The Exchanges

Q Okay. So help me out here. We're both like CEOs as well. How do you think about promoting people then who are great ICs? Do you not promote them to managers? What's the right way to do that challenge them?

A That is a great question. So actually every time I've promoted an IC to a manager in a startup, this was not at Twitter, but in a startup scenario, we've done it as sort of a trial period. And so it's like, Hey, so-and-so we have this opening for this new role. The team's growing. We need some structure. Uh, we are going to have so-and-so take on the role as a trial over the next two months and let's see how they do. And honestly, there's been both outcomes in some circumstances. It's been great and everyone's rallied around them and it's like, great. Okay. Now they're the manager for that team. And I've had circumstances where they haven't, and it was sort of widely recognized that they weren't excelling in that role. And we decided to move them back to an IC and that was okay as well. I think there's a, A really bad perception that being a manager is better than being an IC. I think it is different and you can be an exceptional IC and you should be comped appropriately for that. Or if your career passion is to mentor and guide folks, like then you move into management.

AI assessment note: “we've done it as sort of a trial period”

Q I mean, that five-year journey is one with twists and turns, and the idea that initially was Digits on founding day one is different in terms of the product that we're releasing today. What did you learn that led to your realization of the need to pivot? Like, why pivot, and why was that enough?

A Yeah, it's definitely been quite a journey, and I know you've been on it with us for years now, so I deeply appreciate it. Obviously, trying to make accounting real-time is a lot easier said than done. Uh, when we started the company, we went heads down on R&D, And really struggled with data quality for like three years. And that was because in 2018, when we started the tech to really automate bookkeeping didn't fully exist. I think I was a little optimistic on how it could work. And so we have dozens of patents on it now, but it was like a brick wall. Um, so in 2021, we made the decision to pivot from like the pure bookkeeping automation to collaboration tools. So better financial reporting, better client portals, Better transaction review process. And, and that worked. We got a thousand accounting firms on the product, 5000 downstream businesses, like that was sort of off and running. But what bugged me is that wasn't really why we started the company. We had bigger ambitions. And so then last year, literally all of a sudden GPT three comes out, chat GPT comes out, GPT four, and we started experimenting and we're like, Whoa, hold on. We're back. Like we can actually do what we set out to do. And so literally overnight, just like we're back focused on this and spent this whole year building it.

AI assessment note: “we made the decision to pivot from like the pure bookkeeping automation to collaboration tools.”

Q you. So, uh, you said about IBM there. You said how silly. I, I say how brilliant. That's called a monetization strategy in every environment. I, I think that actually AI implementation services will be one of the biggest categories in the next few years. Do you agree with me, or do you actually think that enterprises will adopt natively? It'll be fine. How do you think about that statement?

A I, I think you're probably right in that it's a large market. To me, it's not a very interesting market. Like, yes, there'll be a lot of consulting to help enterprises adopt AI, and the products won't be that great, and they won't really make that much of a difference. What I see when you come to platform shifts is, again, this opportunity for sort of a net new approach to really take over fast, and what gets me excited is, like, let's build real workflow automation for real people in massive industries that are outdated, like, for example, accounting. Like, are you going to trust Intuit to adopt this and totally disrupt Their own product lines, or do you think a new upstart's going to come in and do it better? And that's where I'm excited to see and like pour energy into.

AI assessment note: “I, I think you're probably right in that it's a large market.”

Q So when you advise founders on the right way to pivot, what would you advise them knowing all that you do now?

A Yeah. So the key is getting your team on board. If your team loses trust in you, you literally have no one to pivot. So it doesn't matter. And they can really sense the uncertainty. And so my, my other advice, as I've said, is like be very intentional and very decisive. And so both times we've pivoted digits. I gathered our core leadership team laid out like, what are the challenges? What am I seeing? What are the options? And we knew within 24 hours what the new path was and what the priorities were. We were all in on that new direction. And so I, it comes back to conviction. Like it is still a bet, but it's like, Hey, here is the information we have on the field. We need to make a decision right now because what kills companies is uncertainty. And if you sort of muddle your priorities and have one team, try this and another team, try this, no one's heart is in it. And both are going to be mediocre. I would rather see founders take like one, To the moon bet on one new direction and it either works or doesn't.

AI assessment note: “So the key is getting your team on board.”

Q you know, when we think about non-obvious things, we mentioned that two commonly said tropes, obviously speed, and then solve a problem that you know deeply. I, I think there's a lot of things that aren't well known about entrepreneurship. Given the fact that you've done Crashlytics, you've now been in Twitter, you've now founded Digis. What do you think is the most like misunderstood Or non-obvious element of entrepreneurship?

A This sounds silly, but honestly, pure execution. And what I mean by that is people know the vast majority of managers are terrible managers, right? The vast majority of founders are simply bad at running companies. And I'm sorry, but it's true. And so what I mean by that is like most founders aren't intentional about how they go through and operate the business, intentional with their time, with their decisions, with who they hire, with what they say no to. And so if you don't have conviction, you're really going to struggle as a founder because you need this like deep seated obsession of what's right and what's wrong and what you believe in and how that informs every decision you make and your decisions may still be right or wrong. They're not going to be perfect, but if you were intentional about them, at least you can trace that back and learn from it versus I see too many founders just sort of going on a random walk. And then when it turns out they were wrong, what do they have to learn? There's not, there's nothing to trace it back to.

AI assessment note: “This sounds silly, but honestly, pure execution.”

Q Okay. So help me out here. We're both like CEOs as well. How do you think about promoting people then who are great ICs? Do you not promote them to managers? What's the right way to do that challenge them?

A That is a great question. So actually every time I've promoted an IC to a manager in a startup, this was not at Twitter, but in a startup scenario, we've done it as sort of a trial period. And so it's like, Hey, so-and-so we have this opening for this new role. The team's growing. We need some structure. Uh, we are going to have so-and-so take on the role as a trial over the next two months and let's see how they do. And honestly, there's been both outcomes in some circumstances. It's been great and everyone's rallied around them and it's like, great. Okay. Now they're the manager for that team. And I've had circumstances where they haven't, and it was sort of widely recognized that they weren't excelling in that role. And we decided to move them back to an IC and that was okay as well. I think there's a, A really bad perception that being a manager is better than being an IC. I think it is different and you can be an exceptional IC and you should be comped appropriately for that. Or if your career passion is to mentor and guide folks, like then you move into management.

AI assessment note: “we've done it as sort of a trial period”

Q you. So, uh, you said about IBM there. You said how silly. I, I say how brilliant. That's called a monetization strategy in every environment. I, I think that actually AI implementation services will be one of the biggest categories in the next few years. Do you agree with me, or do you actually think that enterprises will adopt natively? It'll be fine. How do you think about that statement?

A I, I think you're probably right in that it's a large market. To me, it's not a very interesting market. Like, yes, there'll be a lot of consulting to help enterprises adopt AI, and the products won't be that great, and they won't really make that much of a difference. What I see when you come to platform shifts is, again, this opportunity for sort of a net new approach to really take over fast, and what gets me excited is, like, let's build real workflow automation for real people in massive industries that are outdated, like, for example, accounting. Like, are you going to trust Intuit to adopt this and totally disrupt Their own product lines, or do you think a new upstart's going to come in and do it better? And that's where I'm excited to see and like pour energy into.

AI assessment note: “I, I think you're probably right in that it's a large market.”

Q How do we feel about the cost of compute changing over time? You know, right now it's actually, we mentioned that kind of cannibalization, like cost of queries is significantly higher with, you know, models than it is for search today. How do you feel about cost of compute and cost of query in the next three years? Is it traditional Moore's law? Is it faster? Is it slower?

A Yes. I I'm admittedly not a hardware expert. Uh, so I don't know if it'll exactly follow Moore's law or not. It'll certainly improve. What has been interesting to see is it's very clear the top models are memory bound as well as CPU bound. So you can't just make the CPUs, the GPUs faster. You need to increase memory bandwidth in, uh, sort of on par with that. And OpenAI actually shared a tech talk a few weeks ago that talks about how they were tuning their data centers like this. It's going to be new challenges for NVIDIA, for ARM, for these chip companies to really unpack, but I mean, I would bet on the pace of technology. It will get a lot faster and a lot cheaper very, very fast.

AI assessment note: “It will get a lot faster and a lot cheaper very, very fast.”

Q Jeff, again, help me. What's the single most important way a VC can be helpful?

A I would say by drawing your attention to what's important, but not urgent. So it's been funny because all these VCs over the past few years have touted value-added services in other parts of the business, and they can recruit for you and do these other things for you. I actually, I don't need the investors to do stuff For us, right? Like as an operator, your job is to execute. Your job is to operate. I'm fine with that. What I think is most helpful is when your investors are always looking forward because they're not caught up in the day to day. And many founders get so caught up in the day to day, their context switching, their racing between fires, that they sort of lose track of what's important in the bigger picture. And so to me, a great face C is basically always looking forward for their company, drawing the founders to spend time on what's really important. Even if it's not the most urgent thing right now, and sort of coaching them on how to be more strategic.

AI assessment note: “by drawing your attention to what's important, but not urgent.”

Q Jeff, again, help me. What's the single most important way a VC can be helpful?

A I would say by drawing your attention to what's important, but not urgent. So it's been funny because all these VCs over the past few years have touted value-added services in other parts of the business, and they can recruit for you and do these other things for you. I actually, I don't need the investors to do stuff For us, right? Like as an operator, your job is to execute. Your job is to operate. I'm fine with that. What I think is most helpful is when your investors are always looking forward because they're not caught up in the day to day. And many founders get so caught up in the day to day, their context switching, their racing between fires, that they sort of lose track of what's important in the bigger picture. And so to me, a great face C is basically always looking forward for their company, drawing the founders to spend time on what's really important. Even if it's not the most urgent thing right now, and sort of coaching them on how to be more strategic.

AI assessment note: “by drawing your attention to what's important, but not urgent.”

Q wildly successful and experienced some incredible growth under yours and Wayne's tutelage. But one of the problems that you did say that you encountered within the acquisition itself with Twitter was a breakdown of the reporting structure. So how can founders in a negotiation and in acquisition talks, how can they ensure that they're speaking to the VP of engineering, the CTO, the people who are really making the decisions?

A Yeah, I think this is also a super important aspect, and we thought we were set up for success here because we did come in and we're reporting directly to the VP of engineering. What we didn't anticipate as a startup was that large companies also go through periodic reorgs. Then, of course, during that, there's different priorities for the company. They maybe put you off in some other location because you're not their focus at the time. And so what I would have done in retrospect is sought some commitment Uh, to continue to report to that person for X period of time. I don't think it's fair or actually good to say, Hey, I want to indefinitely report to this person. You want to move around as the company changes structure. But for us, it was very short term. It was sort of a month in that Twitter happened to do a reorg and we just got unlucky. And so I would have sought some sort of maybe six or nine month commitment to stay at that level so that we could really Integrate into the company, understand what was going on, and then identify the best place to report into.

AI assessment note: “sought some sort of maybe six or nine month commitment to stay at that level”

Q And as you suggest that one of your reasons for being favorable towards the Twitter acquisition was their agreement to continually invest in the Crashlytics platform, something that didn't maybe happen within Creo and Box. Yep. So how do you think founders can negotiate for this continued investment in their product and sector when in potential acquisition negotiations?

A Yeah, this is really, really important, and the biggest lessons I've learned from Twitter Um, and to be clear, they are incredibly professional at doing acquisitions. Twitter on average acquires one to two companies a month. They have for years. And so over the course of dozens of deals and dozens of experiences, they've really honed their process. And one of the key aspects is total transparency as to the motivations. And so it's really key for the long-term success of the deal for the acquiring company to be completely open about why they are doing it. Why is this strategically important to them? And how do they see that fitting into their future plans? And it's equally critical for the founders to be totally transparent on what they want out of it. Like, why are they considering this deal and what's important to them? And you're right. Our biggest lesson from in Creo, uh, was, was not being able to guarantee future investment in the technology. And so that was a big aspect of our discussions with Twitter. Like, was this strategically important to them? Were they willing to commit future headcount? And that becomes part of the deal terms. It's like, yes, we will Do this amount of investment over this many years? Um, and were they willing to give us the support and visibility internally so that we don't become some sort of backwater? And one aspect of that is we had started Cr…

AI assessment note: “Were they willing to commit future headcount? And that becomes part of the deal terms.”

Q honestly, 90% of companies that I see are like AI for wealth management, AI for podcaster, album artwork creation. I'm not dissing it, but like, you know, and it's like three weeks old, and I'm like, that can't have been that difficult to build. My question to you is, what's the difference between a thin layer on top of one of these models and a thick wrapper with inherent value?

A Yeah, this is the real problem you're hitting on. So, Startups are gonna get killed because they're very thin wrappers. And I agree with you. Most of what I'm seeing on the angel side right now are these very thin wrappers on top of open AI. And like, if your primary product value is, is scripting GPT, that's a thin wrapper. If you've built it in two to four weeks, that's a thin wrapper. And so the, the key thing to me is like, we are in a hype cycle around AI, much like potentially the hype cycle with crypto and other tech before it. And all of these sort of fake use cases are going to get quickly washed out and replaced and commoditized. It's really important to me that people view this as a technology. It is, it is like your MySQL database. MySQL was very popular 20 years ago, right? Like it was super cool what it could do. And it's still cool, but if you try to go raise money on you've built a form on top of MySQL, you're not going to be successful. And in five years, that's what a lot of these are going to look like. You need to really focus on the market and like solving a core problem.

AI assessment note: “If your primary product value is, is scripting GPT, that's a thin wrapper.”

Q Um, you mentioned about kind of, uh, OpenAI and ChatGPT kind of, I don't know how to say it correctly, but like opening your eyes to the new possibilities that were available to you. When we think about that, I'm just intrigued. Do you worry about a lot of your company being based on an external party's direction development? It is unlike other times in that way.

A Yeah, this is such a special moment and it, I'd say it vividly reminds me of the rise of mobile, which we wrote in Crashlytics from 20 10 to 20 13, the Web two O transition in 2005 to 2007. It's like these platform shifts are so rare and I love them because that's where all the massive opportunities arise. And so the key though, as a startup founder is to be in a position to rapidly adapt to that new reality and you need to be faster than your competition. So like for five years now, I've run digits on a weekly sprint. Every week we get to decide what our direction is that week. And I have to say that's been critical this year, keeping up with all of the changes and evolution of the tech. So I'd say, I don't know. I'm not particularly afraid of it. I love the energy of the tech world moving at a, at a huge pace. And so the key thing that was keep in mind, this is technology. Like we are viewing it as a tool. It's not life or death for the company. It's like databases and so on. It's just evolving faster. And so let's see how we use it, and it gives us more capability, but you ultimately need to stay focused on your customer and what problem you're solving, first and foremost.

AI assessment note: “I'm not particularly afraid of it. Like we are viewing it as a tool.”

Q I'm just kidding. You don't have a billion. Um, it'd be a sad day. My question to you would be like, Ian, bluntly, do you think the best companies will leverage many at the same time, or will it be one that we rely on?

A That is a really good question. Uh, so we built our ML team Three years ago, we're training our own in-house models. You're right. Like you can't hallucinate in finance. We've done a ton of work to make sure our math is always accurate. It's it'll be really interesting to see there's every model has different strengths. Like if you already look at Bard and versus GPT and so on and so on. Um, my sense is there'll be a very common popular open source sort of base LLM and then tools to allow folks to fine tune it easily. And what we've discovered is While it creates, it takes so much time, data, energy, money to train an LLM, fine tuning it can actually be done relatively straightforwardly with a surprisingly small amount of data, as long as your data is very high quality and focused. And so imagine you basically have your sort of default Linux operating system, right? And then every, everyone customizes their flavor of it for their product, market, use case, whatever it might be.

AI assessment note: “my sense is there'll be a very common popular open source sort of base LLM”

Q honestly, 90% of companies that I see are like AI for wealth management, AI for podcaster, album artwork creation. I'm not dissing it, but like, you know, and it's like three weeks old, and I'm like, that can't have been that difficult to build. My question to you is, what's the difference between a thin layer on top of one of these models and a thick wrapper with inherent value?

A Yeah, this is the real problem you're hitting on. So, Startups are gonna get killed because they're very thin wrappers. And I agree with you. Most of what I'm seeing on the angel side right now are these very thin wrappers on top of open AI. And like, if your primary product value is, is scripting GPT, that's a thin wrapper. If you've built it in two to four weeks, that's a thin wrapper. And so the, the key thing to me is like, we are in a hype cycle around AI, much like potentially the hype cycle with crypto and other tech before it. And all of these sort of fake use cases are going to get quickly washed out and replaced and commoditized. It's really important to me that people view this as a technology. It is, it is like your MySQL database. MySQL was very popular 20 years ago, right? Like it was super cool what it could do. And it's still cool, but if you try to go raise money on you've built a form on top of MySQL, you're not going to be successful. And in five years, that's what a lot of these are going to look like. You need to really focus on the market and like solving a core problem.

AI assessment note: “if your primary product value is, is scripting GPT, that's a thin wrapper.”

Q So how do you maintain a strong sense of communication with them and ensure they feel kept in the loop whilst maintaining this lack of transparency that we just said was beneficial?

A Yeah, that's a great question, and that's very much the case. There's a clear reason why your team joined you, chose to follow you, and chose to join This smaller riskier company than going and get getting hired by a bigger one. My belief on acquisitions and, and IPOs for that matter, exits of any kind is that they are not about exiting. The entire point is to become more successful at your mission. And so the reason the box acquisition wound up being quite successful for us is that we were able to stay true to our mission and we were able to show our team that we were staying true to the mission. And so we had been building this document preview technology Box, of course, had millions upon millions of documents, but no way to show them, and so we could actually take what we had built, scale it overnight very broadly, and provide a lot more value to a lot more people in so doing, and so that was the main story you tell your team about how, hey, this actually furthers our mission, and for us to go and actually achieve this independently, we would have had to build a lot of what the acquirer in this Case Box already have. And so this is simply accelerating our roadmap. This is saving years of our journey, and we're going to be far more successful because of it.

AI assessment note: “that was the main story you tell your team about how, hey, this actually furthers”

Q Can I ask you, how do you advise founders on when they have enough data to make that pivot? Because you don't want to make it too quickly where it's like, whoa, hold up horsey, not enough data, but also you don't want to be too slow. How do you know when you have enough data to make a decision?

A That is the million dollar question. And Uh, I'd say it's more of an art than a science. Like you need to have a feel and an instinct as the founder of like, do you see a path to success? And if sort of the windows closing on your path to success with your current business, like that's to me when you have to go pivot and a lot of super successful companies were hard pivots, right? Like Twitter was a podcasting startup. Slack was a game. YouTube was a dating website. Like it is, it's totally crazy. And so to me, it's like impossible to say like, oh no, stop. Like that pivots too far field. Like you have to like have this sort of founder instinct and that's how great companies come to be. The one thing I'd modulate that with is like, you need at least a year of cash left, because if you don't have a year of cash, you're not going to have time to see this pivot through. And so it's really like founder grit and enough cash. If those aren't there, return your capital. If those are there, I would go for it. That's how like huge opportunities come about.

AI assessment note: “I'd say it's more of an art than a science.”

Q Um, you mentioned about kind of, uh, OpenAI and ChatGPT kind of, I don't know how to say it correctly, but like opening your eyes to the new possibilities that were available to you. When we think about that, I'm just intrigued. Do you worry about a lot of your company being based on an external party's direction development? It is unlike other times in that way.

A Yeah, this is such a special moment and it, I'd say it vividly reminds me of the rise of mobile, which we wrote in Crashlytics from 20 10 to 20 13, the Web two O transition in 2005 to 2007. It's like these platform shifts are so rare and I love them because that's where all the massive opportunities arise. And so the key though, as a startup founder is to be in a position to rapidly adapt to that new reality and you need to be faster than your competition. So like for five years now, I've run digits on a weekly sprint. Every week we get to decide what our direction is that week. And I have to say that's been critical this year, keeping up with all of the changes and evolution of the tech. So I'd say, I don't know. I'm not particularly afraid of it. I love the energy of the tech world moving at a, at a huge pace. And so the key thing that was keep in mind, this is technology. Like we are viewing it as a tool. It's not life or death for the company. It's like databases and so on. It's just evolving faster. And so let's see how we use it, and it gives us more capability, but you ultimately need to stay focused on your customer and what problem you're solving, first and foremost.

AI assessment note: “I'm not particularly afraid of it. I love the energy of the tech world”

Q now? Because I speak to some of the largest enterprises in the world, and they're like, dude, I'm not sending my customer data, my transaction data, and I'm not picking on OpenAI here, but to a model that is outside of our bounds. How do we think about enterprise control of very sensitive data in this world where they want to get the benefits, but don't want to lose control?

A Yeah, two different thoughts here. This is a really good question. So they had the exact same reaction to cloud. If you go back 10 years, they were like, I would never put stuff on AWS or Google Cloud or Azure or whatever. That's ridiculous. Why would I share my data with those companies? Right? And now it's just like, not only do they all do it, but it's actually better because those companies core competency is running data centers. Most enterprises have no idea how to operate a data center. And so I think this will go in the same direction. Um, and like you'll, you'll have very clear guidelines around how the companies use the data for model training and it's off limits and so on. And sort of that trust will be overcome. The other angle though, is there's also the danger of every enterprise jumping on AI because it's hot and cool and sort of like they all jumped on blockchain for zero reason, even though it did nothing. Right. And like IBM's at fault. IBM was consulting, like charging for services to consult on how to adopt blockchain into your enterprise. That's ridiculous. And so, like, again to me, focus on your customer, your market, your product need, and view this as a tool, not a panacea, and adopt it strategically on, like, what makes sense and where it's going to push the product forward.

AI assessment note: “they had the exact same reaction to cloud”

Q How do we feel about the cost of compute changing over time? You know, right now it's actually, we mentioned that kind of cannibalization, like cost of queries is significantly higher with, you know, models than it is for search today. How do you feel about cost of compute and cost of query in the next three years? Is it traditional Moore's law? Is it faster? Is it slower?

A Yes. I I'm admittedly not a hardware expert. Uh, so I don't know if it'll exactly follow Moore's law or not. It'll certainly improve. What has been interesting to see is it's very clear the top models are memory bound as well as CPU bound. So you can't just make the CPUs, the GPUs faster. You need to increase memory bandwidth in, uh, sort of on par with that. And OpenAI actually shared a tech talk a few weeks ago that talks about how they were tuning their data centers like this. It's going to be new challenges for NVIDIA, for ARM, for these chip companies to really unpack, but I mean, I would bet on the pace of technology. It will get a lot faster and a lot cheaper very, very fast.

AI assessment note: “It will get a lot faster and a lot cheaper very, very fast.”

Q Can I ask you, how do you advise founders on when they have enough data to make that pivot? Because you don't want to make it too quickly where it's like, whoa, hold up horsey, not enough data, but also you don't want to be too slow. How do you know when you have enough data to make a decision?

A That is the million dollar question. And Uh, I'd say it's more of an art than a science. Like you need to have a feel and an instinct as the founder of like, do you see a path to success? And if sort of the windows closing on your path to success with your current business, like that's to me when you have to go pivot and a lot of super successful companies were hard pivots, right? Like Twitter was a podcasting startup. Slack was a game. YouTube was a dating website. Like it is, it's totally crazy. And so to me, it's like impossible to say like, oh no, stop. Like that pivots too far field. Like you have to like have this sort of founder instinct and that's how great companies come to be. The one thing I'd modulate that with is like, you need at least a year of cash left, because if you don't have a year of cash, you're not going to have time to see this pivot through. And so it's really like founder grit and enough cash. If those aren't there, return your capital. If those are there, I would go for it. That's how like huge opportunities come about.

AI assessment note: “I'd say it's more of an art than a science... you need at least a year”

Q I'm just kidding. You don't have a billion. Um, it'd be a sad day. My question to you would be like, Ian, bluntly, do you think the best companies will leverage many at the same time, or will it be one that we rely on?

A That is a really good question. Uh, so we built our ML team Three years ago, we're training our own in-house models. You're right. Like you can't hallucinate in finance. We've done a ton of work to make sure our math is always accurate. It's it'll be really interesting to see there's every model has different strengths. Like if you already look at Bard and versus GPT and so on and so on. Um, my sense is there'll be a very common popular open source sort of base LLM and then tools to allow folks to fine tune it easily. And what we've discovered is While it creates, it takes so much time, data, energy, money to train an LLM, fine tuning it can actually be done relatively straightforwardly with a surprisingly small amount of data, as long as your data is very high quality and focused. And so imagine you basically have your sort of default Linux operating system, right? And then every, everyone customizes their flavor of it for their product, market, use case, whatever it might be.

AI assessment note: “there'll be a very common popular open source sort of base LLM”