Speaker 1In the SaaS world, it was like, yeah, if you had product market fit, it almost is like your business can scale. But in AI land, it's like the opposite. You can't scale if you're spending more on inference and you are generating revenue.
Speaker 2Just wrapped up a great conversation with Somya Tullo, SVP of sales at Fireworks, one of the fastest growing AI native companies out there, recently raising a Series D from a pretty stacked cap table at an $18 billion plus valuation. In the conversation, we talked through what this inference market is.
Speaker 1Inference, regardless of which company you're buying tokens from, is the oxygen of the modern enterprise. Every single workflow that we are all participating in every day is probably something that can be made better or faster or more meaningful with AI. And inference is the fuel that powers all of those things.
Speaker 2How she chose Fireworks for her next move in the AI native landscape.
Speaker 1I think there's some amazing companies in AI, but like they've reached. Escape velocity, whereas at this stage that I joined Fireworks, there's like 20 people in GTM, like every single person can have an impact.
Speaker 2This fire acronym that they use for different inputs that salespeople at Fireworks use in order to be successful.
Speaker 1And we came up with our fire way of selling, which is.
Speaker 2Welcome to the crew podcast, where we have conversations with the top go to market leaders at AI natives and SaaS startups. I'm Chris Balestris, co-founder of crew, a go-to market recruiting media and investing firm helping seed through series D AI natives grow. Now onto the episode. Samia, thanks so much for coming on. I feel like I've known a Fireworks for a few years now. When I saw you joined, we helped place Lance and like one of the A's over there. And I just feel like inference is having this moment right now. It's like everything that I've heard of the last week. So if you want to think of it like chat, GBT launches in 2023, and then the coding agent starts. So it was like cursor was all the rage. And then the last like six months where my AE friends are going to CTOs that I'm talking to, everybody's talking about inference and I'm not super technical. So maybe you can help explain to me like what has happened in the market that companies like Fireworks have just completely taken off. Like what's been that chasm that kind of got crossed?
Speaker 1That is a great question. Inference, regardless of which company you're buying tokens from. Is the oxygen of the modern enterprise. I think what's happened really in the last year or 18 months is that the average company and the average enterprise has started to attach AI to every part of their business. It starts with demand. Developers and their productivity. It then goes to customer support. It then becomes something that finance uses like wildfire. Our finance team is one of the largest token consumers at our company. And as that has happened, I think every CTO and every CFO is saying, hey, where am I purchasing and how am I powering all of this inference? And that's led to a huge moment for open weight models. And so on the one hand, you have the demand of these customers who are using truly like infinite amounts of tokens. Their demand for tokens is insatiable. And on the other hand, you have these open weight models that have gotten really good. You mentioned when ChatGPT launched, like I think DeepSeek launching, the first DeepSeek was like a huge aha for people. It was in the Wall Street Journal. I didn't know much about inference at the time and I read about it. And in the last quarter, we've had truly state of the art models launching from the open weights ecosystem that are Opus level or better, Astra level or better, especially when you'd post train them to do cool stuff. I think this moment where the supply of exceptions. Models that can like do frontier level work are available in the market at a time where every CTO and executive buyer is saying, I need to control and own more of my AI is exactly why inference is exploding as a category.
Speaker 2Yeah. I think everyone throws out inference. And then if you were to quiz them on, hey, give me like two minute talk track of what that actually is and explain it to like you would to your grandma. A lot of people wouldn't be able to do that. Tell me of how all this interacts with itself. So there's open weight models. There's open weight models. There's open source versus using chat or open AI and anthropics APIs. I kind of get broadly why this all makes sense, but okay, you bring it in house. So inference sits on top of it. Can you just explain to me like on 10, what even inference is, where it sits in the stack and why the emergence of open source models is an accelerant for your guy's business?
Speaker 1Yes, that is a great question. So inference is really the process of a prompt going into an LLM. The LLM basically, you know, it's a process of a prompt going into an LLM. It's really decoding that prompt and giving you an answer. So it's really the process of interacting with these LLM models in real time. It's really the runtime for those models. What our company does and how inference sits sort of in your AI stack is to actually run any kind of LLM workload. You need a few things to be true. First, you need compute GPUs that are powering all of these workloads that are running very, very quickly and fast. Then you need the model that's like. Sitting in that compute and then you need a harness on top of it. The harness is the technology that's telling the model how to behave in your environment. And what fireworks provides is we are really an all in one solution to manage the models and the compute that you're going to be using in your workflow or in your application as a company. And so if you think of one of our big examples, you know, we have helped train the model tenant. That's Harvey's latest model that'll be in there. New Harvey to platform. And when Harvey is working with us, we're providing compute that helps power their application. We're helping provide access to all of these open weight models. One of the things that's really special about open weight models is you can control the weights of those models. And so you can take data and take computing techniques to make them really good for your use case, which Harvey's really trying to be exceptional at legal. And then we help serve it in their harness or their application.
Speaker 2Yeah, I love that. I love like the use cases of, yeah, Harvey or like what Rogo is doing. For finance or basis is doing inferences, the fuel, you all help in the compute and model part of this. Is it because CTO, everybody was kind of Yolo in the first year or two, and then they realized, okay, what's the ROI here? Spend is getting pretty outrageous. Also, these open source models have caught up to the closed sourced ones. So because of this, it's been like, let's use fireworks as an inference partner to be able to, we still need to keep going with our acceleration on using AI. But we. We just need to like have some sort of a cost layer, cost like manager that helps with this. Is that right?
Speaker 1Yeah, that's a great frame of it. I would say there's two big things that our buyers are talking to us about. One is cost. If you look at Anthropic and OpenAI and maybe Cursor in there, that's what, $300 billion plus of revenue that came out of the CTO's pocket in the last two or three years. So they're definitely noticing. But I think actually the bigger thing that. Most companies come to us to talk to us about is they care a lot about their data. They care a lot about specializing. And, you know, the purpose of the Frontier Labs and their stated mission is creating some kind of like AGI super intelligence, like this model and like this ecosystem that's exceptional at all things in the world and like highly generalized. But if you go think of like any random bank, you know, what they care about is having probably like hundreds of agents, if not thousands of agents. That are specialized to all the different things they do. An agent for coding, an agent for risk underwriting, an agent for KYC, an agent for customer support. And our bet is that specialized intelligence, which is models that are really specifically designed and trained for each of those use cases. And then having a collection of those models in your ecosystem is going to be more impactful in the long run for your business, both for cost reasons, but also for getting the most out of your data and being the best for your customers. And so that's really. Like our vision of the world. And I think most of the companies that come to talk to us, it's partially about cost. I think cost is the thing that maybe makes them like realize like, oh, shoot, I didn't like getting that bill. But then they think about it a little deeper. And I think owning your AI stack and having complete control over it is is the bigger and deeper business value driver.
Speaker 2Yeah, I agree with you that like it's like the apps that are going to specialize in certain verticals or use cases. There's going to be a massive market for that. Like we're working with a bunch of these like Harvey for X, for example. Yeah. But then there's a lot of. We see this on the recruiting side where they're really worried. Let's say they can't get into fireworks. They're not technical enough, for example. They then say, well, if I'm an app seller, is this app just going to get taken away by Anthropic or OpenAI? But then there's arguments like yours that no, they're building specialized intelligence for use cases. How would you think through, let's say, in the audience and they're looking to join an application company, which ones have staying power versus maybe ones that don't? If you have a POV on that.
Speaker 1That's a great question. I would say, first of all, I trust. I mean, I'm totally AI pilled and I truly believe we're in the very first early innings of this market and that this is going to be one of the biggest categories ever. When I interviewed for this job, I was really fortunate to meet like Sonia, who's on our board from Sequoia, who's incredible. And Eric, who's on our board from Benchmark, who's also on the board from Cerebrus.
Speaker 2I guess I was that cap table.
Speaker 1Yeah. We're so fortunate. And, you know, I had some of these kind of questions. How can we go compete with OpenAI or Anthropic or in their space? What? I learned from those conversations I would encourage anyone who's listening to think about is this is a massive, massive, massive category. And it's just important to be in it. And it's too big for any one or any two companies to just go win everything. There's going to be a lot of room for applications, for infrastructure, for consumer products, for business products. And every company I've been a part of, you might ask, hey, you're competing with a hyperscaler. Why would you do it? Or you're competing with someone. Why would you do it? And I think that's like the wrong question. Yeah, there's always going to be some big company in the space that's doing their version of the thing. But if you're passionate about an idea and you have a compelling story for how it changes that space, you know, I think that's like the bigger thing to really care about.
Speaker 2Shout out to a sponsor of the crew podcast, MetaView. Hiring is the highest leverage thing any leader does. As they say, the team you build is the company you build. That's why over 7,000 companies from startups to the Fortune 500 hire on MetaView. MetaView gives hiring teams AI agents that do the recruiting work for them. Finding candidates, reviewing every application, capturing every interview, all in one shared workspace, so teams make faster, better hiring decisions. Don't let your competitors out-hire you. MetaView customers close roles 30% faster. Check them out today at metaview.ai. When I looked at your background, you played golf in college. I was very jealous of it. I grew up playing baseball. I'm like, talking about a waste of a sport that made me worse at golf because of the way that I naturally swing. But then you were in finance for two years at Deutsche Bank, and then you started in biz ops roles. And usually, like, I want more finance and consulting, and business ops. I'm a former consultant myself. And it's like, to make the flip into sales, what was even the chasm that you crossed? Was it, I think it was at Twilio, where you then said yes to sales. And like, why didn't it take you, why didn't you do that earlier in your career? And tell us more about when you said yes to sales.
Speaker 1Yeah, that's a great question. So when did I say yes to sales? When I joined Twilio, we were about $89 million in revenue. And we were very proud of the fact that we had no sales. When I joined, I worked with Jeff and Lee and Roy on our Series E pitch. And we were telling investors like, hey, guys, like this thing sells itself. People just come to our website. And how great is that? And it was really great for a while. But we got to a place in our evolution as a company where it was very clear that to grow from several hundred million to several billion in revenue, we needed to go sell to the likes of United Airlines and Walmart and Bank of America and, you know, the names. And George, who's actually now our president at Fireworks, had come on to Twilio's exec team to help us really navigate that enterprise transition. And I was like, oh, my gosh, I'm so excited. And I was like, oh, my gosh, and we took a big bet on creating what we call the sales surge. We hired 100 salespeople in a year. It was an awesome experience. I got to work really closely with him and Mark, who became our CRO, on developing our coverage models and the board presentation and the economic justification to do all of this hiring. I found it really interesting and I found it really fun. And Mark at some point was like, Somya, you know what? There's two things that really matter at this company, sales and product. And I think you should go do one of those things. You know, BizOps is kind of about making slides and making recommendations, but not really about carrying the number. And then I told him, yeah, I'll come be in sales. And he was like, OK, well, you don't know anything about sales, so you could be at AE. And I was like, OK. And it was really humbling, honestly. But I went and I worked as an AE in our strategic accounts team. I turned around some really big and technical accounts that we were really struggling with. I learned so much. I had so much fun. And I really never looked back from
Speaker 2there. Yeah. The quote that you said, you know, being proud going into the Series A, it reminds me of like what Slack did. Yeah. They were like proud to not have salespeople. Yes. For almost too long and ended up teams that are catching up, you know, because of it. I love hearing those stories of when people flip, especially ones that went to great colleges, went more traditional finance background than to BizOps. We need more of those people. I didn't even know
Speaker 1like sales was a real career track, honestly, when I graduated college. I went to a small school in the Northeast and it was like everyone wanted to be a banker or a consultant for whatever reason. And it was I went to college during the mortgage recession. So actually really dumb when you think about it. But that's where I started. And I didn't. Yeah, I wasn't made for that world for long.
Speaker 2Yeah. Yeah. I felt the same way with my short stint. You had a great run at Twilio 2014 to 2021, about seven years. You had some stints at some other companies, actually Zillow, which I thought was just I feel like I'm on that every day. It's an amazing platform, especially as I was buying my first house last year. The Zillow estimate can really get you. It's such a unique algo. How did you choose fireworks? What do you tell people? Let's actually flip it now to say that they're technical enough to get into companies like fireworks. How should they think through? Do I go to an inference provider? Do I go to more of an app layer company? Do I go to like some of these search APIs? Like, how did you think through it and how would you advise people to think through company selection?
Speaker 1I think for me, you know, something that I think about a lot is something George told me like probably 10 years ago now, which is think a lot about the platform you're on. And the best way to grow your personal career is to be at a platform that's going to support a lot of personal growth and development. And the way I think about that, like, what does it mean to be part of a great platform is number one, I think you need to be in a company that's in a massive TAM. I really asked myself, like, can every person use this product? Can every company use this product? And therefore, does it have the runway to be in a very, very big space? I think personally that like exceptional technical founding teams win every time. I know that I'm not going to be probably like the CEO or founder of a company, but I love working as a partner to an exceptional technical founder. Lynn is amazing. She's one of the most amazing people I've probably like ever met in my life. And that's probably the number one reason I'm at Fireworks. And she's assembled in truly like next level technical team from places like the MetaPyTorch project and Vertex to come build this. And I think of the top, you know, I like great minds in the world, like maybe five of them work at our tiny company, which is so cool. And I knew I wanted to be in an environment with that type of a technical talent. And then the third is, you know, I care a lot about being able to personally have an impact on where the company is going. And it was really clear when I was interviewing at Fireworks that the go-to market was so nascent relative to the opportunity. There's some amazing companies in AI, but like they've reached escape velocity and like the every marginal person that joins, like they're helping the company. And I think that's really, really important. And I think that's really, really but like the company will survive without them. Whereas like at this stage that I joined Fireworks, there's like 20 people in GTM, like every single person can have an impact.
Speaker 2Like you do back the napkin, like revenue to AEs was like the biggest I've ever seen. Yeah. It's crazy.
Speaker 1So super, super awesome. And that's how I joined. And, you know, for me, I optimize entirely on the upside of the business. And I trust that my scope, my role, my team, all that stuff just like works itself out.
Speaker 2Yeah, that's great. So I think the TAM piece, I've chosen an opportunity that didn't have a big TAM before. And I remember when we were trying to recruit people, I knew the TAM wasn't super big. And then more senior reps when we were trying to recruit them, I'm like, oh, I never joined them. I'm like, wow, there's like one reason, like the TAM's way too small. I'm like, why couldn't I have known that? Right. And it's paradoxically, the best markets are super competitive because people are choosing to spend their times at larger markets that have a big TAM. And that's why when every time reps come to us and they say, oh, well, isn't that space super competitive? I go, it's usually like a green flag, if it is. I think that's spot on. Yeah. Okay. So you've been there for now, like six, seven months. Earlier this year, you joined. You talked about this FIRE acronym for like the sales inputs, one of them facing the place. I'll let you rattle it off, but the face in the place one really resonated with me. That was a KPI we used to use at Salesforce way, way back in my first sales job in tech. But tell us more about this FIRE acronym and like the things that are important that if you do X at fireworks, you believe that a rep would be successful.
Speaker 1We were starting to scale our sales team. We wanted to have a really crystal clear set of guidelines and expectations that firstly helped our team succeed, but more importantly, helped us differentiate our company in the eyes of our buyer, the way we show up as a team. And one of the things I think is really important in infrastructure that I think is actually really misunderstood is it can, like a sales cycle can go one of two ways, like a speeds and feeds conversation about like how much you cost and what your latency is, or it can be like a partnership conversation. And so we spent a lot of time of like, what are the sales behaviors that make a customer journey feel a lot more like a partnership? We came up with our FIRE way of selling. F is for face in the place. I think this is a really big one. I've spent my whole career selling to developers. They don't want to talk to you if they don't have to. My husband is a developer. He like walks into another room to go on a phone call because that's, that's just how it is. As we're really thinking about it, we're like, you know what, like we are not negotiating on Slack and like doing business online. We want our customers to know who we are as people. And we care a lot about showing up on site. And honestly, when we show up on site and we bring like our FE team, our field engineers, when we bring our product managers and help actually show the technical depth of our company in person, it helps us win. It helps our customers succeed. And it's a big driver of our success. So the F piece of the face in the place is super, super important. I come up here all the time to like walk around to every startup's office. I is industrialized. It's like running everything like a machine. It really comes down to how we're using AI in our organization to scale and to scale rapidly, to enable ourselves rapidly. We have brought in a lot of things from the traditional like playbook culture, like PG Tuesdays. That's also part of running like a machine. That's been really helpful to our results. R is rapid. If we're the first to get a customer to to production live and successful, we're going to win. And so I think being responsive and being quick about how we're moving, especially in this space, is really important. And the E is engineering excellence, which I think is actually like the most important one of all of those, which is we want to have the most technically adept sales team in the industry. Even as we scale, we're not lowering our bar for technical depth. And so you'll see that we're hiring folks who have been SEs before or have computer science degrees or demonstrate like just a genuine curiosity about technology in our space. And a huge part of our interview process is about testing for this element.
Speaker 2Big thanks to our sponsor, Attention. Sellers were hired to sell, not to fill in the CRM. Attention records and understands every sales call, then automatically updates your CRM. Sends follow-ups and scores reps against frameworks like Medic, Bant, or your own custom methodology. Next best actions are proactively generated from there. You get a clean pipeline, coaching on every call, and forecasting you can actually trust without asking reps for more admin. Turn your call recordings into a GTM operating layer with Attention. Try them today at attention.com. I want to dive into those. The face in the place thing resonates. I think you just break a barrier of trust in a good way. Yeah. When you meet somebody. Get to look them in the eye. I think that like it's just a human thing. Like the developers, if you were to ask them, yeah, they don't like sales. But like everybody, like if you're just nice to them and you're like can hold up in a conversation technically, like it's going to only help the relationship in my opinion. Tell me about like the R part. So like first to get to production, how does that work in a deal cycle? So I'm sure that they're evaluating other, hey, like the developers looking at two or three competitors. How do you think through that? The reason why I say this is I had on Liam Mulcahy at Parallel. And he said that. Sometimes they feel like they get boxed out with developers because they want them to let the pie bake in the oven without any intervention. Yep. And then they want to see which pie comes out the best made. And then that's how they make a decision. But how do you like balance working with a developer, not pissing them off, but also influencing a deal? Okay. Trying to get this R of trying to get it first when maybe they might be boxing you out of a specific process.
Speaker 1I have a rep, Chris. We call him Craig Head, who's amazing. He's like, my job is kind of like being the assistant coach. And a huge part of what I'm doing is I'm making a customer's deal go really quickly because I'm coordinating really important pieces very strategically and thoughtfully. And I would say when it comes to R and getting a customer live to production, this is what our field engineering team is supposed to do. And our reps need to be able to know enough to set our field engineering team up for success. For us, a well-scoped, well-defined proof of value is going to go really well. If it's not well-scoped, if we don't really know what the customer is solving for. If the requirements are changing on the fly, I think that's where you create opportunity for your competitor to come in and do something different. I was actually on a call with a really cool customer and he said something to me like towards the end of the call where he's like, well, I'm learning a lot just by the questions you're asking. And I think that to your point about developers and letting the pie bake, like if you can get a customer to say something like that to you, oh, you're asking me questions that actually are helping me progress like safely and soundly through my launch journey. You're going to earn it. You're going to earn the right to be part of their proof of value as opposed to leave me alone. I need to put credits in my account. I need to play with it and I need to come back to you later.
Speaker 2Yeah, I think the way that I would always frame this back, if I'm putting back on my Enterprise AE hat, I was mediocre Enterprise AE. But I would say like, hey, if I was in your shoes, this is how typically our customers evaluate something like this. And then they say, and sometimes they don't choose us because they care more about X, right? You purposely go away from your own interests. Yeah. And you throw that out there. But if they care about Y, then they will tend to look at our solution a little bit more. And like one's not right or wrong. Every company cares about different things. Yes. Your job is to help them make an informed decision. You might lose some of that. Then they open up. Then you realize, oh, wow, we have no shot at this. You just walk away. You work on the other deals that give a higher chance of closing. It's this whole like, how do you become a trusted advisor? It's a little bit harder with a developer. All right. So the E in the fire. Yeah. The hiring. You guys sell a super technical product. What's the technical floor? Is it? Like, hey, they need to have sold like an API before. Maybe done that for a year. Obviously, the ceiling and the thing you hope is the former software engineer or the former solution engineer that like had the same come to sales moment that you did. But how do you think through hiring technical talent and the things that you assess for maybe more of the high slope candidates versus the former engineer candidates?
Speaker 1So we do sell a really technical product. But I really believe that the biggest input to success is. Your curiosity and your intellectual horsepower in terms of developing technical acumen. I don't think you have to come in being an expert in matrix multiplication or fine tuning techniques. Well, well, I actually was a math major, so we could take it there. I don't think that'll be that fun for everyone. But matrix multiplication is what I basically see. That's the kind of thing you should say in an interview for fireworks that will probably hire you anyway. You know, I would say for us, we have one part of our interviews like circuit that is all about technical curiosity. And we tend to have people come in and really explain their own product. Why is it different than other products? What is an architecture level example of why your product versus someone else? So being able to dig into how the technology actually works versus sticking to like a true just business driver. So a good example is like if I was kind of explaining Twilio, I can go really deep into how Twilio Super Network works, how we were using machine routing algorithms to choose between underlying providers to get the best cost quality for customers, how we thought about managing throughput of messages against telecom networks, how we thought about all these different elements of building your actual application that ended up with a business outcome of customer engagement. And so those are the kind of things that we really test for. And, you know, if you've been able to go in depth in your prior company and after the two, three, five year stint you did there, you can tell us in depth and you come off as someone who is going to take that time and effort to go learn a hard thing. I think you're going to be able to be really successful at fireworks. And we have people who are successful on our team that are not from like the super technical companies. Honestly, I've had people from those really technical companies fail in our interview process. And I think that's where it comes down to, like, you can get by at a really successful technical company and not be technical yourself because the company is very successful. You're taking orders. You've got a lot of resources, whatever. That's just not how our company is right now. So a lot of our process is about what's your personal horsepower, curiosity, grit.
Speaker 2Let's say there was somebody listening who has traditionally been an app layer. Layer seller and they want to break out of the app layer and go into more of the technical layer. What are some things that they can show or demonstrate? You brought up a couple of them, but like peeling back that layer, what can they do to demonstrate that they are technically curious?
Speaker 1One is like come with a great explanation of how your app product that you've been selling works and be intellectually honest. I loved what you said about like this is where it's not good and this is where it's going to show that you have real command and control of how something you're actively working on. I think that's one of the things that really makes it work because most app companies actually, once you dive into it, there's something that's complex about it or hard about it. That's why someone's buying an app instead of just doing it themselves. So be able to articulate that in a compelling way. I would say the other thing is for me right now, like everyone can be a developer. It's so empowering. When I was at Twilio, one of the coolest but like hardest onboarding experiences was everyone had to build an app. And I mentioned my husband, he wasn't my husband yet and I were dating and he's a developer and like literally just downloading the Python libraries took me like two hours. I was like, I'm going to fail out of Twilio on my first month and eventually I got there, right? But now it's like you just fire up Cursor and you're like, I want to fine tune a small model, take me to fireworks documentation and like go try it. Like you can use plain English to go figure that out. And I think that that's super, super empowering. If you came to an interview with me and you were like, I taught myself this, I would think that's badass.
Speaker 2All right, y'all quick break. If you're getting value from this, do me a favor and hit the follow button. On YouTube, Spotify and Apple, it helps way more than you think. Now back to the episode, assuming that they're technical, so they pass that. Are you scanning for deal motion? Sounds like you might be doing both where, Hey, I might want the data driven PLG seller, but for our larger deals, getting me like the medic Mongo person. And it sounds like you're merging both of those cultures because you're running different motions on the go-to-market side. How do you think through the background as it relates to like sales motion?
Speaker 1Yeah, we need both. And I think this is where AI is different than anything before it. I had a candidate in my office and he was like, yeah, you're getting like, whatever, $10,000 PLG lands and then making them $20 million contracts. Like no other company was doing that, you know, within one quarter or whatever. And I think that's why you need both at a Twilio, at a Stripe, at any of the best, best, best, you know, Slack PLG companies. You can get a pretty big deal done and quite a lot of revenue with like very junior people at that company, you know, and we would like go talk to someone whose title would be like engineering. And they're spending $400,000 a month on Twilio or whatever. That's not AI. AI is very much a CEO and board level issue right now across all the vectors we talked about before. It's a budgetary thing, but it's also like this existential thing of using AI is going to allow you to succeed in the future. And I think that there's this major data issue that everyone is trying to navigate through. And so I think we're like the command of the message, med pick kind of stuff really gets you going is it's really training you to dig deep for a truth. And I think that there's this major data issue that everyone is trying to navigate through business pain, attach it to a buyer that's a senior buyer who has the authority to sign off on a $20 million deal or a $40 million deal and navigate that entire process very quickly. Whereas like I would say the PLG approach is like, oh, everyone's already basically bought it. I have it. technical win i even probably have a business win and now i'm just helping my developer advocates like get through their own procurement process to get a bigger deal and better discounts what is the
Speaker 2typical business win that is relevant for a fireworks sale so i understand the cost considerations but what about if you bring on fireworks you have this product competitive differentiation is it like oh we can launch just because now our margins make sense like what is that actual business win from i'm gonna help you all win because if you do fireworks you'll get why you know all of our
Speaker 1ai native companies are seeking competitive differentiation and lynn talks a lot about scaling into bankruptcy and in the sas world it was like yeah if you had product market fit it almost is like your business can scale but in ai land it's like the opposite like you can't you know scale if you're spending more on inference than you are generating revenue and so for them it's it's very important and i think that's like the business pain is having overall a successful sustainable differentiated product that you can bring to market in the enterprise the business pain is that i think every enterprise needs to grow and deliver results and keep up without spending more money and they are sitting on massive workloads for whatever it is claims underwriting customer support legal work everything to do with basically knowledge work and the pressure in our world right now is to do all of that five times faster at the same cost and that's really the pain point that we're solving there's no way to like close that gap i would say today yeah
Speaker 2it's interesting for you to say that about even the ai natives care about margin profile because like you know i'm like on twitter and i'll see like oh this startup's growing super fast for revenue but they're selling a dollar for 80 cents or whatever like i'm sure now they're getting a little bit more pressure as to like okay you know maybe you're a little bit uh unprofitable right now but like what's the story of how to get that and that's maybe where you guys are helping to like inflect
Speaker 1that totally and i mean the cost of compute is like rising very rapidly the opposite happened in the area uh era of cpus got it right the price of cpus went down unit costs dramatically as like platform as a service type companies were scaling gpus are increasing substantially so like the same job is going to cost you twice as much than it would last
Speaker 2year you said that there might be a rep that is selling 10k deal one day and then 20 million the next quarter how do you compensate for that and i think that's a really interesting question because i think it's a really hard problem to say to rep when you have that variance of deal size honestly
Speaker 1it's a really hard problem yeah i think this is why the labs like don't really have variable comp
Speaker 2can we actually why do you think that some of the labs only had a flat this is what we're going to pay you that's
Speaker 1such a good question i think yeah
Speaker 2give me your i give you mine just so you get the juices phone since i like this is so lame i like ask the question that asked me to answer it i just think that reps would be making way too much money if you put them on variable comp and then the engineers would be like what the heck these reps are order taking and they'd be almost mad that like a rep is not doesn't seem like that they're allowing effort to pay out essentially you know
Speaker 1it feels to me that a lot of the transformational deals companies do especially as they're scaling are truly like huge team efforts to go close and the rep is playing a role the traditional kind of like outbound pg hunt into these companies like that's not the genesis of a lot of the labs i would say it feels to me more like there's massive market awareness and demand for their products and so maybe that shifts how you think about compensating people you know my understanding from my friends who work at the labs and there's just amazing talent that's going there is like people want to be part of the big story they're not there for a w-2 and that's part of the cultural motivation is like not having as coin operated of a sales team yeah and having people that are thinking about the long-term business and just being part of it i think there's so many ways to to build a great company and as you know you and i talked about when i was at twilio at that beginning era we didn't really have sales and so we also went through a lot of these questions i think a lot of technical founders as they start to build sales they're like you know yeah why do why do i want these people who have to get paid for every little thing they do you know maybe there's just a different way to motivate and like find people who are intrinsically motivated to just win and and get big deals so
Speaker 2i agree with that but it's just funny because those people will end up making a ton of money like they know that it's like go to the company that pays you that doesn't pay you variable comp because it means the company is growing so fast they can't pay you at variable comp and because the company is growing so fast your equity will be worth a lot which means that you actually will get paid a lot of money and then the hack is like the tax that you will pay a lot less in taxes actually if you get that in equity incredible talent that is going there and like the best reps choose companies that have a tailwind like you don't get points for playing hard mode i agree with a lot of what you said and i think like last thing on hiring like is there anything else of like maybe learnings that you've had of hiring sales people throughout whether a twilio some of your other runs or even now at fireworks that you
Speaker 1care to share sure i mean i i bet this has been i know this has been shared on your podcast before but i think indexing on the resume and experience over how someone shows up a hundred percent is a lesson i've been like pretty burned by i care a lot about like our discovery interview sales challenge as part of the process i would say there's people who show up amazing in a conversational interview and then they come to present their work and you're like oh my god this was not good and so i highly recommend you have people show you their work and have a role play in every process that you're in and honestly i think there's a lot of sales reps who are like oh i don't want to do this like part of the process it's a little bit of a deterrent and i'll just tell you that's by design like if you're not wanting to put work into the interview you're probably not going to be wanting to like grind on pg tuesday like seven weeks in or 12 weeks in those are two things that have been really important yeah
Speaker 2how would you advise somebody who's setting up an interview process that agrees that you want to see their game tape or how they perform what's too early but what's too late in the process if it's too late then you burn a lot of resources of a sales rep potentially you know wooing you in behaviorals but if it's too early maybe you do like there is a little bit of a courting process when do you think about introducing a role play in an interview
Speaker 1it for us is basically our second step so we do the hiring manager screen and then a prep call for the disco process and the prep call is part of the interview it's like what's the quality of questions and prep you're doing and then the disco and then you go on from there i think it's
Speaker 2so smart i think one thing that i've heard from other leaders on the podcast or even just people that we're about to interview is that they're introducing this but they don't know when to do it because they worry about weeding out reps too early but then again like the best reps are looking at this is like hey i'm finally raising my hand to interview at one of these companies i know it's going to be a lot of work and i'm willing to put that in yes so it's okay that you might lose some reps they're probably not ready to interview just yet then right that's right i remember i got a job in 2021 the interview process was so easy and i looked back at that and i was like what a red flag like yeah it's painful to have to like spend a weekend learning how to demo this product but it's the best company is going to have hard interview processes so if
Speaker 1you're serious about working there it's only going to help you in your job when you have to go sell it so if you think about it that way it's a no-brainer you only want to work somewhere that has a high bar for talent if you yourself are a high performer yeah i would say like to your point about choosing the right company and all the research you need to do like if you actually want to choose a good company to be at that is your prep for this type of interview anyway so like i would just view it as if you don't want to do it then you're probably you're right it's not the right time to hire i also think it's
Speaker 2unique like you're going to force yourself now to like if it doesn't work out fireworks going to make you more dangerous in the next interview yes so it all builds on itself i'd love to pivot to the i and the firework i'd love to pivot to the i and the fire acronym industrialize i think which was just like how you guys are using ai for leverage in sales i'd love to like learn a little bit more not only the sales team but tell me more about this finance use case that you said like oh yeah how are you thinking about the knowledge workers at fireworks and how they're using ai one of our like most
Speaker 1popular launches this year has been our nexus platform and what nexus is is like basically a way to take fireworks's inferencing layer and model routing and put it underneath all of your existing coding harnesses and so part of now what we do to onboard everyone into sales and we're hosting 60 people at our sales onboarding this week in our office is we get you onboarded to fireworks itself and we have a bunch of applications that we built on top of it including our gtm agent we have an amazing amazing gtm engineering team and a big part of our sales onboarding is showing you like all of the agents that we're using to do sales we have a prospecting agent that's amazing we're hosting our first user conference forge on november 3rd and we have a prospecting agent to help us get the right attendees for that so we keep building all these agents that power our different pg tuesdays which makes life so much easier we have agents for analysis so it's like never ceases to amaze me to just like go into my little cursor and be like what's wrong with my pipeline this week or this or that or all the questions that i can just ask we are doing a lot of really cool enablement agents for that we're using some of the cloud like voice mode and so we did one that i was playing with it i was just on a trip to europe and i was doing it on the flight um home where you know you just record yourself like doing a live role play on competitive intel and it was great so you know we're just scratching the surface of all the things that we could do with you guys having such
Speaker 2a baked out and technically deep engineering team how do you think through build versus buy because even in what you just said there it sounds like you are using cursor but then some custom build on top of it. Like, how do you think through that mix of, buying certain tools versus then taking, you know, a layer on top of that to customize it?
Speaker 1Across the company, we have a couple options for what, like, IDE you use. Okay. So some people are on Cloud Code, some people are on Cursor, some people are on whatever. They're always, like, kind of switching around. Within GTM, like, we have a lot of tools. We have Gong, we have Cloud, we have Cursor, and then we have a bunch of Slack agents that show up. And so we're really just trying to give people a lot of options and then let them incorporate it into their workflows in a way that makes sense. Right now, for us, it has made sense to build a lot of the applications we're using on our own inference with, like, V0 front-ends or things like that, lovable front-ends, whatever it is. We'll see as we keep scaling, you know, what makes sense. Like, I feel like we're just constantly learning.
Speaker 2Yeah, I didn't have much to do this weekend because I live in New York and I had interviews on Friday and then interviews today. So I forced myself in a room with Grokbot on Saturday. And that's the next wave of things. And people would say, like, well, why can't you just do it with Cloud Cowork and Cloud? I think that it's just too much. It's not specialized enough and it wasn't snappy enough. So if you guys are using Cursor, you should check out. That's, I think, in the next frontier is, OK, we have these Slack agents, but then not needing to natively do it because I don't want to have to worry about Salesforce's development capabilities. And then I think Cloud is a little bit too horizontal and it wasn't quick enough. I was using it for stuff. So great for my I use it for my daily chat. And I still use Cloud Cowork and stuff. But that's the next frontier. I haven't used Instinct or Muse. But this is the whole next thing that, like, I am so excited about. I feel like we just move the ball forward.
Speaker 1Yes. The frontier is just shifting all the time. And that's why I'm excited to be at Fireworks. I think we're here to help people take advantage of all of those changes. And I think the future is kind of. What you outlined, customers are not going to want to live in one ecosystem. They're going to want to be at the edge, at the cutting edge as all these new things happen. And that's our job is helping them get there.
Speaker 2Any specific use cases for the GTM team? So you had mentioned a couple. Hey, for PGA Tuesdays, is it that you surface up data and then kind of give the reps, hey, these are the ones that you're going to want to reach out to today? Is that right about that? And then what are the other things that, like, you guys are using for AI and GTM?
Speaker 1We actually had. We closed the quarter in person. Like, everyone came into the office every day last quarter. And one of those days, we ended up carving out a couple hours to have a little, like, Fireworks hackathon using our own product. And the winner was called Ting Ting is our finance leader who does DealDesk. And it was called Ting Ting AI. And it was just like a version of DealDesk that we hacked together in a couple hours. It's like does all of the work that Ting Ting does.
Speaker 2Will this get approved?
Speaker 1It was just so awesome, you know. And pricing out dedicated, like, inferencing solutions is complicated. There's token. And math. There's GPU math. There's what kind of hardware? What kind of model? So it's different for different kind of models. So it's actually quite complicated. And I remember when I joined, our product team was like, we just spend a lot of time on pricing. And I was like, that's weird. That seems like something I could do. And now, yeah, I really can do it. And so we're constantly building new things like that, where especially when the bottleneck is waiting on another person to do it. We're like, if an agent can just do it, even if the agent is getting you 95% there and then a human has to review it, as opposed to just waiting for someone to do it. I'm going to do a full loop that really helps. And because we have so many new people coming online, like, we really want them to feel like they're in control and they have the tools they need. So deal desk has been a big one. Prospecting is a big one. It's integrated with lemless. So we're like trying to just automate as much of the outreach as we can. Enablement is the other really big one.
Speaker 2Okay. And then it's enablement similar, like the piece that you had said of how there's competitive dynamics. Is that like actually a voice first or is it more like they had a call and they can chat with it? Is it when they're in Slack, the developer asks a question and there's like a little side panel. What do you mean with enablement?
Speaker 1So a lot of what we're doing right now are like voice role-play enablement things. And I think the reason for that is it's like, it's hard in the moment to come up with like a snappy answer. I even notice it when I go and I do, I'm a big believer in like, I do every enablement thing. I don't ask the team to do things I don't do. Yeah. The enablement voice starts talking to you in this chat bot and you're like, like, do I know the answer to that? You know? And I think practice. This is like the most important thing. Cause then when a customer asks you a hard question, you can just be cool as a cucumber and answer it. So for me, like role plays is the thing that I like, think every sales manager says they're going to spend time doing with their team, but like never has as much time to do as they think. So like, that's been a big one for us.
Speaker 2You need like scar tissue and muscle memory. And the only way to do it is by doing it. It's like reading a book on soccer versus playing. It's just like, you can read it. Then when it happens in the moment, you're, you need to get used to like the brainwaves firing a little bit. A hundred percent. Yeah, yeah, yeah. All that. Yeah. Those are great use cases. You've been interviewing a ton of people, fireworks, you guys are growing a ton. And then you did a lot of interviewing at Twilio. When you think of like the mafias or the, the sales orgs that you've been the most impressed with, whether that was in the SAS era or the AI native era, like what come to mind? And like caveat being ones that you haven't worked at. So, ah, good.
Speaker 1So you can't say Twilio. I won't say Twilio, although it was an awesome experience. I would say for me, I actually have like a ton of respect for the John McMahon, like school of play. Book companies. And we're really fortunate to have Sahir on our board from index who was Mongo's chief people officer. And so I've learned a lot about Mongo sales culture and we've hired some exceptional sellers from Mongo because of that, or from the Mongo kind of playbook universe. So that would be really high on my bar and the people that we've specifically hired, like databases are hard. You're really selling something. That's going to be very disruptive to your customer to like rip this thing out and move over. You know, those reps have been great. Databricks has been another. Um, team where I just have huge respect for what they've built as a company. They're also an investor in fireworks and, and I have a lot of respect for the sales crew over there. For me, it's like both those teams kind of like have the best of like a lot of hustle and grit and a culture about like winning, but also known for being technical.
Speaker 2Any of the AI natives that come to mind, maybe of sales leaders that you respect that recently went somewhere or like anything like that.
Speaker 1Yeah. Great question. Well, I think you had Steve.
Speaker 2So I was going to actually tell you, you can't answer Steve. He was just sitting there this morning. So. I'm going to ask you a question.
Speaker 1I think perplexity is an awesome, awesome, awesome product. Um, I think it's great. You know, uh, a company that I admire that's doing just exceptionally well in the enterprise is Sierra. I think that they are doing something I didn't necessarily think was like super possible in the sense of like, you know, at Twilio, we tried to sell contact centers really hard and they're going in and totally reinvented customer support at very large, very slow, slow moving enterprises. And they're moving quite quickly. And I love what Harvey is doing. I think that who would have thought that like old school law firms would be like the first to completely modernize with AI. And so I think what they've done is really interesting. I love that they have like forward deployed lawyers and stuff like that on their team. And I think they're taking just a really creative approach to sales and making that happen.
Speaker 2Agreed. What is Sierra doing differently that like other AI natives in support?
Speaker 1You know, my understanding, just supporting them a little bit, is that. When they're in a POC, they win. Their product is really good, but like their team is really good. They have really great FDs. They show up and really help customers integrate the solution. Got it. In the best way possible. And I think for me, like the buying journey is going to continue to go in the direction of the customer needs to see value first. And the commercial conversation comes after that. Yeah. You know, 10 years ago, it was more like you show them the slide deck and then they're like, can I try it? And you're like, yeah, like only after you sign. In a contract, you can like put your.
Speaker 2You got to sign a hundred thousand. Yeah. I mean, that was my life at Salesforce. Like there were so many times where you were selling off of. All right, well, I'll sign this because you guys are Salesforce and like you guys will make this right. We trust you. But especially as you go down market to these startups, you have to do it the opposite way.
Speaker 1Yeah. So. AI is just changing everything. I mean, everything is changing very, very quickly. But there's I mean, there's just products out there that I think are so exceptional right now. And when you kind of extrapolate. It out, you're like, wow, we're already at a place where like what lovable does blows my mind. What Sierra does blows my mind. What Harvey does blows my mind. Like what cognition does blows my mind or we're just getting started. So it's a very exciting time to be in this space. There's probably never been a better time to build a company.
Speaker 2Totally with Harvey got me thinking, I want to bring it back. So you said your finance team is using a lot of tokens. Uh, what are you seeing internally at the knowledge worker, specifically a finance team use AI for.
Speaker 1I mean, there's so much that they're doing. Like we're starting to like close our books. Like a grown up company get gap, you know, all these like various proper ways of doing business. And they're using AI to like do a ton of things around like reconciling numbers and pulling systems together and like changing how we're doing billing. We just raised a big funding round. And so that was like another thing is there's just agents that can be part of anything. I think finance is very close to code in a sense, and that it's all sort of like numbers and a lot of those problems can be rendered as essentially code. And if a problem can be rendered as code, it can be like tackled with AI. And so, um, totally agree.
Speaker 2We're using it now. We're not as nearly as big of a team as you guys are. Any of the companies we interview, but, uh, we're using now these agents graph, but I keep bringing up graph by you use any of these agents to just like kind of give you nudges of where people are maybe over on invoices and as an air cover, seeing if you're making any mistakes, like we'll use it. So when we send an invoice, it scans mercury. Hey, is that reconciled with the contract that we had sent? So we're starting to get a little bit better at that. There's so many things we've been able to do.
Speaker 1It's like, you know, one of our most precious resources is our compute. And so we. like use agents to make sure our fleet is being perfect performance optimized at any given moment that we don't have any idle gps that shouldn't be and all that kind of stuff we have this like perf agent that's like figuring out how to just like constantly squeeze more throughput out of the system and it's just like it's so cool it's like anything can be made better with ai yeah
Speaker 2oh yeah this is great thanks so much for coming on the crew podcast you guys are hiring for sales yes sales what is it sales fd like what are the the roles we're hiring for sales for field
Speaker 1engineering we're hiring leaders of all those teams we're hiring in north america europe and
Speaker 2singapore great sweet thanks for coming on the crew podcast yeah thanks for hanging with us if you like this one follow the show on spotify apple podcast and youtube and we'll see you next week cheers
PodcastTranscript AI