Airtable Co-Founder and CEO Howie Liu

Post on

June 14, 2023

In this week’s IA40 spotlight episode of Founded & Funded, Investor Sabrina Wu talks with Airtable Co-founder and CEO Howie Liu. Airtable is a low-code platform that enables teams to easily build workflows that modernize their business processes. The company launched in 2012 and has been on a product-led journey since then. Last year, Airtable ranked number three in the growth stage section of the intelligent applications 40. And just in May, the company announced new embedded AI capabilities to make it possible for teams to integrate powerful AI into their data and workflows.

In this episode, learn about Howie’s transition from a first-time founder to a second-time founder, the lessons he took with him from that journey, and how he decided to go up against the dominant forces in the low-code productivity tools space when he was only a few years out of school. As Howie explains it, to be a founder, you really have to have the perfect balance of naivety and pragmatism, but you’ll have to listen to hear his explanation.

This transcript was automatically generated and edited for clarity.

Sabrina: Hi everybody — my name is Sabrina Wu, and I’m an investor at Madrona Venture Group. I’m very excited to be here today with Airtable CEO and Co-founder Howie Liu. This is a particularly exciting conversation for me because I am a huge fan of Airtable, and I would bet many people listening to the podcast today also are, and if not, people have some homework to do to go check it out. So it’s been a lot of fun for me watching the progress and the growth and seeing how many use cases have really emerged over the years, and recently with the launch of Airtable AI, which we’ll spend some time digging into today toward the end of the podcast. So Howie, congrats on the success, and welcome to the Founded & Funded podcast.

Howie: Thank you all. Thank you for having me, Sabrina.

Sabrina: Howie, I’d like to start by going way back. So you’re not actually a first-time founder. In 2010, you founded your first company, a company called Etacts, which was an intelligent CRM. Etacts later sold to Salesforce, I think about a year after the founding, and you spent about a year at Salesforce before leaving to found Airtable. It’d be great if you could share with the listeners the journey of deciding to become a second-time founder. What made you decide to take the jump again, and what was the original inspiration behind Airtable?

Howie: So in many ways, I see Etacts as a warmup act to Airtable. As the first-time founder, I really didn’t know what it was like to start a company — I worked on some small web apps, but nothing that was really formal. And Etacts was the first company that I co-founded where we actually went out and we raised some money, went through YC, hired some people, launched a product, got some real traction, and, near the end, even turn on monetization on one of our features. It got some small amounts of real revenue. But in many ways, it was trying to do all of those things for the first time. And not just the first time as a founder, but even the first time as a product operator. This was the first job I had really meaningfully out of college. So it’s not like I had built great products before, knew how to scale them, etc.

So I think, in many ways, I had to learn as I went along and was able to apply a lot of those learnings to the second time around with Airtable to do things with more of a deliberate approach. If I could characterize that first company Etacts as just trying to figure out what we’re supposed to do at every part of the company. With Airtable, what we wanted to do was start with a lot more conviction of here’s the opportunity, almost like create effectively a business plan and a roadmap, and have more forethought of if we build this, what’s going to happen? How are we going to validate every step of the way?

And in fact, the time that I spent at Salesforce after being acquired by them directly inspired a lot of the ways that we thought about Airtable — both the massive opportunity of if you could democratize the process of building business apps and distill it into these elegant building blocks, which in a way Salesforce did, but just to a very different side of the market, much more complex, heavyweight applications. You build it on Salesforce and it’s a really great platform for that. But with Airtable, we saw this opportunity to actually disrupt that and democratize the building of apps. So actually getting to see from within Salesforce, here’s what it’s like to build a great app platform to take it to market to many different industries and use cases, etc., was definitely a direct inspiration to Airtable.

Sabrina: And I think one of the things that I really have found fascinating by Airtable is that it makes it easy for all users regardless of how technical they might be. I think you used the word no-code in terms of being able to build applications in a really powerful way without having to write the code. But when you founded Airtable in 2012, I think the idea of taking on somebody like Microsoft, who has predominantly been dominating the productivity software market for decades, must have been a scary concept, especially as you noted maybe a couple of years out of school. So I’m curious, how did you think about creating a new collaboration tool? Can you tell us about the journey of tackling this problem and some of the challenges that you had along the way?

Howie: First of all, I think philosophically, when I reflect on the journey of being a founder, I think you have to have this perfect balance of naivety. So you actually think that you can do something as bold as take on these massive giants, whether it’s Salesforce or ServiceNow or Microsoft or whatever, yet pragmatism. So you’re not just doing it in a completely unstrategic way. You’re finding a place where either structurally or otherwise, there’s a weakness or a gap where you can exploit it. For Airtable, when we thought about the productivity landscape, there was a lot of incremental innovation. If you look at G Suite and what they did with Google Docs and Google Sheets, etc., it’s really cool, but in my opinion, incremental innovation on the offline version of Word and Excel, etc. They brought it online. There was a lot of technical magic that had to go into creating real-time collaborative versions of those products. There was a technical thing they came up with called Operational Transforms that allows you to deal with all these real-time editing people online and handle all of their merge conflicts in a really seamless way. And yet, from the product standpoint, it didn’t fundamentally unlock completely different use cases of Excel or of Word. It certainly enabled more collaboration. It solved a lot of file saving and sending headaches, and yet it was still basically the same product experience.

And the opportunity that we saw for Airtable was that most people actually are using Excel in one of two very different ways. One is number crunching. And if you think about the original origin of the spreadsheet, Lotus 1-2-3 or even before that VisiCalc, it was like this glorified number-crunching tool for accountants — actually a computerized version of something that happened very offline manual, you would literally be crunching numbers by hand or with calculators. And yet, as Excel became more and more mainstream, I think people ended up using it as their makeshift database. They would come in and build customer lists, inventory lists, or even event or wedding RSVP lists.

So in practice, there was this split in terms of how spreadsheets were used. And the side that we wanted to take on was when people were using spreadsheets not as the number crunching tool for which it was originally invented, but instead as almost like a lightweight database workflow type use case. For those, we knew that we could do a much better job because we didn’t have to compete head-on. We didn’t have to go and recreate all of the advanced number-crunching functionality of a spreadsheet. We could just pick off all of these tabular workflow use cases and do a much better job of building a product that was, actually, at its heart, more of a database and act platform metaphor disguise or masquerading as a spreadsheet interface because we knew that would be a really intuitive way for people to just start using our product.

Sabrina: I think that’s a really important point that, at its core, Airtable is a database, as you just alluded to, and that’s one of the reasons why you can do and use so many applications and it crosses across a variety of different audiences as well. So I’m curious, how did the customers and maybe users surprise you in terms of ways that they leverage the product over time? I’m sure there are some really interesting learnings that maybe… Did you reprioritize or pivot how you thought about building the product over time? How did you think about that?

Airtable CEO Howie Liu: I think the starting thesis that you have for a product then often creates a self-fulfilling prophecy. Meaning because the product was very clearly designed for tabular use cases, we didn’t actually support any number-crunching functionality initially. You could not crunch numbers in Airtable if you wanted to. Now we have some ways of doing that, like formulas or you could create reports and so on. But when we started out, there was no way that somebody could use Airtable as a traditional number-crunching spreadsheet. For instance, when we got our early alpha customers and we discovered the use cases they were building, it was stuff like a nonprofit building programs management and donors management workflows. You know, being able to administer across many different locations their operations. And these are much more apt use cases that otherwise might be powered by something like a workflow.

And I think over time, as we discovered more of these use cases, we leaned into them and built more functionality to really enhance that. We built templates at some point once we discovered initial use cases that were bubbling up, we would go and templatize it. And especially early on, a lot of it was SMB-oriented or even consumer-oriented. So we would take all of the great usage that we heard from the community and make it easier for people to build that if they were to sign up later. So I think you start with this thesis, and then if you’re right or if there’s any inkling of being right, I think you start to see some organic growth around that. And ultimately, we just lean more and more into it.

Sabrina: As I mentioned earlier, I’ve had the privilege of using the product now for probably close to five, six years. And I think one of the reasons I really love the product just because of the Clean UI/UX, I mean, there are many different reasons, but it’s so intuitive in terms of how easy it is to use without any knowledge of coding or even any knowledge of how to use Excel. And I think that is a testament to your understanding of how to solve a customer pain point. When did you realize Airtable had this incredible product-market fit? How did you realize you had this conviction that you actually were solving a really big, important problem?

Howie: I think there’s almost two ways that you can go about finding scalable product market fit. And I would say the first way is maybe what Twitter did. You build something you’re not really sure where it’s going to go. You start with something small and you just get some traction. And if it works, you fan the flames and you build from there. And I think there are plenty of great companies that have been built that way. They start with something really almost seemingly like a joke or a toy that becomes something really, really big. I think for us, we started the other way, which is to identify a really large opportunity that, on first principles, just should be solved. And that opportunity for us was, we identified this need for apps in every part of every company. Functional apps, little apps that currently weren’t being built because it’s too expensive to go and build an app with code or even to go and take a very heavyweight solution and customize it to your needs. So instead, they’re getting solved by makeshift spreadsheets and documents and people emailing things and just not really having a very structured way of doing their work when they should be. So we did a lot of research on that space, and we came to a pretty high conviction that this should be a thing that exists. In fact, there were little glimmers of it in the past. Some of the earliest software products when computing first came to the fore were database products like Ashton-Tate’s dBase. There was Lotos Notes, Microsoft Access, FileMaker Pro, etc. So there were these glimmers of this opportunity. And, of course, on the large enterprise side of things, there were big platforms that solved this problem, but we really felt quite confident that there was a gap in the market for us to go and fulfill.

We felt like if we could just unlock the near-term product usability and the onboarding and the growth mechanics of the product, there would be a big light at the end of the top. It’s not like we’d built this thing and have no TAM. So really, we broke it down into the different phases of product market fit finding. Initially, can we design a prototype of this that actually is intuitive enough for somebody to immediately start using and building a real workflow, real app. So some of our early alpha tests were really designed to do that. It wasn’t about getting as many users as possible, it was making sure that we could actually solve real workflow problems for a small number of invite-only alpha customers. And then from there, and they get every step of the way, even when we launched on Hacker News and got maybe, I want to say, 10,000+ organic signups and started getting a trickle of additional signups on our waitlist — dozens per day from there on. Or once we launched publicly and got even more signups and even more daily organic signups, there wasn’t a single moment where it was like, okay, we’ve made it, this is a thing. It was more like at every phase, we were unlocking the next phase of growth. And we were figuring out clearly the initial product has value and some people are able to figure it out, but a lot of people get stuck because they don’t know how to build or what they’re supposed to build on the platform. So we have to do a better job of onboarding, we have to build more templates. We have to add more sharing functionality for the product so that once they’ve built something, it’s easier to collaborate with others in it. And I think along the way, we built up enough of these unlocks to actually continue sustaining growth. Initially purely from a bottoms-up PLG standpoint and later from also enterprise go-to-market standpoint as well.

Sabrina: I was going to ask about the go-to-market model because Airtable has this really interesting mix of the PLG bottoms-up motion where a user like myself can go on, try out the product, and test it out. And then maybe if enough people from my company come, then you could do more of this enterprise sales motion. But having two motions can sometimes be challenging or maybe has its own set of challenges. So I’m curious, how did you guys navigate that, and is there one that was easier to implement than others? Or in today’s day and age, everyone talks about how the PLG motion is the way to go because you get this long lead of customers, and you don’t have to do the top-down sale. What was your thinking around that? And what light can you shed to the listeners on that point?

Howie: I think so many of our decisions early on were made probably naively. But naively, we thought, “Hey, if we just build a really great product, people are, of course, just going to come and want to use it.” And there were a couple of prior arc examples of PLG companies at the time we started. Dropbox and Evernote were probably the most notable ones. Slack actually didn’t launch until after we had started going. So we started in 2012. I think Slack probably launched in 2014/2015, probably just right before we launched. So there weren’t that many great, especially B2B and team or even department or company-scale applications that had proven out this PLG motion.

So it was definitely early days, and thus very naive for us to assume it would work. And yet somehow, it did. I think, in this case, because we had a low enough barrier to entry for somebody to just pick up the product and start using it. And that was both a usability thing. It didn’t require you to learn this complicated manual to be able to build on Airtable. It was easy to get immediate value. So we really tried to front load how you get an MVP of a use case in Airtable up and running and have it be actually demonstrably better than the prior arc. Let’s say, using a spreadsheet or not using anything at all. So we really tried to front load a lot of the very easy-to-use yet powerful features like having rich text fields or dropdowns or even be able to visualize the content in a way that was not just limited to the spreadsheet grid.

And I think over time, the product funnel just continued to compound. So PLG took us a very long way. We got to tens of millions in revenue at the time that we went out and raised our series, our unicorn round. And this was I think the time when no-code, low-code was starting to become more legitimized as a category and also just the PLG engine was more recognized as maybe a plausible path to building growth. And yet I think one of the limitations of PLG is that from a product standpoint, sometimes you get stuck in smaller-scale use cases. It depends on the product, but in some cases, the mechanics of PLG and any particular product are that you get great bottoms-up adoption. But sometimes you need a little bit more of a push to actually consolidate a bigger data set. In our case, becoming a system of record for something really mission-critical and also becoming the way that an entire larger maybe department-level process as opposed to team-level workflow, is built. And sometimes those things do emerge organically. We were very lucky to see early PLG traction carry us forward into these bigger meteoric use cases within larger enterprises.

But what we also recognized is that we didn’t want to just rely on that organic momentum to bring us there. We wanted to go and start more directly engaging in enterprise-level sales conversations to get those higher value meteoric use cases because we knew that the real opportunity was not just to go and serve lots of little smaller fragmented use cases, but actually to scale up and raise the ceiling of what you can do in Airtable. So it starts small, but you can also grow into a true system of record — something that’s really, really powerful at a departmental or even company-wide scale. So we had to shift into a very intentional mode of execution, both for our product and a go-to-market standpoint to really move, it’s not just up market into larger companies, but really up use case into bigger and more valuable use cases within the enterprise.

Sabrina: That leads to an interesting place to pivot the discussion a little bit to talk about AI and ML because you can do a lot with different data sources across the organization if you’re able to connect in different pieces within product and marketing and sales and how can you enable and create this feedback loop. And I also don’t think you can have a conversation with a tech founder these days without talking about something related to Generative AI. So I know Airtable just announced Airtable AI. I’d love it if you could just tell us a little bit about what some of those features are. What are some of the embedded AI capabilities and maybe tie it into what you were talking about building on that concept of connecting different data sources within the broader organization?

Howie: So our approach to AI is that we think, first of all, the modern models, especially LLMs, are capable of really profoundly useful knowledge works. We’ve gone from maybe over a decade ago, AI being a very narrowly applied thing. If you got a large dataset, you could do predictive analytics. You could create a better recommendation engine. I think of the Netflix data size prize when I was in college as well, an example. We then entered the space where you could do really powerful machine vision, you can identify what’s in images. That was a big breakthrough. But I think now the big moment for LLMs is that they’re not just capable of outputting text in a certain stylized format or writing emails, etc. I mean, sure, those are some of the use cases, but I think we’re actually just scratching the surface where most people who have interacted with ChatGPT are just scratching the surface of how much deep reasoning and creative work these LLMs are already capable of.

If you imagine the product roadmap use case in Airtable, you’re coming up with feature ideas. Maybe those are informed by user research that you’ve done. So you can track both user research and a feature backlog in Airtable, maybe also the release marketing workflow. Every step of that process probably has multiple points into which you can inject AI. AI is not just for very superficial things, but actually really meaningful reasoning work. So I’ll give an example of in the user research tagging phase, you can take actually user research snippets or insights and have AI categorize each of those. We have an AI field where you can actually take any input from a record and then output something that’s basically prompted from the LLM. So in a way, it’s having a little LLM brain embedded in every single cell of Airtable as a primitive and taking whatever inputs you want from the localized data that you have in Airtable and then outputting it seamlessly in the context of that workflow. Another example might be, okay, now you have these insight summaries of user research for each feature. Now pull those together along with the high-level goals of this feature into a product requirement stock and actually generate the first draft of that. And it’s more than just stylistic formatting. It’s actually going and thinking a PM would, what does this product feature need? You’re Uber, and you’re trying to create a new feature — like a loyalty program — what should that entail? And so our goal is really to integrate LLMs into the context of your data, your workflows, and into our interface all with a no-code UX around it so that it just becomes another primitive in the toolkit you have to build apps. And ultimately, I think our thesis is that these LLMs are really powerful, but the real value gets exploited once you put them into the context of data and workflows. And that’s really what we’re all about.

Sabrina: I think that the point around being able to integrate directly into your workflow is a really, really important one. No one wants to leave their workflow to go look for an answer, go ask a question, find the output, or go to ChatGPT and paste it back in. So if you’re able to show that directly in your native Airtable workspace, then it becomes much easier. But I think one of the questions I have for you is how do you think about the UI/UX when it comes to that. That’s one of the big questions is there’s so much going on, h ow do you give the right outputs and continue to gain user trust as they are maybe using the product? These models can tend to hallucinate, for example, so maybe you get the wrong tagging, which may not be the end of the world and this workflow, but how do you think about some of those things to keep the user really engaged?

Howie: So first off, I think there’s been a lot of speculation that AI is going to obviate the need for traditional user experience design. Everything’s going to be replaced by natural language interface as the input, and then you’re just going to magically get the output that you want. It’s going to do all the work for you and perfectly hand it to you on a plate. And I think to your point, LLMs are very, very capable, and you can get accuracy up through a number of means, whether it’s fine-tuning, giving it a few shot examples, or just plugging it in with the right prompt and the right context. And maybe there’s some pre-and post-formatting tweaks. But I think, ultimately, we’re a long way off from having AI that’s so powerful that it can just do everything you want without human intervention. And I think the more powerful applications, at least in the foreseeable future for these LLMs, are going to be making the output of it very visible and interactive so that the error tolerance is very high. When I think about Copilot in GitHub as an example, it’s a really great application of AI because the worst that happens if it generates bad code is the human coder can just review that code and edit out the part they don’t want or change it. You can even have it generate 10 different examples of code and you can use that to inspire your thinking. And I think that’s the best way for LLMs to be used, especially in our context where Airtable is primarily an internal application builder platform. You’re not building external customer-facing use cases on Airtable typically.

So in the internal use cases, you don’t have to worry as much about some of these other issues like is the content output copyright safe? Is it appropriate? Is it going to hallucinate? Our goal is to, in the near future, deploy LLMs or encourage LLMs to be used in the context where the output can be seen by human very easily and edited. And it’s more of almost like a very, very advanced auto-complete step where it can generate the first draft of something, but there’s still very much an expectation that the human comes in. And this is, by the way, where the fact that Airtable is a very visible product. Everything in Airtable is very visible, the data is visible, the steps in a workflow are very visible. You can compose an interface. You can create fields that chain off of each other and the output of one AI field, then you can see before you go and pass it into a formula field or another AI field or trigger some action with it. The fact that all of it is very interactive in terms of the human, I think, helps in the cases where the AI’s output is not perfect, but can be usefully wrong or at least a good starting point. So I think it’s a really, really good call out, and it probably increases the importance of having really strong UX around the human feedback loop.

Sabrina: And then I’m just curious, can you share with us generally, we’ve talked a lot about large language models. There’s obviously been an explosion of new models that come out, seems to be that there’s larger models that are trained on even more parameters every day. There’s now these open-source models that are coming out. How are you guys thinking about the technology, and how are you building the infrastructure so you can maybe easily swap in different types of models based off different use cases, even when we think about different types of data types. Some models are better for structured versus unstructured. How do you think about the tech stack, and what does that look like?

Airtable CEO Howie Liu: We want to be fairly interoperable with any model and initially, it’s going to be LLMs, but in the near future, we’re going to do text-to-image and other models as well. And I think the idea is our strength strategically is that we have really good no-code UX to build apps. And with our existing customer base, we have good data and good distribution in the context of specific customers. So our goal is not to aggregate that data and train our own supermodel on that data. Our goal really is not to go and do anything particularly fancy or deep at the model layer, but really to be quite interoperable with any model.

Right now, we’re really focused on making the product experience very seamless with open eyes model. So I think GPT-4 is a really, really capable model that can do so many different things out of the box. And in many ways, that’s really important to us as a platform because Airtable is also uniquely horizontal. We have all kinds of use cases and almost every industry function company size, we’ve had cattle farmers doing cattle tracking in Airtable to lawyers doing case mapping in Airtable, all the way up to some of the larger scale enterprise processes we’ve talked about.

Sabrina: And I think you mentioned an interesting point there around data, and one question or concern that we’ve heard from some enterprise customers that I speak with is how they want to be able to leverage data into those models. They want to train it because, obviously, if you put in your own data, the model becomes smarter and is able to solve in more contextually aware ways. But with that becomes this question around data privacy and security. I’m sure you’ve thought about this, so I’m curious how the enterprise customers that you work with might be thinking about how they can leverage their data and make sure that the data that they have that’s proprietary to them isn’t fed back into the model. So if I’m using Airtable, I want to make sure that doesn’t happen to me. So how have you guys thought about that?

Howie: All of our offerings, by default, will not have data retention, so your data will not be used to train models. That’s going to be a really important default guarantee that we have, just so that you don’t have to worry about putting your most trusted and high-value data into your Airtable. That should be a given. I think secondarily, there’s still going to be a lot of different preferences within enterprise customers. So I’ve spent a fair amount of time talking to CIOs or CXOs at different enterprises, and I think every company has a slightly different stance on this, and I think it’s quickly evolving. I mean, nine months ago, probably most enterprises didn’t even have a strategy around LLMs and what are the LLM providers that we’re going to partner with or leverage. We need to train our own or use one of the open-source pre-trained models to deploy it on our own infrastructure.

It feels like the beginning of the cloud revolution where everybody’s trying to scramble and figure out what is our cloud strategy. I think the smoke’s going to clear a little bit in the next, call it six to 12 months, and there will be some stabilization of different enterprises falling into a few different buckets of preference. Some are going to want to have in-house, in private cloud deployed offerings, whether it’s something like the Microsoft managed offering or AWS Bedrock offering, etc. Others are going to be fine using OpenAI’s own offering. And our goal is to really be interoperable with as many of those different options as possible, including if an enterprise wants to post its own model. It’s our goal to figure out ways to be able to talk to those models in a secure environment and be able to give you the best of both worlds. So I think the landscape is very quickly evolving, and it’s premature to call where things are going to settle.

Sabrina: With the landscape quickly evolving, one thing that I think where Airtable has an advantage is that you have a large reach, a large customer base, and you have the distribution, and that’s what a lot of early-stage startups are looking for. But with that being said, I think there’s also a lot of innovation happening at a really fast pace. So I’m curious with all these new companies popping up that are built with large language model technology at the core, what keeps you up at night as it relates to AI/ML? How are you continuing to stay ahead of the curve and educating and making sure that you’re building Airtable and positioning yourself in the best way possible?

Howie: I think, on the one hand, rationally, I can say the LLMs are going to continue advancing at pretty incredible speed, even if not just increasing the size of the data sets that they’re being trained on, since we’re exhausting the number of available public domain tokens that we can use to trade them with. But even just improvements, for instance, to how to fine-tune them and improving the performance in specific applications, I think that’s continuing to advance at such a rapid rate. We’re going to see multimodal become a very widely available option for most of these models. And I think I miss it all. The rational thing we could say to comfort ourselves as Airtable is as long as data distribution and that UX of how you present that model, how you integrate it into a useful use case, remains valuable, we’ll still have a role to play. And we need to make sure that we’re keeping up to date on the latest advances in models, what are the new models we need to support, etc.

The paranoid version of me, which I think is similar to the dichotomy of naivety and pragmatism, I think you need a little bit of rational certainty and then also some paranoid uncertainty to always be on top of the game. I think the paranoid version of me says, “Well, at what point do the models become so disruptive that there’s a completely new experience possible of building apps?” And I want to say in the near future or even midterm future, I think again, the no-code UX is actually the ideal way to build apps with LLMs. And you’re still going to want, if anything, more UX around the feedback loops and the affordances for how people can build and then use these LLMs in practice. But I think we want to be very, very plugged in. And I’m personally spending a lot of time in the ecosystem learning from one of the most interesting and disruptive startups in AI, spending time at really every layer of the stack from app companies all the way to the LLM providers just to make sure we’re staying a couple of steps ahead of the game. It’s really exciting because, in many ways, it feels like in that entire world of AI, nobody really has a good census view on where things are going to shake up. And, certainly, in terms of where value will accrue, it’s really not quite clear. In many ways, it’s both terrifying and also very, very exciting because it’s like anything could happen, and we can’t fully even imagine what product experiences and business models will look like five years from now as a result of all these continued advances and compounding of AI capabilities.

Sabrina: Totally agree. Even as VC investors, we always say we try to predict what the future will hold and make bets based off that, but it is incredibly difficult to predict these days. But it makes it really fun, as you point out, to think about all the innovation happening at each layer of the stack. I think it’s a really fun time to be a builder. Just to wrap up here bit. We ask all of our intelligent application i40 winners a lightning round of questions. So going to ask you a few questions here. The first is, aside from your own, what startup or company are you most excited about in the intelligent application space and why?

Howie: I think there’s a lot of really interesting AI app companies that are finding some very specific use case built around AI. One example is Galileo. It’s a way to design interfaces with AI and eventually you can output either a Figma design or code. I don’t know where it’s going to go. I think the founders maybe are actually still figuring out in the big open-ended world of possibilities, where can you take this? And I think that’s actually part of the excitement. There’s so many different entry points of where you can apply an LLM and then build up all of the more specific product functionality and go-to-market execution to turn that into a real business. It’s a lot to be figured out, but I think it’s really cool to see a lot of these specific app companies go and try to find one use case to take and specialize in.

Sabrina: Outside of enabling and applying artificial intelligence to solve real-world challenges, what do you believe will be the greatest source of technological disruption and innovation over the next five years?

Airtable CEO Howie Liu: I think it’s hard to even say because AI itself is so big. In a way, it’s almost like, what are the different permutations of AI? And AI can be applied in both very top-of-the-stack ways. Like hey, let me take one of these LLMs and build a transformative consumer product experience or enterprise. But also, there’s going to be a lot of innovation around taking transformer model architecture and then training it with new data, whether it’s biomedical data or it’s self-driving car data, etc. So I’m going to give a non-answer, which is it’s going to be AI and every single permutation of AI — applying models to new use cases, applying existing models to more interesting consumer-level UX innovation. It’s all of the above.

Sabrina: Last question. What is the most important lesson likely from something you wish you did better, that you have learned over your startup journey?

Howie: I think the importance of moving quickly is not to be understated. In a way, Airtable benefited from being very thoughtful and methodical with our product roadmap and really the TAM and de-risking it. At the same time, every day really counts. And the more that you can start compounding your learnings, that doesn’t mean always go into hyperscale mode right away. I think it was actually a good thing that we took three years to build the product and launch it. We’re very intentional about our early days ,product-market fit finding before we turned on the gas of let’s scale this up.

All that being said, the more you can accelerate that rate of learning, and I see this in the AI space where all these new startups are launching and very, very quickly gaining user feedback, learning what works and what doesn’t work. And maybe not all of them will have durable advantages right away, but I think the faster they get out there into the market and learn, the better. Especially as the world starts accelerating in its pace of change, I think being able to learn very quickly and scale up that process as opposed to just focusing on scaling revenue or growth in traditional terms, I think becomes one of the most important core competencies as the landscape evolves.

Sabrina: Awesome. Well, Howie, this has been a lot of fun. Really appreciate you joining us today on the Founded & Funded podcast. Thanks again.

Howie: Thank you, Sabrina. It was fun to chat with you.

Coral: Thank you for listening to this IA40 Spotlight episode of Founded & Funded. If you’re interested in learning more about Airtable, please visit If you’re interested in learning more about the IA40, please visit Thanks again for listening, and tune in a couple of weeks for the next episode of Founded & Funded.

Other stories

Share The IA40
Copy link