Christian Stano

Christian Stano. C-H-R-I-S-T-I-A-N.

I am the Field CTO at Anyscale.

Anyscale is the company behind the open source framework Ray, which powers a lot of the data processing, training, and inference. When you go and look at some of these large models and foundation labs that are all the hype these days, a lot of them are powered by Ray. Anyscale is the enterprise platform around it. So we work with some of the largest, biggest, baddest foundation model labs in the world to help them achieve essentially the next frontier of intelligence.

When you want to build these types of AI pipelines, you have to think about the compute layer, the data layer, the developer interface, and then, as you scale this across organizations and teams, you also have to think about how you actually operationalize it beyond just standing it up and letting it rip. When you're talking about these ecosystems and environments that are billions of dollars worth of capex investment, all of that is amplified tenfold. Ray really simplifies that equation by giving developers an interface into those giant compute pools, to build out the pipelines that power what they want: I want to take all my data, I want to train this huge model on it, and I want to actually offer it as a business.

Ray has been around for about 10 years. Anyscale, the company, has been around for about six. We were founded out of the same lab as Databricks, which a lot of people are familiar with. Ion Stoica actually sits behind me in our office, which I think is pretty fun. There's a very similar ethos: take this open source, improve the world with it, and Anyscale can capture a small amount of value from that.

Great question. We help blend between the infrastructure layer and where the developers sit. The way we like to describe it, there are five common workload patterns that fall into those three buckets of data processing, training, and inference that a lot of developers are trying to achieve today. Then they have this infrastructure layer, which is all the compute, the CPUs, the GPUs, whether you're running on cloud or on prem. But between those two layers is a massive gap that requires orchestration, scheduling, all those kinds of things I just talked about. That's where Anyscale sits.

I like to boil it down to three things. One is speed. If you want to build this yourself, whether it's open source or alternative frameworks, we typically see customers spend anywhere from 12 to 18 months just to stand this stuff up, and then another six to nine months to actually operationalize it. So in a world where we're measuring things by the weeks, you're looking at almost a two-year runway to get value out of the massive capex investment you just made for GPUs. We've dropped that down to about a month, so by month two you're shipping your production inferences and models. That's value prop one. Number two is the overall cost of the system: how many people do you need to maintain it, how efficiently can these systems run, how can I share compute across the different teams I'm working with? We help simplify that to run basically better, faster, cheaper. And the third thing is for the developers themselves. Something I always strove for when I ran the AI teams at Attentive is customer satisfaction. CSAT is a lesser-talked-about metric when we're talking about GPUs and capex and all these dollar figures, but are my developers happy, do they like the tool they're using? Anyscale has invested a lot of time and effort into the interfaces we provide developers to make their iteration and experimentation cycles much faster, much easier, and much more of a good experience.

Exactly. One of the fun things when I moved here: down the 101 there are billboards about AI, AI, AI. In the last couple of months, if you've been driving down the 101, it's all talking about own your intelligence, stop renting, own your intelligence. So what does that actually mean, and how do you actually get there? That is the question we're starting to answer. When you have this massive pool of compute and you want to run this application layer, we help you bridge that gap to actually own that intelligence layer and then action on it.

We provide a part of that. Usually when our customers think about using Anyscale, they're on this maturity curve: I've been writing prompts, I've been using API providers, and that's given me some ROI on my business. So I've proven there's business value in going with these LLMs and AI providers. What comes next is, I want to predictably control how that cost scales. We don't just provide the answer to the cost problem, we provide the actual systems that let you control that cost.

We have customers in every vertical, from AI-native foundation model labs all the way up to banks and Fortune 500 companies, and everything in between. Where we see the most success is in physical AI: robotics, ADAS, autonomous vehicles. Torc Robotics is a huge customer of ours, doing autonomous trucking. I was just on a panel last week with them talking about the incredible things they're doing to operationalize autonomous trucking at scale, all powered by Anyscale. And companies like Spotify and Apple, who are really large open source Ray users and power a lot of their tooling on the Ray frameworks as well.

There is a large portion of value that unlocks. We also see a lot of value in the traditional digital-native company that's running large-scale CPU processing. My role before Anyscale was head of infrastructure at a company called Attentive. We were the company that, when you get a text on your phone saying hey Josh, come buy these shoes because we're recommending them to you, we were powering all the personalization behind that. That was primarily CPU-driven. So even at the quote-unquote lesser scale of investment on the CPU side, customers still see value. I would say the biggest value is when you start thinking about massive parallel scale. That's where it really gets unlocked.

I love this question. I'll start with what a field CTO actually does, because when I talk with customers, and even with people within the company, it's like, okay, this is a cool title, what does it actually do?

Exactly. Even my family, when I used to talk about working at Attentive, the SMS company, they're like, oh, that makes sense. I say I work at Anyscale, there's a thing called Ray, it's massive-scale GPU, and their eyes glaze over. So a field CTO essentially has three roles. The first: I'm a strategic advisor to our largest customers, usually operating within the engineering leadership circle. I take my experience running these types of teams and sitting in their seat, and apply those patterns to make sure they're successful. The second role is that I influence a lot of our product strategy. Taking those two together, it's about making sure we're building the right things for the right customers, so we're providing continuous value, especially as things change really frequently. The third is, as the world we live in moves increasingly faster, how do we make sure we're putting out the thought leadership around where Ray wins, where Anyscale wins, where we see this market going as we move from a rent-to-own ecosystem? My route here is very non-traditional. I was head of infrastructure at Attentive. I joined and started the platform organization from scratch. There was one engineer who joined the team, and from there we built the team up, built the platform up, and were able to achieve real-time personalization in about six to eight months, which is pretty unheard of in the space.

Before that, I spent a good deal of time when I was in DC working in and around govtech. I stood up one of the first AI platforms at the Department of Defense, a big initiative called the Joint Artificial Intelligence Center. Back in the day it was a big competitor with Palantir. That was a lot of fun, like six or seven years ago.

From there I've taken on different parts within the infrastructure and systems domain, all of which have given me a ton of different experience that I can apply as a field CTO, not just in the AI space but in security, compliance, regulated environments, systems architecture, and also building and running an organization.

I love being customer-facing. When I was an internal engineering leader, I had so much fun building the team, building the system, seeing the output. But for me personally, when I have the opportunity to work with some of the best customers in the world pushing the frontier of AI every single day, with a framework like Ray that as of next year will power about 50 billion dollars of compute capex, that's an opportunity that's unheard of. As I thought about where I want to go next and this opportunity presented itself, it was a fantastic chance to take everything I've learned and grown into from my past experiences, bring those patterns to the customers we work with, and also explore this entirely new space of AI infrastructure and these intelligence loops that are getting paved every single day.

I do like people. I won't say this is my own quote, but I like to say I'm an outside cat, so I like to socialize.

I've been here about two and a half years.

Growing up in Cleveland was such a great childhood, a small city, everyone's close. One of the exciting things about San Francisco, outside of the weather, is you can just feel the energy in the city. I know during COVID it took a bit of a downturn, but just walking here from my office you can hear people talking about AI, about ROI and business value and the next big thing, billboards everywhere talking about the next frontiers. Working in the space, it's just such a cool place to be at this point in time.

It's usually hit or miss, it depends on if they've used AI before. One of the things about San Francisco is that we live in a bit of a bubble. AI adoption is everywhere here, everyone uses AI, but if you go outside the bubble there's less usage. When I tell people who have used AI before, and I can equate it, like, hey, the ChatGPT thing, we help companies build things like that, they're like, oh, that makes total sense, that's really cool. The overall sentiment is always interesting, because not everyone has as much of an in-your-face AI presence in places like Cleveland. There's an overall optimism that I hear, and there's overall curiosity. A lot of people are curious about what this actually does and means beyond just a cool chat where I can ask what the dress code means for the upcoming wedding I'm going to.

That's a great question. What I'm seeing, as I talk to people both here in San Francisco and in places like Cleveland, of course there's that hesitation and anxiety around, is AI going to take our jobs? I like to think back to when Microsoft Word came out. Was there an implosion of the economy? That was a pretty revolutionary thing, you could type a document and share it around. AI is at a totally different scale and surface area, so it's very hard to say what the ultimate outcome will be. But the way I like to think about it, from the position of Anyscale when we're helping these companies, is there's going to be so much economic opportunity created by AI, both to build the AI, run the AI, and maintain it. There are going to be new segments of jobs created by AI. One of my best friends does AI enablement now for family offices. So amid that anxiety, I think there's a light at the end of the tunnel. Am I naive that there's probably going to be some impact and fallout from AI? No, I think there will be. But I do think the next step is the economic opportunity of our lifetime, as this creates new and different types of jobs in a future we probably can't even imagine today.

I usually like to center around, imagine all the things you go into work and say, I hate doing that, I don't want to do this, crunching through spreadsheets, whatever it may be. Imagine if you had an unlimited assistant that could do that for you and free you up to do the higher-leverage, higher-impact things for your business. For a small business, it could be something like, rather than running inventory, maybe you have an AI system running inventory and you can focus on marketing, or the thing that energizes you about that business. It all comes down to that energizing piece. If we can use AI to unlock that energy again for people, there's a huge amount of value people can bring into their businesses and the way they approach their day-to-day life.

Exactly.

There's some truth to that. In my opinion, and I'm obviously biased because it's the seat I'm in, the field CTO will be one of the most important positions of the next couple of years, for a couple of reasons. One, as we look at AI creating more and more leverage, with individual contributors and specific people creating widespread impact within companies and for their customers, having someone in a field CTO seat who can essentially 10x or 100x their impact just by nature of that is something really incredible. I also think that when it comes to the forward deployed engineering motions, that's also, whatever you want to call it, rebranded professional services. There's a lot of hype, unhype, anti-patterns and patterns around it, and I can dig into that if it's interesting. But the concept of the forward deployed engineer, basically saying, hey, you have a business objective, a business outcome, let me help you bridge that gap with our software and our expertise, it's hugely valid. I'm seeing it every single day with companies we work with. We help robotics companies raise their next round of funding by helping them build a VLA model. There are really cool use cases like that of the forward deployed engineer that I think people miss when they say it's just rebranded professional services.

It's actually pretty interesting, because we're seeing very good efficiency of these resources, especially if they're experts in their field. For my seat, my goal is to embed with the executive teams and say, here's how you should think about this problem, here's the strategy I've found that works, and here are the things to avoid. And by the way, I have a forward deployed engineer who can come in and make this happen for you if you want to move quickly, get to that outcome faster, and supplement your team with someone who's done this a hundred times before. That's more of my role, rather than orchestrating across all these different implementations.

It varies. We have forward deployed engineers who that's all they do. We have some folks who are experts in their specific industry, so they wear multiple hats. But we're talking in the range of 20 to 30 engineers going out there, working with customers every single day to help them achieve their next business outcome.

It really depends on the engagement. We're globally distributed as a company, and it's very hard to get on a plane to Asia for some of the Asian customers we have every single week, especially if you're based here in the US, so it's a mix. I find there's a huge amount of value to being in person. A lot of our forward deployed engineering engagements will fly into the customer's office, sit with their engineers, and basically run a two-week sprint: between now and then, let's keep cycling through it and identify what we actually need to build. We also have forward deployed engineering engagements where it's very clear, we don't need to fly in, we'll just get the work done and present it as a deliverable. So there's a spectrum between those two.

We are globally distributed across the US, primarily in our field and go-to-market organizations. We have an EMEA presence and an APAC presence. By nature of every business in the AI world these days, there's not a geographical boundary on who needs support on these AI systems, so we've aligned our teams and our locations with them.

This is a great question, because our head of engineering and our CTO play an extremely important role, and we have very different roles within the company. The CTO and head of engineering, their main focus is, am I internally executing on the things against our product vision? That's the number one goal: are we building the things we need to build to put a product out in the market? My role as field CTO is on the inputs and outputs of that. On the input side, making sure I'm talking to all of our customers and prospects and understanding their needs, where their biggest problems and pains are, and turning that into a product roadmap. On the output side, as our product gets released into the market, my goal is to make sure our customers are successful with it. So the way I describe it: I'm on the inputs and output side, and the traditional CTO and head of engineering are in the middle to actually make the things happen and execute.

I always like to think in the outside-cat metaphor. A lot of engineers like to focus on the engineering problems, and they're de-energized by the sticky, soft questions it takes to relate that to a customer. So I like to say I'm both a translator and a storyteller: make sure we're telling the right story, the right narrative. What does our product do, what value does it bring, and what problems is it going to solve for you?

I see a couple of different flavors of this. There are a lot of engineers moving into that forward deployed engineering role if they're energized by those customer interactions, but I don't think it's a requirement. What I'm seeing more, from the internal side, is less of a retraining and more of an evolution of the role toward architecture-focused systems engineering and systems design. How do I orchestrate these systems and become more efficient with them to achieve in a month what would have taken three? So I think there are two flavors we're going to see engineers evolve into: the outside cats going toward that forward deployed engineering motion, and the people who want to stay on the execution and internal side, focusing on how to multiply themselves and get into orchestrating these fleets of agents, instead of getting into a coding IDE and crunching on a six-month roadmap themselves.

I absolutely think there's room for all the different shapes of engineers still. For me personally, I'm less energized by the internal execution, getting into an IDE and solving the really hard, detail-oriented problems. There's always going to be a space and a need for people like that, because it's a different skill set with different outputs. I can't be successful in my role if those people aren't successful in theirs, so there's always going to be a synergy between the two.

I think the baseline is a team member that's essentially an agent, someone who can operate as that layer underneath the team to support everything happening day to day. Then a product manager influencing what we're building and why, and then smaller, more condensed teams of three to five engineers usually rolling up to some type of leader. I'm seeing more of this pod style: take a really hard business problem we can execute on quickly, give it to a lean team supported by agents, and let them be successful with it. Sometimes there's a forward deployed engineer in that equation, working directly with a customer, a design partner or an existing customer, to help close the loop from the inputs to the outputs.

I'm really excited about the flywheel effect we're going to see from a lot of these frontier model companies. What I mean is, as we think about this maturity curve that a lot of companies outside the ones you hear about in the news every day are starting to get into, there are life-changing flywheels they're starting to create: in the life sciences, drug discovery, protein folding, autonomous vehicle safety systems that aren't in the news every single day but I think in the next one to two years will be. The things they're doing are like science fiction. I can't even believe some of the things these companies are doing, and the opportunity to work with them and support them from the Anyscale and Ray side, for me, is a life-changing opportunity.

Let's chat again in like six to eight months, and I'll have some.

Yeah, thanks so much, Josh.