density. Like there's a ton of stuff on our roadmap where, you know, the, the type of speed ups that you're talking about are coming very soon. So the performance story only gets better. The economics of the offering only get better over time. The more to come on that later in the year.

Stephen: I love it. You know, we have a lot of payment tech founders,

Ian: Yeah.

Stephen: to this podcast. A lot of them, obviously similar to how we thought about crypto, are

Ian: Yeah.

Stephen: about ai. You seem like from the, you know, last time we talked maybe in December to now, like you're well versed, you're talking about light and density, you're talking about semiconductors.

Ian: Yeah.

Stephen: was your path to learning about ai? Was it

Ian: Yeah,

Stephen: white papers? Was it

Ian: yeah,

Stephen: books? Was it watching videos? Was it just listening to other people talk? How, what was your learning path like, and maybe we can steal some of your,

Ian: Well, number one, just go start using it. Like, don't worry about the words, like, oh, I don't know what inference is, or, I don't understand how training works. Like I. Doesn't matter. You don't actually need to know any of that. So don't be intimidated by, by the jargon or the underlying technology. Like pick a thing that you're doing and it can be trivial.

My wife and I are taking the kids on spring bake vacation coming up and the place we're going, you know, we get sort of a, a menu. We can kind of build our own menu for the, the time we're there for like breakfast, lunch, and dinner. I was like, I don't have time to sit down and figure out, you know, four and a half days worth of meals.

And so I just took the menu and, and I, I sent it to one of these LLMs and I said, Hey, here's what we like to eat. Here are the options. You know, build me a four and a half day menu. Leave out one dinner. 'cause we'll go out and you know, one breakfast is Easter breakfast, so it needs to be pancakes. And like 10 seconds later it spit back a menu.

Stephen: I love that. I love

Ian: Right. And, and so stuff like that is how you learn this stuff. You start to see where it's good, where it's bad. And I would say, you know, sign up for a bunch of the services. Most of 'em have good free tiers, but go around, you know, pick the problem that you're interested in. I had another one recently where we, we were attending a conference and we got a list of all the attendees.

It was like 2,500 people. And, and we wanted to be able to go filter it down and, you know, pick out the people that were most relevant for us to try and schedule meetings with. And then we needed to track had we set up a meeting, you know, had we got a response to outreach, like kind of a, a flow chart of like where we were in the progression, but the list was sent to us as a PDF file

Stephen: Right,

Ian: and it, it had some really weird formatting in it.

So it wasn't like you just copy paste and drop it.

Stephen: The L might be an I or you're calling people by different name.

Ian: it was, it was a total mess. And, and I, you know, I don't write code for a living, right? I, you know, as much as you call me technical, I don't, I don't feel that technical most days. But I was like, you know, what I need here is like a, a parser to actually run this. And I was like, you know, I'm sure there's like a library and Python that can just take a PDF parse all this data out, you know, and it'll, it'll give me a CSV file.

I can open in Google Docs. I was like, but I don't know how to write that code. You know, I went to Claude, described the problem, gave it the, the document, it didn't get it right on the first time, so I had to give it some more feedback. But at the end of it, I had a piece of Python code that I could execute and I got, you know, all the data that I wanted.

Stephen: Yeah, that's crazy.

Ian: Right. Yeah.

Stephen: it. It doesn't, it might not get the answer the first time. So it actually forces you to think a little bit more critically, which only makes you smarter as a human being. 'cause now you're like, oh, I got to stage one.

Ian: Yeah.

Stephen: get to stage two? And it's like that critical thinking partner now that you have

Ian: Yeah,

Stephen: you didn't have before.

'cause if you couldn't get past stage one, that was it. Right? You

Ian: that's,

Stephen: laptop and call it the night.

Ian: that's totally right. And, and I, that, that's maybe the biggest piece of advice I would give to people is like, when you pick a problem and you go try and solve it, using one of these tools, there's a good chance that you don't get the output you want on the first try, or even the second or third try and.

I've seen people kind of dismiss AI and be like, ah, it doesn't, you know, it's a toy, it'll never replace me, therefore I don't need to engage with it. And I think that's the wrong approach. Like, I, I would actually suggest that you think about it as an intern, like if you've ever had an intern work for, you know, somebody who's you know, totally green, probably still in college, maybe even a high school intern.

If you think about the LLM as being at that level of expertise now when you ask it to do things. You know, provide the same level of granularity and specificity that you would to an intern who's never done the task before. And if you operate in that way and you get like, oh yeah, I got back like 70% of what I was looking for here, like, that's pretty good for an intern, right?

On a task they've never accomplished, you know, now like refine the request in such a way that, you know, you move it from 70 to like 85 or 90%. Then if you start to think about it, like how long would it have taken you on your own to get to like 90% good on the task, and how quickly can you now do the last 10% on your own?

And you've probably saved, you know, hours, maybe days worth of work. So that to me, that's the best way to learn. Go get hands-on, actually try and use this in some real day-to-day tasks. The second thing I would say is like, there are a ton of really smart people out there. Who, if you want to get into the technology, you wanna understand, like, you know, when Ian says a training run, like what does training actually mean?

Like, how does that work? Or you've, you've heard about like transformer architectures, like what is a transformer? Like all this stuff is accessible. You know, there's some great people to follow. Like, I would say Andre Carpathy. Is probably one of the, the deepest technologists who has an incredible way of explaining this tech to people.

So if you Google him, you know, he's put out recently a couple, three hour video series that takes you through all of this stuff. And, and so if, if that's what you're interested in, like, you know, go look him up. Start there. The other person I would mention is Andrew ing with ng. Andrew has built free training courses.

He's built a whole academy that start, you know, you can start at a very high level and go as deep into the technology if you, as you want. I would go follow those two people and just start consuming that content. And I, I promise, in a few weeks you will be you'll be probably above. You know, 70, 80% of the general population on this stuff.

Stephen: I love that. And you know, nobody's been listening to these people for like six years. So like they're dying to talk to somebody now that they have a, a captive audience. 'cause for six years, you know, they've been building and hustling and

Ian: Totally.

Stephen: teaching and there's been nobody in the classroom really.

So

Ian: Yeah. Yeah,

Stephen: you talk to me a little bit about like any regulatory requirements? Obviously you've been at chain

Ian: yeah.

Stephen: regulatory landscape with blockchain, we hear about like the safety and guardrails of ai. We

Ian: Yeah.

Stephen: to spool outta control. What, if any, guardrails or any requirements that you're thinking about as you look at GR.

Ian: Y you know, I am I, I'm a little bit skeptical of the people who are running around claiming that AI is gonna take over the world. I, I've always been pro technology. I think every time we've had a major kind of technological. Improvement or shift. We've seen major increases in quality of life and productivity.

And I, I look at AI the same way. I think there is massive upside and limited downside. So anytime I hear the folks advocating for like, increased AI safety or like, let's not let the, you know, companies train models at an advanced pace or like, we need to limit the number of chips that are available.

Like, all of that to me feels like total nonsense. It, it's the equivalent of saying, no, no, no, we can only manufacture 10 cars a year because the you know, we're worried about the, the horse industry. It just, I'm sure people made the same arguments a hundred years ago, but it, it makes no sense to me, so I'm,

Stephen: the internet came, they were like, oh my God, what's gonna happen to everyone that's making paper? Nobody's gonna use write letters or use documents anymore,

Ian: yeah.

Stephen: office is gonna close

Ian: Yeah.

Stephen: I remember that. I was there for

Ian: Yeah, it, and look, things are gonna change. I'm not saying that everything's gonna stay the same. I'm not saying that people's jobs won't be affected. I'm not saying that there's, there's not gonna be some temporary or even midterm kind of downside for certain, certain places or cases. But the net effect for society is gonna be major increase in productivity.

Which I think will unlock an incredible amount of opportunity and wealth for society at large. Just the, the access to knowledge and information instantaneously for everyone who's connected to the internet around the world, which is now, you know, most people is such an opportunity that, that alone, if we just stopped with education and said none of the, like, you know, knowledge worker productivity tasks.

I would say that AI is totally worth it. Every dollar that's been s spent to this point is totally worth it. So I, I'm, I'm all about as fast as possible you know, not, not slow us down.

Stephen: We have to remember our kids are gonna grow up in this. Just like, you know, we have those crypto native

Ian: Yeah.

Stephen: know, the social media native kid. Like our kids are only gonna grow up knowing that they can get their answers quickly, fast. And they

Ian: Yeah.

Stephen: think about what the next question's gonna be.

But you talked, you know, interesting enough about consumption. What are your thoughts? 'cause you know, as soon as Bitcoin came on the scene, there's people protesting on how much energy is going into Bitcoin. Seems like AI is consuming a

Ian: Way more energy.

Stephen: minting block, than minting blocks on the blockchain.

So what are your thoughts? I don't hear any like protestors about the actual consumption as much is, do you have any thoughts?

Ian: people are talking about it. It's not as loud because I think the, the difference between crypto and ai. Is that there's a more obvious use case, applicability, productivity for the average person, right? Like my wife is using AI to, to plan our vacation, as I was mentioning earlier. So for her it's like, oh, this is great.

Whereas crypto, she was always a skeptic. She's like, I don't understand why this exists. You know, magic money on the internet is not like a thing that I have a use for in my life. So I think when you, when you have that skepticism toward a particular technology, you're inevitably gonna look for the downside.

In the case of Bitcoin, power consumption is kind of an obvious one to, to pick on. In the case of ai, I just think there's, there's fewer people that are really focused on the downside 'cause it's so useful, it's so obvious. Like why we want this thing to exist, that you're, you're actually now, rather than concerned about power consumption, you're concerned about availability of power.

You're like, we need to fire back up the nuclear reactors. Build more. Build baby, build.

Stephen: Right, not use less, find more, I think. Yeah. Where it's like, Hey, we're using what energy to mint a a, a cartoon camel with a, with a cigar like that doesn't have the same effect as like, Hey, this person just helped me do my job

Ian: Yeah, exactly.

Stephen: more time with my kids.

Ian: Exactly. Yeah.

Stephen: are you looking forward to, whether it's at gr, what are you seeing?

Is there anything that like completely blew your mind? Like, Hey, I didn't even think about AI for this,

Ian: Yeah. Oh man. There's so much stuff. I, the future is bright here. It, it feels like every week we're seeing new models be released. You know, just in the last couple months people probably heard about deep seek where this kind of, you know, relatively unknown Chinese lab that was started by a quant trading firm.

Built a state-of-the-art, you know, kinda world beating model for a fraction of the spend that we've seen from some of the US labs. And I think they kind of prove to the world that, you know, this race isn't over. It's not won by two or three American companies. This is a global phenomenon that that really anyone can participate in.

And so, you know, the, the. Capabilities that are coming out in every dimension. Microsoft put out a model the other day that is able to generate video game scenes kind of live from taking samples of, of games, so entirely new levels being created kind of dynamically. You've got, you know, video companies building videos that I can't tell the difference between a, you know, professionally shop movie.

And and something that was generated from a text prompt. And so all of these technologies are just at the beginning, right? We're just figuring out how they work and how to apply them. And so as I look forward over the next year, like all of these things are just gonna keep advancing at such a speed that you know, the world is bright.

It's gonna be a lot of fun.

Stephen: I love it. And you're posting more actively on LinkedIn. We see you on Twitter. We'll make sure to include those in the notes 'cause it's fascinating to watch you going through this journey.

Ian: Yeah.

Stephen: you help so many people go through the same journey in blockchain and we're waiting for your podcast to come out with ai.

I'll be, I'll be devastated if it's not with me. But I'll understand too, 'cause you guys are going through the moon. Ian Andrews, thank you for appearing on the Around The Coin podcast and giving us practical things that we could take away and get into AI today.

Ian: Stephen, it was a pleasure. We'll talk soon.

Stephen: Awesome.

‍