Howie Liu: Decoding Airtable's $11B Valuation; The Impending AI Revolution in Enterprise | E1053

1. 0:00 – 6:03

   Personal Insights and Product Strategy

   1. HLHowie Liu0:00

      I tend to be a believer that AI has the potential to actually lower the cost of goods and services production, and therefore increase also the demand. There's a potential world, though it's certainly not guaranteed, where we find ways to use more human potential and employ more people because each person is made more productive by AI.

   2. HSHarry Stebbings0:20

      Howie, I'm so excited for this, man. I always love our chats, so thank you so much for joining me today.

   3. HLHowie Liu0:26

      Yeah, it's great to be with you.

   4. HSHarry Stebbings0:28

      I always love to start with like the founding stories, but I wanted to do something a bit different today. I wanted to start with, if you could call yourself up the night before you started Airtable and gave yourself a piece of advice, what would you tell yourself?

   5. HLHowie Liu0:41

      The thing about advice is, it's always, uh, 20/20 in hindsight. Um, so there, there's a million little things that I would tell myself, but I think, I think the biggest one is, you know, it's really important to think not just about product market fit. You know, every first-time founder is concerned about product market fit, and there's this adage that like, you know, the second-time founder thinks about distribution. My take on it is a little bit more nuanced, which is, it's still product market fit, but it's about figuring out the right product strategy that marries with an effective go-to-market model that, that lends well to that product, right? So certain products work well in a bottoms-up, viral, you know, organically adopted way. Certain products are single user versus team-centric. You know, accordingly, I think you need to design a go-to-market model, whether it's outbound sales or performance marketing or organic and viral growth, um, that really pairs well with, with that particular product dynamic. In hindsight, we didn't think enough about go-to-market model, uh, in the early days of the company. We just focused on building a good product and, you know, happily and, and luckily, that was good enough to get us to the next phase where we did start thinking about go-to-market. But I think it's something that, that, uh, we could have thought about a lot more.

   6. HSHarry Stebbings1:51

      If you had have thought about it a lot more, what would you have done differently? Because to the outside world, it seems like a perfectly executed PLG strategy. What would you have done differently had you had have thought about it?

   7. HLHowie Liu2:01

      I think one of the differences was, you know, we built the product with the ability to become team-centric. So we built, you know, we spent a lot of time engineering the backend to be real-time collaborative in nature, which was not a small technical feat. We thought about the team-centric use cases, uh, and when we started putting out templates, this was a little after we launched, initially, you know, we built, uh, some templates that were more solo user-centric, some that were more team-centric. In hindsight, you know, I wish we had put even more emphasis on the team scale use cases, and especially those use cases that, that involved larger teams, right? I think we got to those sort of organically, but it was a little bit more, um, diffuse. So early on, we got a very large range of different adoption, and, uh, I wish we had guided our adoption more and earlier towards the, the team-centric use cases with the anticipation that those would be the ones that, um, A, monetize better, uh, B, um, you know, lend to more, uh, you know, more go-to-market models, right? You know, if you're taking a single user product out to market, um, it's just harder to make performance marketing work. It's harder to make... You know, you can't really make outbound sales work, right? The economics of, you know, pitching a single user who pays a single and relatively small dollar cost to you, uh, is just hard. So I think being more, um, committed to, uh, from the earliest days, this, the team-centric use cases and even larger teams, um, would have, uh, would have lent better to early on a, a pairing of, uh, more aggressive go-to-market, uh, models.

   8. HSHarry Stebbings3:38

      We're gonna move into the topics that we originally planned, but I do just have to ask this. As a VC, people actually come to me for advice, which is bizarre and fucked up in so many ways, uh, but my question is, I have so many PLG founders who say, "We have a horizontal product," similar in the way that Airtable is, so many different use cases. Should we go for two or three verticals and product market aggressively towards them? Or should we let a thousand flowers bloom and just let the world try our products and be much more horizontal? What advice would you have for those founders? 'Cause I never know the answer.

   9. HLHowie Liu4:12

      Uh, I don't think it's, it's for lack of, uh, good thinking from you. Um, I think it's a genuinely tricky question. Uh, there's nuances, right? You know, some horizontal products, uh, are actually, you know, a vertical of their own, right? When I think about, you know, a Slack or a Dropbox, I think they actually fall into a very well-defined category in my mind. Like, you know what you're gonna use Slack for, you know what you're gonna use Dropbox for. So even though they're not function or industry specific, they kind of occupy a well-defined category of their own, right? You know, you know you're gonna do chats and, and, uh, communication on Slack. You know you're gonna do file sharing on Dropbox. Um, I think for, for other products where the use cases are not as self-evident, um, the importance of verticalization, especially early on, is more important. Um, it's kind of the, the crossing the chasm versus inside the tornado metaphor, where early on, I think, especially when the market is still trying to figure out, "What is this thing?" Um, you know, it's, it's helpful to really go deeper into a few different use cases and make sure you fully solve those, while at the same time preserving the optionality in the long term. You know, not over-committing the platform or the product roadmap in such a way where, you know, you're so brittle, you're so hard-coded for those narrow use cases that you can't expand outward. So, I think these are, uh, pithy statements that are easy to, to, uh, to speculate on in the abstract. But when you get really specific in terms of what is the use case, what are the, the functions, um, what are, what is the product roadmap one way or the other, and the go-to-market model, I think that's where it gets really interesting.

   10. HSHarry Stebbings5:43

       Speaking of it getting really interesting, we first kind of reconnected and decided to do this next discussion be- because of a tweet that I did on AI and essentially how it's integrated into enterprise. I wanna start with kind of this thread, so to speak, and it's, why are advancements in AI as significant as the introduction of cloud computing?...

2. 6:03 – 30:52

   AI and Technology

   1. HSHarry Stebbings6:04

      mic over to you.

   2. HLHowie Liu6:05

      Yeah. Yeah. I mean, I, I would even venture to say they're, they're, uh, they're different but potentially, um, going to be more profound. And I think the reason for that is, you know, the, the benefits of cloud compute are, are pretty easily grokable, right? Um, you don't have to manage your own servers. There's infinite scalability. You know, as hardware costs go down, you kind of instantly realize the, the benefits of that, right? Like, we just pay our AWS bill every month and, you know, over time, over the long term, you get more compute, more scalability for cheaper, right? And you get more and more resources that are provided that solve a lot of not just, you know, physical headaches, uh, with managing your own, um, uh, servers, but even software headaches increasingly, right? Um, they, they give you architecture out of the box. So by all means, there, there's a lot of power and value to the cloud compute model. I think AI, while still in its infancy, and especially LLMs, are gonna be profound in a very different way, which is, you know, when you think about the broad range of knowledge work that's done today, you know, every, every, uh, function, right? Whether it's legal or finance or even the more creative functions, marketing, et cetera, you know, you can just see the glimmers starting to emerge of how a really, really broad set of, uh, work can be automated or accelerated with AI, with especially GenAI. Um, you know, I think, uh, we've probably all spent way too many hours playing around on Midjourney with fun prompts and just being kind of, uh, having our minds blown with what AI is capable of. I mean, it's seemingly creative, right? Like, now whether you, you, you, um, you actually believe, uh, you know, you're willing to ascribe it the word creative or not, I think objectively, it is able to produce useful output, right? And, um, and it's easy to squint and imagine as the capabilities of these models only get better and better, that's gonna, that's gonna increase. So, you know, then coming back to the, the question, you know, it's really about the, the breadth and the depth of value that can be created with GenAI, um, and if you believe that we're just at the beginning of, of this curve of not only adoption but technological improvement, right? I mean, cloud in a sense, you know, while there were improvements to what you get from the cloud, the cheapness of it, the scalability, et cetera, you know, to me it felt like more of a binary shift, right? Like, you go from on-prem to cloud and you realize a lot of those benefits from that switch. I think with GenAI, we're gonna see more and more use cases get, uh, you know, get powered by AI, right? Um, and it's gonna attack all these different functions, industries, et cetera, um, piece by piece. And so the, the total amount of disruption in that is, I think just a lot bigger.

   3. HSHarry Stebbings8:44

      So if we think about piece by piece in enterprise, you know, you speak to some of the biggest enterprise leaders in the world, what are the commonalities, I guess, in what enterprises wanna w- achieve through the implementation of AI and GenAI, do you think? What are those commonalities?

   4. HLHowie Liu8:59

      For sure. So I think it's super early. I mean, I've, I've spent a lot of time talking to the, the C-suite of Fortune 500s about AI and, and, uh, you know, Airtable has some AI capabilities that we're excited to, um, uh, you know, to get into the hands of both enterprise and, and self-serve customers. Admittedly, I think, um, one thing I've learned is, is just that it's super early. I think, um, a lot of customers are still trying to, to figure out what you can do with AI. So to some extent, I think we're still limited by the broader, um, understanding of what GenAI is capable of, what its affordances are. So, you know, what are its limitations? Obviously there's, you know, hallucinations and, and accuracy issues that are a major challenge, right? When you're trying to, you know, implement a, you know, a use case that requires a, a high degree of accuracy, uh, for instance, if you wanna use retrievals use case where you're able to ask, uh, you know, a smart AI bot, uh, questions about whether it's HR benefits or, you know, wealth management information, these are things where you wanna have it be right, you wanna have it cite the sources, um, and you want it to, to give you a good answer too. You know, there, there's all these limitations for sure, and there's also just a, um, you know, we're, we're still in the education phase, uh, where I think every enterprise is, is getting, uh, savvy as we speak or trying to get savvy as we speak, in terms of what, what is, you know, what is an LLM? You know, what are these basic primitives, right? You know, I'm hearing the words even factored database and, and other, you know, kind of technical, uh, concepts, uh, be talked about in, in, uh, in the enterprise. And to me that's really exciting because I think once there is this baseline level of understanding of what exactly this, this technology is and what it's capable of, I think that's when the real fun will begin, where we'll see enterprises actually get really, um, uh, smart about applying this powerful technology to specific problems that they have.

   5. HSHarry Stebbings10:54

      So I'm in Europe, Howie, uh, and, uh, we were chatting about kind of how shows have changed over time and how I've changed over time, I'm much more honest, we're still like, decades behind a lot of US enterprise and people are gonna kill me for this but I don't care. Um, (laughs) uh, but like we don't have Slack in a lot of large enterprises. Like, w- you know, I, I have friends who are CEOs of 50 to 500,000 companies, they don't know what Slack is, um, let alone Airtable. And then if GenAI and AI is the next one, my question to you is like, how far away do you actually think we are? And is this tech bros getting a little bit excited too soon?

   6. HLHowie Liu11:32

      One difference between GenAI and, uh, you know, traditional enterprise tech is that GenAI is not gonna be confined to just enterprise deployments, right? In fact, you know, the, the way that we've all become, uh, aware of and, and, you know, it's become top of mind, is through the consumer applications, whether it's ChatGPT, Midjourney, you know, these products ha- have gotten real scale, right? I mean, compared to any consumer product like Instagram or Twitter, et cetera, um, the growth curves as, as we all know, are, are pretty profound. And I think to me one of the interesting things has been...How broadly, um, you know, gen AI has touched, uh, consumers. So, it's not just the, you know, Silicon Valley elites who are adopting this product, but, you know, I've- I've come across, you know, just really surprising, um, you know, kind of friends, family, uh, people that you wouldn't otherwise think of as super early adopters, especially of AI, you know, finding interesting applications of- of, uh, of AI, right? I have a friend who, uh, has a dad who's a lawyer, um, and, you know, not- not in tech law actually, um, just in kind of traditional, uh, you know, kind of personal law, and not on the West Coast either. And- and yet, uh, you know, this- this dad, uh, has gotten really into using ChatGBT to- to help with, um, uh, you know, kind of triaging legal cases, you know, doing research, et cetera. It's been really impressive to me, and I think a real testament to the nature of AI, which is, you know, by- by default it's a much more accessible technology, right? Um, I mean, gen AI especially, the whole idea is that you don't have to have technical domain expertise to go and build or train your own model. It just kind of works out of the box, right? Um, sometimes it's wrong, but- but it- it's helpful.

   7. HSHarry Stebbings13:17

      I'm just rolling with this one. It's the end of the day, and it's Monday, so fuck it.

   8. HLHowie Liu13:21

      Yeah. (laughs)

   9. HSHarry Stebbings13:21

      Um, employees don't want it.

   10. HLHowie Liu13:23

       Yeah.

   11. HSHarry Stebbings13:23

       I'm speaking, I- I'm, you know, I'm invested in several companies where I speak to the CEOs and I'm like, "Eh, are we using X? Are we using Y?" The teams are pushing back.

   12. HLHowie Liu13:33

       On what basis? 'Cause I- I've heard it, I've heard all kinds of different, um, you know, perspectives on this, uh, from employees, including our own.

   13. HSHarry Stebbings13:41

       Media companies to gaming houses, where asset creation, where content creation, where graphic creation, they're just, they're getting rid of 90% of the team o- over time and replacing them with these tools, and the teams know this is a gradual transition.

   14. HLHowie Liu13:59

       Yeah.

   15. HSHarry Stebbings14:00

       What- what- uh, what happens when the employees don't adopt because they know this is a transition? It's almost like the screenwriters right now.

   16. HLHowie Liu14:07

       I think we have to, you know, get very, um, uh, you know, solemn and- and serious about the- the, uh, the very real economic implications of AI, right? Uh, and again, we're- we're still super early. You know, people are worried about, you know, the- the, um, uh, eventual, uh, super intelligence capabilities of AI and what that could mean for humanity, and that's a very valid and- and real concern. But I think even nearer term than that are the economic and- and disruptive, uh, implications, right? What happens if AI, in a much faster way than any other, you know, disruptive technology in the past, kind of, um, you know, changes the job landscape, right? I tend to be a believer that, uh, you know, AI has the potential to, you know, actually lower the cost of goods and services production, um, and therefore increase also the demand, right? Like, we can just produce more and better, uh, products. And I think, uh, while it allows us to leverage labor, labor productivity goes up, there's a potential world, though it's certainly not guaranteed, where we find ways to- to actually, you know, use more human potential and- and, uh, employ more people because each person is made more productive by AI, right? I mean, I, all analogies are quite crude, um, since nothing is quite the same as- as, uh, as AI, but, you know, if you think about the advent of personal computers, of course it completely disrupted, you know, some- some traditional knowledge work, right? Or work in general. Filing, you know, papers and cabinets, you know, like basic information retrieval and manipulation and document, you know, uh, editing in- in a typewriter format. You know, all of that obviously went out of the- the, um, out the window. Fast-forward a couple decades, and I think, uh, you know, there- there is more gainful employment than ever, right? That's, um, arguably enabled, uh, by the- the, um, you know, the productivity-enhancing capabilities of computing. So, I think we're in this weird transition phase where everything can h- Like, the- the risk is that everything happens very quickly before we as a society know how to kind of adapt. But I do think there is potential, you know, for- for, uh, for- for every person to- to figure out how to augment their- their capabilities and ultimately become more productive and- and yet more valuable because of- of this human and AI symbiosis.

   17. HSHarry Stebbings16:26

       For those that are educated in enterprise and those that are aware, what are the biggest reasons for them not to adopt? What are the biggest implementation challenges that they face? Why are they not adopting it if they're post-education phase?

   18. HLHowie Liu16:41

       We're candidly still very much in the education phase, in- in I would say, you know, the first half of that, uh, uh, of that journey. You know, keep in mind, th- this technology is so new in the- the awareness of, uh, you know, mainstream or- or, uh, if you don't think we're in the mainstream, then even in- in the, you know, the minds of, let's say, enterprise decision-makers, right? You know, I think everybody kinda had this wake-up call, uh, roughly a year ago, um, and- and, you know, the- the, uh, just the buzz around gen AI's capabilities, and I'd like to think it's not just because of an accidental excitement that built up because of- of the, uh, consumer-facing products, but also because the- the capabilities of the newest models just got really good. I mean, if you look at GPT-4's performance versus GPT-3, for instance, the, you know, in- in, um, in pretty profound ways, it's actually able to do things like, you know, pass the bar, pass job interviews, solve problems that, uh, you know, previous generations of models just did not do well enough, right, to be useful. You know, the- the technology has actually reached this breakthrough point, and we've only really had less than a year, um, as, uh, as a world, and- and especially within enterprises, uh, to- to keep up and figure out, what is this thing that even now, uh, public market investors are pricing into our stock, you know, based on whether we're gonna be an AI winner or loser? And that's not just for tech companies. You know, it's of course even for traditional companies, right? If you're a retailer, the ability to gain OPEX margin i- is, uh, is driven by, you know, uh, your ability or- or the, uh, perceived ability to implement AI, uh...... um, uh, and improve the leverage of your business. So I think first w- we are very squarely in the, you know, early innings of even that education phase. That being said, I, I think some of the other bottlenecks are gonna be, you know, around data privacy and whether enterprises are comfortable trusting, you know, these, um, uh, cloud-posted providers. OpenAI, for instance, uh, has an amazing model offering, uh, but right now they, they don't have a way for you to deploy in a, you know, self-hosted way inside your own infrastructure if that's something, um, that, that you care about, right? And of course, there are gonna be open source models that, that help bridge that gap. Um, you're gonna be able to adopt your own, uh, models, uh, that are pretrained. There's gonna be challenges around, you know, what is the nature of the training data, you know? Are there ... Is there copyrighted content on which this model is trained? Does it have the risk of plagiarizing content, you know, especially if you're producing content that's gonna be public-facing as opposed to just inward-facing? And ultimately, uh, you know, how do you get the accuracy, uh, and the safety of these models high enough to be useful for the intended application?

   19. HSHarry Stebbings19:22

       I, I totally agree with you and, and get you on all of those challenges. I think a lot of it also comes down to kind of handholding for the enterprise.

   20. HLHowie Liu19:30

       For sure.

   21. HSHarry Stebbings19:31

       I think they want it and despite what you said there about those challenges, they might try it. And w- we kind of, uh, chatted because I put a tweet out saying, uh, "The biggest companies building AI will actually be services companies helping integrate AI into large enterprises." Do you agree with that statement, that services companies helping integrate AI into large enterprises will be some of the biggest winners of the next few years?

   22. HLHowie Liu19:59

       I think services companies will definitely play a very important role. Um, for how long and how much of the cake they take versus application companies, your guess is as good as mine. Um, but I think especially in these early innings, the services companies are going to be the handholders, right? They're gonna help these enterprises both with the technical know-how of, you know, for instance, how do you implement a vector database? What's the right embeddings model to use? How do you chunk your content, um, appropriately for the intended use case? How do you do citations? Like, these are all technical implementation details that actually matter a lot, right, to getting a useful solution. And short of having, uh, you know, a, a, uh, SI come in and help you, you build it, or having a really, really robust, uh, platform or solution, uh, that comes out of the box with all of these capabilities, which, you know, candidly I don't think exists yet, I think enterprises are gonna need some help, um, you know, figuring this out. Otherwise, you know, I've, I've seen some, uh, you know, go off and, and do this, uh, themselves and it's possible, but it's also a lot of, of heavy lift, right? It's a very uphill journey to go and gain the internal technical expertise, um, you know, to, to do this.

   23. HSHarry Stebbings21:12

       Can I ask you, when we think about enterprise and startup or smaller company, I think it comes in, like, two different lenses.

   24. HLHowie Liu21:18

       Yeah.

   25. HSHarry Stebbings21:19

       One is in the, like, providers of AI themselves and incumbent versus startups, and then the other is in terms of the actually normal company world, big toy company, startup toy company as an example. If we start in the traditional tech world, who wins? Like, who does AI favor more? Does it favor Adobe or does it favor the next generation creative cloud company? Does it favor Airtable or the, you know, AI First Airtable from two months ago in YC?

   26. HLHowie Liu21:54

       One thought I have is it grows the entire pie. Um, when you think about, like, Microsoft's Copilot offering, it's gonna add significant ARPU to every single seat of office, which is a massive install base, right? So it's actually growing the pie in dollars, um, because it's creating new economic value in the world, right? So we're actually expanding GP by enabling people to be more productive, um, and, and creating a new market. You know, first, uh, I don't think it has to be purely a, you know, winner takes all, you know, Adobe wins and Canva loses type of equation, right? I think AI could actually enable both Adobe and Canva to grow and to actually create disruptive experiences that both deepen the, the value prop. I mean, Generative Fill, for instance, in Adobe is amazing, right? It's amazing for existing Photoshop users in a way that I think, uh, you know, I personally would be willing to pay a lot more money for. Um, so probably, you know, lots of monetization upside there. You know, some of the AI capabilities are also gonna be amazing from the standpoint of creating disruptive, uh, capabilities that, that, uh, that open these products up to new user bases, right, and even new use cases. And I think that's where, uh, you know, potentially the, the startups, um, or the, uh, the, the newer, uh, companies that have more ability to, uh, to lean into the tech and, and kinda build products, um, without the need to support existing customer expectations, distribution models, et cetera. Um, you know, when you think about, uh, products like, you know, Gamma or Tome, for instance, disrupting, you know, how you create slides, um, you know, I think there is a possibility that, you know, while Google Slides and Microsoft PowerPoint are for sure going to implement gen AI capabilities into their own products, that, you know, the- these, uh, these new upstarts, um, implement them in a way that actually goes after novel use cases. They're not competing for existing PowerPoint use cases, but they're actually going after completely new ways of sharing, expressing information that arguably are, are not even really about slides, right, so much as, as, uh, you know, a better way to communicate visually. I kinda think of it like Prezi which, uh, you know, I, I never personally liked that much as a product 'cause it always made me dizzy, the flying around the, the screen. Um, but what they were doing was not going after, you know, uh, traditional PowerPoints, but rather creating a new and, uh, an arguably improved way of, of, uh, communicating content in an engaging way.

   27. HSHarry Stebbings24:21

       Can I ask, in- in this situation, you'd be considered the incumbent, if we're being blunt. Are you able to move as fast as startups are? Everyone always says with incumbents, "Well, they can't move as fast. They have the distribution, but they don't have the speed." When we look at Adobe, speed of execution, Google, in many ways, speed of execution, for you as Airtable state, do you think you can move as fast as startups can?

   28. HLHowie Liu24:44

       Yeah. So the short answer is yes, and I think it's- it's sometimes a rational choice that every company... And- and, you know, I think it's really a gradient, it's a spectrum of how much of an incumbent versus an upstart you are. And, you know, like, in- in all spirit of, uh, humility, I would put us very (laughs) much still on the upstart, uh, side of the equation. You know, I think, um, any company sub-one billion in revenue is- is, uh, is an upstart compared to the enterprise titan. You know, whether you're a Salesforce or a ServiceNow or- or obviously, um, you know, the, uh, the big FAANG companies, we're just still operating at a much smaller, leaner scale than those companies. So, I like to think, first of all, that- that- that we do have a higher pace of execution velocity than a very, very large company. But I also think, um, you know, within Airtable, there is this conscious choice that we always have to make of, you know, we- we still have a finite number of resources, right? We may have raised over a billion in capital, we may have, you know, many more employees than we did when we started the company at two and then three and then, you know, five people, but there's always resource scarcity. And- and so I think it's- it's always this question of, okay, we could go and put, you know, 10 people, um, let's say, uh, on a completely new project that's building literally a separate product, separate code base. You know, we- we could build something new from scratch and then cross-pollinate that into our existing customer base, right? And, you know, it's a viable model. Some companies have done it. Or we can go and think about how to build, uh, functionality that is incremental to our existing product but still adds, uh, or changes the- the value prop of Airtable to our existing customer base, and maybe even unlock, um, our ability to sell into or get adoption from new customers. You know, it still creates disruptive value. So I think it's... You know, it's ultimately a choice of, you know, if we wanted to cut corners, launch something completely separate and new, we could move really, really fast. I mean, maybe some startups are able to- to, uh, take, um, even more aggressive shortcuts, uh, like, you know, not having to worry about security, uh, as much, right? Uh, or stability, SLAs, et cetera. Like, we- we would have to have a certain minimum level of expectation, um, that we can guarantee to our customers, uh, around those things, uh, which, you know, ultimately is- is valuable, uh, I think, for us. Though it does slow us down versus, you know, a truly, uh, you know, kind of, uh, cavalier new startup. Um, but I think, you know, when- when we're building stuff into the existing product, into... you know, for the existing customer base, I think it necessarily is a little bit slower than building it, you know, as a clean slate, cut all corners, new product entirely. Um, but- but you gain the advantages, of course, of- of, uh, you know, compounding the value that you already have. So, I think it's always a choice.

   29. HSHarry Stebbings27:30

       Howie, I had Emad at Stability on the show a while ago, and he said that, "Enterprise will adopt AI next year, and when they do, it will be a fricking train." If we break that up into two separate parts, um, do you agree that enterprises will adopt AI en masse next year, and how do you think about that timing?

   30. HLHowie Liu27:52

       Personally, I- I don't know if there's anything particularly special about next year. I mean, I think, um, every month that goes by, enterprises are getting more savvy about, uh, about AI. You know, I'm- I'm seeing it in my conversations, uh, with enterprises, you know, today versus three or six months ago. Um, there is this- this constant and- and kind of rapid, uh, progression of- of savviness, right? Understanding, you know, what are the types of use cases that we as an enterprise should deploy AI into first? Lower stakes, higher upside, easier to- to get, you know, some product into our- our, uh, our workforce's hands, um, and start experimenting with. I see this as more of a, um, uh, you know, kind of a agile or an iterative approach to- to, uh, productization, right? Uh, whether that productization is done by the enterprise themselves or with the help of an SI or, uh, you know, by startups, uh, or existing software vendors that are gonna provide AI-powered solutions into the- the hands of the enterprise. To me, it, it's- it's a, um... it's a feedback loop. And, you know, as we go and see what works, uh, in the enterprise, and Airtable is- is definitely, you know, leaning very much into this, uh, mentality. You know, we- we don't think we can predict exactly what the killer use cases are gonna be. I mean, there are some known- known use cases that- that, um, that are gonna be valuable, but we think it's too early to really know everything that's going to- to happen with AI and what are all the- the- the good use cases. And so, I think it has to be this, you know, this dance where, you know, we're gonna put out product value, uh, you know, in the form of both, uh, basic building blocks of AI. So, AI Field is one example, uh, at Airtable where you can compose a- a workflow using AI as easily as you could just add a new column, right? And then you can, you know, chain the output of that into an automation or put it into an interface. That's a very flexible and powerful primitive among, uh, you know, many other future primitives that we're gonna build. But we're also coming into enterprises and, you know, spending time figuring out what are this few starting point use cases, right? In marketing or let's say product management, um, that are great entry points for an enterprise to start using AI, right? What- what does that look like if an enterprise wants to, you know, start by trying to use AI in a product workflow, you know, to, uh, synthesize insights from customer feedback, to generate new PRD ideas, right? Um, to- to summarize execution progress, right?... you know, what does that look like? Um, and can we lead with a vision of, of product value and then test the market and, and get feedback as customers actually start adopting, um, you know, the, these solutions with AI? And then we learn from there, bam. So I think it has to be much more iterative over the coming months.
3. 30:52 – 43:08

   Business and Growth

   1. HSHarry Stebbings30:52

      are the hardest parts?

   2. HLHowie Liu30:53

      Going back to the, one of the early questions you asked, um, you know, I think it comes down to product go-to-market model alignment, um, meaning, you know, you can have a product that's amazing, uh, you know, for let's say individual users or small teams. And, and thankfully, Airtable was always built to be at least team capable from early on. But you know, let- let's just say like hypothetical product, uh, X. You know, product X is purely for individual usage. There's no team collaboration built in, at least from, from day one, and you get all this viral adoption or just all this, this, uh, top-of-funnel adoption for this product X. But, you know, at some point when you try to go into enterprise, you know, it becomes really tough to sell the value on X to a senior buyer within the enterprise, um, if its value is only experienced by individuals, right? And then you have to do a productivity sale, which is to say, like, you go and you say, "Well, like every person that uses X gets 10% more produc- productivity," right? Or saves two hours a week, or whatever it may be. Or they like it a- and you should pay more money for this, right? Or buy it for more people. And maybe that works, although I think in, in, um, uh, increasingly in this environment where every enterprise is trying to rationalize their tool set, right? They don't wanna have, you know, a million different products. I mean, some enterprises I've talked to, they literally have thousands of SaaS products and they're trying to consolidate down, uh, you know, 50 different collaboration tools into three, right? For good reason, right? It's easier than to develop internal know-how around that one product, right? Um, instead of having to support 20 different ones, um, and you know, you, you have one vendor really to manage, et cetera. I think, you know, for, for product X, you know, if you're coming in and, and if the value that you're selling is purely a function of, you know, "Well, we have, you know, X number of people who use us individually and there's no team or org-wide value that, that becomes more than the sum of its parts or that matters specifically to the executive buyer," I think that's gonna be a really tough position to be in if you're trying to transition into the enterprise, um, uh, and especially, you know, sold enterprise, right? Which, you know, save for Atlassian, I think every great enterprise product eventually has to, you know, be, be sold to a, uh, a senior, a strategic buyer, right, in the enterprise. Whether it's the, uh, head of IT for ServiceNow selling ITSM or, you know, initially for, for Salesforce, head of sales, uh, now head of support, et cetera. I think you have to define that senior buyer who really has a business case for, for adopting your technology.

   3. HSHarry Stebbings33:28

      Can I be a dick? Um-

   4. HLHowie Liu33:30

      Yeah. (laughs) It's your show.

   5. HSHarry Stebbings33:32

      Everyone talks about... E- well, true, but I'm a Brit so I feel very uncomfortable-

   6. HLHowie Liu33:35

      (laughs)

   7. HSHarry Stebbings33:35

      ... asking quite direct questions. Everyone talks about like the bundling of CFO purchasing decisions. How do you think about that? Would Airtable be a bundling or an unbundling? Like, would you be vulnerable to a Google Suite-

   8. HLHowie Liu33:49

      Yeah.

   9. HSHarry Stebbings33:49

      ... or would it be in favor of you? I, I don't really know the answer there. And, uh, and is it true that there is some bundling happening?

   10. HLHowie Liu33:55

       So I think bundling is going to become, uh, really, uh, i- i- you know, it already is very important, um, for products that fit into that core productivity suite. So if you, you know, if, if you're a product, um, that generally provides very horizontal, uh, but in my mind, shallow val- or, or commoditized value, meaning, like, there's a lot of different products that do the same thing, right? Um, so certainly if you're doing video conferencing, there's, you know, Zoom is great, um, but there's also other products that offer the same thing, whether it's Teams or Meet. You know, there, there were, there were many other products like BlueJeans, et cetera, that, that predated, um, you know, the, the current er- uh, uh, products. And I think, um, for those products, or if you're doing whiteboarding or if you're doing any kind of like freeform or, or, uh, very generalizable document editing, um, I think those products are screaming to be bundled, right? Because they, they're, they're very broadly applicable. Every company wants some form of it. Maybe not for every employee. Like d- does everybody need video conferencing? Yes. Does everybody need whiteboarding? Maybe, maybe not. Uh, unclear to me at least. But either way, you think of it as an aggregate decision that you make across the entire company, right? So the CIO can say, "Look, I'm gonna go with Teams and not use Slack. Um, I'm gonna go with, you know, Office. Um, and I'm gonna buy it all together as a bundle and not necessarily need every single specific product that otherwise would be unbundled." Because the benefits of each of those products just are not that differentiated, right? Like, I need one big bundle. It's, you know, uh, it's really about finding simplicity and cost effectiveness. And if some of my users would have preferred XYZ, you know, uh, let's, let's just say like document editing product instead of Word, too bad because it's close enough and I don't hear a, um, a strong enough argument pro- for the business impact of, of using one or the other, right? Um, I think that's different from making an ROI sale, which is by definition about differentiation, of saying this is not a commoditized product. But in fact, you know, in our case, we're saying Airtable can uniquely solve these end-to-end process implementation problems, right? We are an app platform. We're a differentiated one in that we're much faster to build on and deploy than say a ServiceNow or a PowerApps. You know, this is not spreadsheets. You're really building apps, right? And these apps, um, actually create significant business value whether it's, you know, powering the content production ERP at, you know, many of the top media companies out there or powering end-to-end marketing supply chain use cases at, uh, really large marketing organizations, right? Some of the biggest-... uh, brands or retailers out there. Um, you know, those are use cases where, you know, we actually are creating a lot of differentiation, we're creating real impact that we can quantify as, you know, what is the value of being able to ship content faster with fewer mistakes, reuse more assets better? I mean, that's very, very high business ROI. Sometimes the cost of Airtable licenses is, you know, a, a ve- very small percent of the business impact that we can have if we deliver on that end-to-end, uh, use case. So, I think as we move upmarket, it allows us to fight the bundling effect by creating differentiated value that can be sold in, in a business ROI, uh, frame to, to a single buyer.

   11. HSHarry Stebbings37:24

       When you're selling into enterprise today, to what extent do you lead with, "Hey, we have all these cool, sexy AI features that make Airtable great," versus, "We are a foundational, uh, tool that you will use, and we have AI integrated."

   12. HLHowie Liu37:39

       So, candidly, I've, I've tried with and it doesn't work. While there's a lot of excitement, uh, in, around AI in general, and, and, you know, maybe you'll get conversations that, uh, are just more exploratory in nature, right? I think there's plenty of, plenty of enterprise buyers who, you know, would love to, to just have, like, an informational conversation about where they get educated on, like, AI capabilities, right? Especially if we, we have expertise around it. You know, interest and excitement alone don't close deals, right? Um, a real business case, uh, real justification of budget, of value, uh, closes deals, right? And, you know, we're not in the business of just creating excitement or hype. We're in the business of, you know, delivering real value and, and monetizing that value and then building a business. So, what I thought is, I think it can be a conversation opener in the sense that, you know, if we come in and say, look, like, uh, you know, AI has the ability to disrupt, uh, you know, a lot of key pieces of this specific, uh, use case. Let's say that marketing supply chain end-to-end process, right? Or it could be product operations, right? We talked about, you know, collecting feedback, syn- like, synthesizing insights from that, generating PRDs, helping with every stage of the end-to-end process of coming up with product ideas, executing on them, and tracking all of the above. So I think when we come in and we, um, we tease a point of view of how AI can specifically help in a certain use case, I think that's a lot more, um, interesting and actionable than just talking about AI as this very abstract, you know, thing. Because then we're, we're back to square zero of enterprises are still trying to figure out what is this AI thing and what can we use it for. So, we are, we are nowhere near, in my opin- opinion, the tornado of every enterprise just knows they want AI in, in, um, in heaps, and is very ready to go and throw their own resources at deploying AI into every corner of the company, right? Uh, maybe that'll happen, maybe the train will arrive, uh, at some point, but it's my, my sense, uh, from my conversations that we're not, we're not close to that yet.

   13. HSHarry Stebbings39:51

       I think the hard thing for me is I meet a lot of AI companies, obviously, as my role and as an investor, and they have these massive logos, your Walmarts, your Nestles, your Coke colas. And you're like, "Wow, fuck." But truth be told, they're two seats in one department, in one off of... And actually, it's 10K spend for these enterprises for them to test, see, wha- what, it means nothing. I guess my question is, to what extent do I start taking those logos seriously?

   14. HLHowie Liu40:19

       In my mind, uh, you know (laughs) , in, in my crash course on, uh, enterprise that I've gained over the past 10 years of, of, uh, you know, of, of, you know, doing the PLG thing and then shifting into an enterprise, uh, centric model, I think a million dollar logo is really the, the, um, uh, the threshold of being kind of a real enterprise account. Now, you know, you could argue like there's a earlier safe milestone before that that's still meaningful. Like, you know, I, I would think of even like 250K, 500K as a meaningful milestone. Uh, but the truth is, what I've come to appreciate is that anything less than a million is really a speck of sand for a really large enterprise, right? Um, and frankly, even a million is just the starting point. I mean, you've got really large enterprise vendors like, uh, ServiceNow or Salesforce that literally are doing multi-deca million or even $100 million deals with, with some of their largest enterprises, and that's, that's where you're really powering a really critical part of, of the, um, uh, of how the enterprise runs, right? You know, to me, there's always a bigger fish, but, uh, uh, at least the way my, uh, my thinking is now, I think a million is, is kind of the real threshold at which you're a, you know, you're a, uh, a, a truly battle-tested or value, like significant value-delivery, um, enterprise vendor.

   15. HSHarry Stebbings41:47

       How long did it take you to get in a million ARR contracts?

   16. HLHowie Liu41:51

       I think it happened fairly early for us, um, I would... Well, I think probably 2019 roughly, so maybe a year after we got our unicorn valuation. Um, you know, that was four years after launch, admittedly, and we had gotten a lot of PLG adoption. Uh, we were well at the, the, you know, mid to high tens of millions in, in revenue, I, um, I believe. Certainly, it wasn't like we, we got a million dollar contract right away, e- especially given the nature of how our product was adopted. Like, we got there because we had this groundswell of organic momentum within these larger enterprises, and by then, you know, there were many Fortune 500s that already were using Airtable to the tune of thousands of active users who are easy to monetize.

   17. HSHarry Stebbings42:35

       Final one before we do a quick fire. I had Henry as an info on the show, and he said, "In the good times, everyone just-"... renewed and put more seats on. Happy days. And now, what-

   18. HLHowie Liu42:46

       Oh, that sounds lovely.

   19. HSHarry Stebbings42:47

       Yeah. And now, not only are they not adding more seats 'cause the teams aren't growing, but they're also wanting data to prove ROI, to prove usage, to prove value. And it's actually changing how he structures his teams b- they have to be much more CS heavy, uh, in particular. Uh, do you agree with that and are you

4. 43:08 – 46:41

   Insights and Reflections

   1. HSHarry Stebbings43:08

      seeing the same?

   2. HLHowie Liu43:09

      You know, we touched on the, the tool rationalization before, and I think it makes sense. I mean, in a way it's, it's interesting now, you know, uh, when, when you operate a company at scale, you start to realize how much macro plays into enterprise behavior and therefore your, your own execution behavior, right? When interest rates were low and every company was, um, you know, going on a little bit of a binge of, uh, spending across the board, um, but including, you know, software where, you know, and maybe there was a rational argument for it during COVID, every company had to digitize their workforce really quickly, shift into remote work, figure out how to enable c- you know, enable their workforce to be effective even when they weren't in office and they had to disrupt how, how they worked. Um, so that, you know, there was kind of this massive groundswell of adoption and just, you know, dollars being thrown at a lot of different products, right? And I think now, there is a very understandable rationalization of, "Wait a second, we kind of had to go during the mad rush over the past few years of adopting tech, spending on tech, um, and also, you know, less budget sensitivity. We sh- we, we, we adopted all these products and now we're feeling, you know, a little full. Uh, we- we've gorged ourselves and we have to go in and, and, and understand what product is addi- adding what value, which ones are duplicitous, uh, or duplicative, and which ones are actually meaningful and, and need to stay." I think the thinnest or the shallowest, um, analysis that enterprises are doing is that active to usage... uh, or sorry, active to paid ratio, uh, calculation. So how many of our paid seats, if you have a hundred paid seats, are we using 50 of these actively? Um, and which departments do those occur in? Do we, you know, does this department really need XYZ product? Um, so I think that is one of the, the analyses. But I think the, the, the harder work is actually saying, "Well, even if they're active or not, you know, are we getting business ROI from this product," right? You could have, let's just say, a c- a collaboration product or a chat product that has very high activity, and yet, you know, maybe arguably is actually costing you in terms of productivity because everybody's just kind of going and, and, uh, using the product. Um, but is it actually enabling real and better work results to get done, right? There's this, um, uh, funny, uh, quote I heard from, uh, from one enterprise CIO, which was, "Look, if, if I added up all the supposed time savings that every collaboration product, uh, you know, claims, you know, my, my employees would have negative hours of work to do every, uh, every week. You know, like save five hours with this tool, save 10 with, say, five. You know, it adds up to more than a full workweek." So the math can't possibly be true, right? Um, and so I think instead the more that you can ground into business outcomes, um... And, you know, for us, that means when we talk about the product operations use case, you know, we're really thinking about not just are, are, you know, teams using Airtable, but is it ultimately enabling product organizations to build and ship better products faster, uh, with fewer mistakes and with better visibility? And what are the best measures for us to, to show that to the customer rather than just tell, right? So I do think it requires a different mode of thinking, especially from the early PLG collaboration and active usage only centric world, uh, to how you do, uh, you know, kind of post-sales engagement.

5. 46:41 – 56:43

   Quick-Fire Round

   1. HLHowie Liu46:42

   2. HSHarry Stebbings46:42

      Howie, I could talk to you all day, but I do wanna move into a quick fire.

   3. HLHowie Liu46:45

      For sure.

   4. HSHarry Stebbings46:45

      So I'm gonna say a short statement and you give me your immediate thoughts. Does that sound okay?

   5. HLHowie Liu46:49

      All right. Yeah, let's do it.

   6. HSHarry Stebbings46:50

      So you're also an angel. Uh, you can put all of your money in one angel investment of yours. Which one do you put it in and why?

   7. HLHowie Liu46:57

      It is way too early to go all in on, on any one investment, especially in the frontier of AI, which is, you know, personally what I'm very excited about. So I love the fact that, you know, I can get involved with a lot of different really interesting founders and companies 'cause I don't know who the big winners are gonna be. It's, it's still so early.

   8. HSHarry Stebbings47:16

      What's been your biggest lesson from angel investing?

   9. HLHowie Liu47:19

      You know, I think, um, it's, it's that everything looks easier from the outside. You know, when you're a startup founder, it, it can feel very lonely because you feel like, "Oh man, everything's so hard and there's so many things we have to figure out." And when you look at other companies, and if you're only reading what they put out on, you know, press releases and the big announcements and so on, everything looks like it's just so easy. And I think when you come to appreciate when you invest in companies and you're kind of plugged into the community, it's like e- everybody has, um, you know, got, got, uh, challenges, right? Like every... It's never this perfectly smooth sailing, um, the whole way at least, right? Maybe there are easier and harder phases. Um, but I think I've just come to, uh, both, uh, empathize with, uh, with every other founder going i- and figuring things out, especially from the early phases. Um, but also normalize that, that experience for myself.

   10. HSHarry Stebbings48:07

       How did you get Phantom to write a series C check? He's incredible, but he writes series A checks.

   11. HLHowie Liu48:13

       I think we, we were at a very unique inflection point as Airtable and, uh, believed and still believe that there is a massive, uh, market ahead of us, right? And so, you know, I think what really matters in venture is, you know, what is the upside, what's the IRR potential for, for this investment?And, you know, if you're in a market where the biggest you can ever be is, you know, $100 million revenue business, right? And then you're gonna plateau, you know, it's hard to come in and- and get excited from an early stage returns profile standpoint, you know, if you're already investing at, you know, basically a valuation where that's priced in. Um, you know, maybe you get a 2X, 3X on it. When we, uh, when we raised that round, um, we were very much, and still are, at this early phase, this point of inflection, where things were starting to work, we were starting to monetize. We were seeing a very quick ramp on the revenue growth curve. I think we went from, you know, what to $10 billion in revenue in- in, uh, a little over a year, uh, where we're be- well on track to multiple 10s of millions. Um, and I think just the- the nature of the adoption, you know, we had everyone from talent farmers, to law firms, to non-profits, to agencies, but also really large enterprises starting to adopt Airtable and paying for us. So it just felt like a very large ... uh, that, you know, could easily or could potentially have 100X even from- from that point.

   12. HSHarry Stebbings49:38

       What's the biggest piece of startup advice that you hear that you think is bullshit?

   13. HLHowie Liu49:42

       It's this idea that, you know, if you just find product-market fit, all else is- is gravy, right? I think, like, product-market fit is just the beginning, and there are so many more hard parts of- of building a business. Now, for sure, like, you- you shouldn't have to worry about, you know, like, setting up, um, you know, kind of every, uh, bit of, like, legal infrastructure before you've found product-market fit. I feel like there's this, uh, you know, mentality or myth that, you know, i- i- the hardest part is that first phase where you're chasing product-market fit, and once you see the takeoff trajectory, you know, that- that's, you know, everything's a downhill battle from there. And, in truth, uh, you know, at least in my experience, like, that's when the real fun and learnings, but also the challenges, really start to emerge, like, to build a real business and to scale it.

   14. HSHarry Stebbings50:32

       What do you know now that you wish you'd known when you started?

   15. HLHowie Liu50:35

       I think it's really about, um, you know, thinking o- on, uh, on- on longer time scales. Uh, you know, we were always a very patient company. Airtable took two and a half years to build the product before we even launched. But what I wish I had done more of was not only have that patience, be willing to think on, you know, 10-year-long time horizons, but actually be more measured about holding ourselves to certain goals along each of the, you know, at- at milestones along that way. So, um, I remember, I don't know if this is true, completely true or not, but I remember, uh, you know, LinkedIn supposedly had put out a business plan before they started, um, I think Reid Hoffman had wrote- wrote, uh, written this- this, uh, this plan. And, you know, it had, like, M-A-U counts and revenue goals and- and, uh, and just in- in full detail, um, and in exactly the right chronological order, um, you know, they spelled out, "Here's what we're gonna be five years in. Here's what we're gonna be seven years in." Like, the MBA's dream of a business plan, where it sets, like, the next, you know, ten years of- of growth expectation and, you know, product execution expectation in. And usually that never works, right? Like, you- you, when you try to predict the path of a company, it's like, everything's volatile. You- you're not, you know, there's so many things out of your control. But apparently, in their case, it actually worked surprisingly well, right? Like, they were able to hold themselves to it. So, I think some of that is, um, in a good way, a self-fulfilling prophecy, where if you hold yourself to, you know, an expectation of, "We have to be able to figure, you know, or get to this milestone, uh, by X date," right? If you're SpaceX or if you're Tesla getting off the ground, you can't just say, "We'll figure it out when we figure it out," right? You have, you're burning cash. You have limited time to- to get a product out the door. You have to make that product good enough to get sales, and then you have to follow that, you know, in the case of Tesla, first product out with a second product that's not a loss leader. I think that kind of discipline could be applied more to, uh, to even software companies. And- and so just holding yourself accountable to this longer term plan of, you know, what is the sequencing and at what point do you need to make that transition, for instance, from PLG to enterprise sold? What does that entail? What are the revenue milestones, et cetera, that- that you could hold yourself to?

   16. HSHarry Stebbings52:51

       Do you feel pressure about scaling into valuation? Your last valuation was $12 billion. (laughs) It- it is high. Um, do you feel pressured to scale into that?

   17. HLHowie Liu53:02

       I feel pressure to drive durable growth in this business. I think valuations are, um, you know, are always, uh, you know, an outcome metric, right? It's- it's, like, the m- like, it's the output metric that is a function of all of your execution efforts. And to some extent, like, you know, if you focus too much on valuation, you're trying to chase the tail and not the dog, right? It's hard to directly impact valuation. I mean, maybe you could do a better job of pitching the company out there and- and that- that, uh, that helps. But I think the- the hard work that goes into building a great business takes a long time, right? You know, Airtable, at least, was a company that, you know, we spent a lot of time building our initial product before we even raised our first round. We then, you know, spent a lot of time getting product-market fit before we went and raised that unicorn round, right? You know, this was not a, "Let's go and get the valuation first and then justify the business after." You know, we very much focused on trying to build a really great product and a business, um, and that- that's what we're doing now, right? I think to some extent, everybody has- has kind of, uh, uh, you know, kind of had a- a reset in terms of, uh, valuation, expectations, revenue multiples, like, what's normal in this new era. Certainly, like, if we do well at executing on the durable growth playbook, we figure out, you know, the, how to sell differentiated value into the enterprise and do it repeatably, um, you know, there- there will, you know, for sure be a break-even point where we are worth our- our previous, uh, valuation and more. You know, the great companies are the ones that- that durably compound. And no matter what the revenue multiple is, right, revenue multiples can compress from, you know, 20X to 10X to 5X. You know, maybe they expand during, uh, you know, certain phases. Like, right now, we're seeing a bull market, especially around AI.... uh, you know, related companies. But I think the one thing to focus on which is in your control, whereas those multiples are not always, um, especially when they're driven by macro, I think it's- it's really focusing on that durable growth, you know? If you can consistently grow, um, and e- efficiently grow, uh, inevitably you will, you will justify any valuation that you pick, right? It's- it's just easier said than done.

   18. HSHarry Stebbings55:17

       Listen, dude. I wanna do one final one. Uh, and it- it's a very simple one. What would you like to be remembered for as a leader?

   19. HLHowie Liu55:26

       You know, I think I care less about being remembered, um, and instead about doing a good job, meaning, you know, I care about, uh, our- our team, um, as a constituency. I care about, you know, doing right on behalf of the- the company's shareholders, you know, having a fiduciary and- and also, like, a moral duty to, uh, to build a really good, uh, business and company. I think ultimately, you know, what I've learned is there's much more to this than just building a- a product, right? You know, it's- it's fun and easy in some ways to- to build a good product. I think to build a great company, um, and, uh, you know, and one that, you know, is not only a great business but also is, uh, you know, a place where great things are built or are done, um, I think that's- that's a very different challenge. And so I am learning to become a company builder. And while it's- it's less about having a legacy attached to my name around that, I think it's something that- that I, uh, want, you know, I- I hold a lot of, uh, you know, both pride and accountability around getting good at.

   20. HSHarry Stebbings56:33

       Howie, listen, I absolutely love this. I always so enjoy our conversations. Thank you so much for putting up with my more direct style this time around, but I've loved it and you were fantastic.

   21. HLHowie Liu56:42

       Thank you for having me.

Episode duration: 56:43