Airtable CEO Howie Liu - Excel vs Access, Founder Mode, and the Future of SaaS

Hello, fellow data nerds. Welcome to World of DaaS. I'm your host, Oren Hoffman, CEO of Incubate and GP of Flex Capital. Discover more episodes, get weekly data as a service news, original content, articles on data, and more at worldofdaas.com. That's worldofdaas, D-A-A-S, .com. Hello, fellow data nerds. My guest today is Howie Liu. Howie is the CEO and co-founder of Airtable, a collaborative spreadsheet and database platform that has raised over $1.4 billion.

Wow.

Airtable is at the forefront of no-code, low-code movement. Before founding Airtable in 2012, Howie sold his first startup, eTax, to Salesforce when he was just 21 years old. So now we know how old you are now, if you can do the math. Howie, welcome to World of DaaS.

Thank you. Really good to be here, Oren. And I should do a shout out, which is we first met in the context of, uh, your prior company, Rapleaf, Because that startup of mine, eTax, actually was a customer of Rapleaf. Your data powered our product. So full circle.

Yeah, exactly. I wish I was smart enough to invest in Airtable when it started back then in 2012. As you know, I'm a huge fan of Airtable. I use it every single day. Tons of our companies use it. So I'm in it all the time. The no-code, low-code movement. Promise to democratize software creation. Where are we at on that vision?

So I think it's really interesting because the first wave of no-code, which I like to think that we helped to pioneer, was really about taking all of these complicated but powerful app concepts. So data model, logic layer, automations, and then interface. And if you're a programmer, you know that those 3 layers are effectively the building blocks of any application, like model view controller. And distilling them into a form that was really intuitive for any non-technical person to interact with.

By non-technical, you really mean a non-software developer. They still could be a relatively technical person or something.

I think that it really comes down to like how well designed the product is, because in our case, we wanted to make it for everyone, literally everyone. Early on, we had such a broad range of users, cattle farmers building cattle tracking apps, replacing— apparently there's a whole vertical solution category for this. CattleMax and all these other kind of brittle tools. We had small legal firms, lawyers themselves going in and creating their own rolled version of CaseMap, which is a LexisNexis product. So they built their own case management software. The coolest random use cases too would emerge. I remember being on a screen share with a woman who I think was an octogenarian cataloging family memoirs, literally figuring out how to drag and drop into Airtable, like these scanned attachments. So I think it really comes down to a test of how truly easy is the product to use, because there's been low-code platforms for much longer than we've existed. Before Airtable, there were all these other products like Quickbase. Even Salesforce itself is technically a low-code app platform. You can configure it to build any app you want. And even before then, back in the day, FileMaker, Microsoft Access, Lotus Notes, Ashontate dBase. And so I think what we really wanted to do was truly lower the floor to an unprecedented level. It really didn't require a technical admin IT type person, let alone a programmer. Truly anyone could go and build their own useful apps.

And then where are we at? How much of that vision is realized? How much of it is still to come?

So I think it's a perpetual journey, both in terms of Lowering the floor, and I like this Bill Gates quote. He said, "There's only been two truly astounding technologies that I've seen in my life, truly jaw-dropping moments." And one of them was the GUI. When he probably saw Xerox PARC and the original GUI, it obviously sparked the inspiration for both Microsoft and Apple to go do the same. And now ChatGPT. We've bottled up intelligence and broad reasoning capability into an instantly available. Fairly cheap API that can be scaled up on demand. And what that means for no-code is literally if you interpret the vision of no-code, not as like, okay, we have to just go and create GUI metaphors for app development, but instead now you can do it without code, but through agentic conversational app building. And you see some of this with the vibe coding apps as well. Like you can go to Cursor as a non-developer and type in the sidebar chat, just build me an app and it will try to do a good job. The range of expression that you can do is just broadening and broadening and the barrier to entry. Is dropping. The vibe coding products, they do have some limitations. If you're using agents to generate actual code, it's somewhat unreliable. There could be data issues, there could be security issues. If you're not a developer, you can't actually inspect the thing that it's generating. You can't inspect the actual code itself. You can only look at the rendered output and hope that underneath the hood, it's all working as it should be. And I think that's where we have a unique take, which is blending this concept of no-code GUI components Where you can see and directly interact or edit with the data layer, the automations layer, the interface layer, literally drag and drop things if you don't like how it was generated. But also on top of that, this agentic conversational layer where you just talk to our new agent Omni to do anything, literally anything that required point and click in Airtable. Now you can just conversationally talk to this agent to do.

Where do you think AI is going to change the most and where is it going to lag behind the least, let's say over the next 3 to 5 years?

I think that every software category is basically up for grabs. And if you think about the disruption of SaaS versus on-prem software, I actually think SaaS was really important, but not actually nearly as big of a deal as AI and agentic product experience, meaning we massively reduced the overhead and cost of maintaining software, of maintaining this deployment. It was much easier to adopt and distribute, but the software itself, the functionality was the same.

Or it was easier to collaborate because there were a lot of things where like download to my computer and pay $100 or something before, but I couldn't collaborate with other people. So it's more of like a single player mode type of thing.

Absolutely. That's very fair. And in the hacks they had to like do multiplayer mode, think about Microsoft Access. You literally have to save a file to a shared LAN. disk drive and then like a network disk drive and then collaborate by multiple clients.

Before Git, I would use SVN, super hack it just to share files with people and stuff.

So not for actual code commits, you would use it for your own Dropbox.

Yeah, exactly. That was my Dropbox that we use internally in our company. It was terrible, it'd break all the time. It was super hacky.

I think you've officially earned your uber nerd badge from that. When you use SVN for basic file sharing and collaboration, you definitely earn uber nerd status. The distribution, like the sharing collaboration layer was probably the single biggest disruptive force of SaaS. Our contemporaries, like think about Figma versus Sketch. It was the same functionality, but one, Figma was much easier to sign up for, adopt, and then share. Anyone could open a collaborative file and same with Airtable. To me though, that's still very incremental. Relative to the massive disruptive change of how product experiences, how software experiences can behave when it's fully AI native to the core. And so when I think about a lot of the current software landscape, much of it is just like we built a vertical solution for one category, be it plumbers, be it dog walkers, whatever, be it like small businesses, yoga studios, like mind-body type stuff. And we just built all of the data model and the business logic rules for this particular use case into our software. And now we have a sales and distribution channel to sell it, but fundamentally it's just hard-coded software. Now, this massive disruption with agentic app building, first of all, the barrier to building your own truly bespoke software that's quite advanced, not just simple, has radically gone down. So that obviously tilts the spectrum in favor of custom software. But then on top of that, the software itself should feel and have a different form factor, it should be agentically infused. So if you build a CRM today, it shouldn't just be your age-old CRM. It should actually have agents automatically running within it that go out and do web research about your customers or go and automatically listen to call transcripts or use MCP to grab data from other sources. The radical improvement in the product value that you can create with a truly AI-native baked experience Is so big of a difference that it makes all of the current incumbents, especially the legacy enterprise incumbents, up for grabs.

You mentioned MCP. How bullish are you on being able to bring in data and collaborate as opposed to needing access to the browser to go do it for the user or something?

So I think both will be done agentically. We've seen what browser use looks like in the context of agents, not just literally scraping the HTML like what Deep Research does. products like Operator from ChatGPT or Manus, et cetera, that now have literally the ability to scroll through a website, hand it the user back control when it needs the human to go and do a CAPTCHA, uh, type in your credentials, et cetera. So I think, first of all, browser use doesn't mean you can't do it fully agentically. I think it's a huge deal that agents are now able to effectively go and automatically do increasingly more and more, if not all of what a human would have had to do with a desktop and not just like on the web browser through, uh, MCP, but anything on your desktop. And I think it's very clear that we're quickly approaching that point where at least anything that's fairly deterministic or rote, an AI agent can definitely do. I mean, we're seeing this, um, big discussion in the e-commerce space of what happens if people start not going out and doing comparison shopping manually as they do today. They search for some item of clothing. Google pops up its sponsored results. You look through those, you click on them, you examine them. What if you just tell an agent everything about yourself and what you like, your style, or it analyzes you and then it just goes out and looks for the perfect products for you and then can even transact?

I, by default, buy probably most, almost everything on Amazon and probably 70% of the time or 60% of the time it is the cheapest price, but there's a lot of probably ways I could probably get a cheaper one if I had an agent doing it for me and it just told me, hey, by the way, you were going to buy on Amazon. You just saved $50. I would be happy to share a percentage of my savings with the agent.

I think the common misconception I see is AI and agents can only be used for very logical tasks. Literally find me the lowest price. It's very utilitarian, but I think agents actually can have a certain sense of taste. You see this with the creative applications of LLMs where ask it to compose a poem about your company in this style with this flair, and it does a very good job. So I think it actually can learn.

I use it to pick restaurants, pick hotels, things that are very, very personal to me. And I try to keep it, okay, here's what I'm doing. Here's who I am. Maybe it's not perfect, but it does a pretty good job. It does as good of a job as if I had a personal assistant do or something.

Exactly. It's not just a comparison shopper. It can actually be a personal shopper where it's finding products that it knows you might like because it understands your tastes and preferences and style in subtle ways. that are not just a keyword match. I think it gets really, really interesting when you start to look at the disruption with agents as not just being strictly rote utilitarian tasks, what we could do with like RPA, for instance, where it was very, very scripted, and instead much more subjective and soft-feel tasks.

Now, back in the day, if you think of Airtable 1.0, it was Microsoft Access. Microsoft Access was an amazing product. It was so good. It was so well done. And then at some point, I don't know when, 20, 25 years ago, Microsoft just killed it for some reason. Do you have any idea why? It could have been amazing. Why did they do that?

Funny enough, when we were founding Airtable, instead of just looking to the future of where we thought things were going, we also looked to the past to like try to understand the existing, the prior art, the Microsoft Accesses of the world. And by the way, like Lotus Notes, also conceptually similar and a very, very successful product for its time, groundbreaking. In terms of the paradigm they defined there, the first collaboration meets structured data type product. But long story short, I ended up talking to a number of people who were literally involved in Microsoft Access, Excel, et cetera, like Steven Sinofsky, Adam Bosworth, who actually created the very first GUI database as his own company and then led, led a different, one of the other leading database companies early on, and then later joined Microsoft to run Access. Steven was a leader at Microsoft that I think oversaw the Office suite at one point, you know, and beyond. But the story that I ended up piecing together was that actually at one point, Access and Excel were effectively 2 peer products that were really, really important. One's a database, one's spreadsheets. Spreadsheets more for number crunching, ad hoc analysis. Databases for structured list data, CRM-y, ERP-type use cases that you build yourself. But if you look at the products superficially, they kind of look the same. They both have this like tabular view, this grid view, and Excel actually had started building more table-like functionality into Excel. So it wasn't just number crunching. You could kind of use it to make a list. And what I heard was like, there was a literal moment where Bill Gates called a meeting because he was hearing all this consumer confusion. And the, the head of Access and the head of Excel were in a room together with Gates. This is way too confusing. We gotta decide on one product to be the default product. That subsumes both of these use cases. And then the other one will kind of relegate into edge case or power user product.

Okay. So I always thought the alternative history is Access would be amazing, but potentially the alternative history is that Excel wouldn't be as good as it is today because they wouldn't have been investing in it as much, or they spread their bets. It wouldn't have been as good.

Yeah, maybe. Microsoft also had enormous resources at the time. I think they could have made both products really great. And I think it was more of like a consumer confusion issue for them. But I think there could have been an alternate world where Excel actually got relegated into a power user tool for finance geeks. It stayed mostly a number crunching tool rather than becoming the dominant default. In fact, I remember as a kid when I installed Microsoft Office 95, the one that came with Windows 95, I don't think we got Access by default with the bundle. I think it was like Excel, Word.

It was like a professional to get Access. Yeah.

What if Access had been the default bundled product? More people need effectively a database than an advanced number crunching spreadsheet that can do IRR, but could have been very vice versa.

I mean, the first web application I ever built was built on Microsoft Access in the cloud. And then I built a whole web application on top of it. And it was the database of not a very fast database. So obviously it didn't scale that well, but the database running underneath it. Then you upgrade, they forced you eventually upgrade like SQL Server back then. So you had these other servers, which is like a terrible product and really hard to use. And now another question I have for you. So 4 years ago, when I was considering hiring a marketer, my basic question to them was, how good are you at Zapier? And if you were not good at Zapier, first of all, you'd never even heard of it. The interview would end. I would just like, show me your Zaps. If you had great Zaps, okay, you're hired. And if you don't, you're not. Now today, outside of Airtable, what's the stack that all good employees should be using?

First of all, I think that's really interesting because to me it relates to this founder mode concept, which I think is a misnomer. I don't think founder mode is just about like founders running a company the way they want. I think it's actually about saying executives in general should be very hands-on and in the details and be experts at the actual craft rather than just delegators and managers. That idea of ask a marketer, especially if they're, even if they're a marketing leader, I assume you would have asked a CMO. If you can't get hands-on with the modern stack of tools that make you 10x more effective, then you're clearly not up to speed. You're not in it enough. No time has that been even more salient than now with all of the AI tools that you can use. So, you know, one, we use Airtable a lot for everything. One of the top use cases in general for Airtable out there in the wild is marketing operations. So everything from managing a content pipeline for this podcast, you could say, here's all the upcoming speakers. Now I can have AI agents in Airtable do research on each person and find recent news. But you can also have it— the cool thing about what you can do with Airtable is that think of like any one-off prompt that you can go and create in ChatGPT, including prompts that leverage web search. So let's say you're researching one speaker that's upcoming on your podcast. Here's a specific prompt that in a very, very nuanced way finds relevant details about them and their history, et cetera, using basically AI web search. And once you have that prompt kind of optimized, the current status quo with chat products is you'd have to go and remember to copy paste that exact prompt and give it the right input context. There's more data that you have about your own notes. From your own notes about a speaker, it's quite laborious to go and replicate that same AI use case again and again. Effectively, what you can do with Airtable now is bake that prompt into a workflow, define a field that performs this agentic action, in this case, web search, and do it every single time you add a new podcast guest or upcoming episode. And so I think that's the really powerful idea that we have with Airtable is take any AI use case and we're all standing on the shoulders of the giants that are The awesome foundation models that we all now have plenty and cheap and scalable access to, and they just keep getting better, cheaper, faster. And on top of that though, how do you build a true application, which is effectively just a container holding the right data model, the right workflow, and the right human interface to be in the loop with the agent and AI kind of applications. So we do a lot with Airtable, and I would expect any marketer or frankly any role a person that is coming into Airtable today, if not to be an Airtable expert, of course we can't only hire people who already know Airtable that well, but I would expect them to be able to, including in the context of an interview process, to like whip up a pretty good Airtable implementation for a relevant use case. So you could be a lawyer or somebody in our legal team. Can you go and build a contract review workflow in Airtable really quickly? That's the kind of systems thinking required. I think to really leverage AI in the modern era.

If you're using like a Zapier or maybe a newer version of a Zapier, like a Pipedream or an N2N or something like that, they charge on a per-action basis. You get X number of actions and you pay. And then if you go above that and you can get your entire team on it. So we have 100% of our team using it and they're all using it and they're all benefiting from it. Airtable is still old school. It's per user. It's very limited. Think of like Airtable versus Google Sheets. I love Airtable. I way prefer Airtable over Google Sheets, but Google Sheets probably has over 100 times the use of our company because Google Sheets, there's no marginal cost. I don't have to pay an extra $1,000 per user to go use it. How do you think about pricing? Because I think you would explode in usage. Your revenue would go up. It might not, it wouldn't maybe go up relative to the explosion, but it'd go up. How do you think about pricing? And I'm sure you've thought about this a lot.

I think pricing is misleadingly complicated or deceptively complicated, meaning you want to have a pricing model that is ultimately simple and grokable by the customer, but there's a lot of, it's multidimensional optimization underneath the hood. Now, especially if you want to do consumption-based pricing, Okay, but then what's your actual inference cost, especially since usage now sometimes correlates with actual hard costs, and it doesn't always correlate one-to-one. Let's say like you price for every zap in the Zapier model, but some zaps take 10 times more cost, including inference costs in between, because they have a reasoning step with an AI agent. And so are you absorbing that kind of tolerance for a variable margin on each of these costs?

But that's the same on a per-seat basis too, though, right? Seats are just going to cost you way more than others. Some people are logging in all the time.

For sure, exactly. And maybe even arguably more so with the seat-based model, like then the seat-based model is effectively on the one extreme of saying, look, we're going to absorb the variability of cost structure and abstract it away from the user or the customer so that they just see a very simple— they get a very simple mental model. Okay, I have people and anyone who wants access to this product, we're going to pay this amount. And of course there's some sliding scales of which plan do you want to be on? And for the most part, it's like, How many people do I have that need to use this product? And I want to pay for each one. And that makes a lot of sense, I think, to a lot of customers because they still think about the value of software as really an enhancement to human productivity. There's a whole argument around how quickly are we going to see agents actually take over jobs and outright replace humans, in which case the value of an agentic solution is not correlated to the number of people using it. In fact, it might be inversely correlated. If you can displace 100 roles with one agentic solution, that's actually worth more. You need to do outcomes-based pricing.

I'll give you an example of one of the companies that we have. So they have about 60 employees and every month they have between 3 to 6 Airtable seats. So they scale up and then they immediately scale down because it's really expensive to have these. They should have 60. So they should have 60 people using it. It is an extremely powerful tool. They have 60 people on Zapier. Or whatever they're using, Pipedream. They have 60 people on it and they're constantly using it all the time. But maybe you have 5 expensive seats and then you have the super cheap seats for people who just need to do a little bit here and there or something like, I don't know the way to do it, but most of the company is not using your product, which is such a great product because of that price barrier.

That's very fair. And we do have a free tier and you could add different people to a free workspace. In the company, 50 of the people could always be on a free workspace and they can use the Airtable free plan.

If you have a paid workspace with 3 users and you want someone else to be interacting with it, then you have to pay for that.

That's right. We do have free read-only users, but if you want them to be a full editor, they do have to pay at the same plan level.

Yeah. Even at one tiny thing somewhere, then they have to pay, right?

Yeah. So that's fair. And it is a question. That predates GenAI and agentic or consumption-based pricing around AI. I recall a really interesting conversation I had with Zuck where, you know, he was giving me some advice about thinking about the business and radical proposal for me at the time. This was probably at least 5 years ago. And it was basically like, what if you just made Airtable entirely free and just go for mass adoption? And instead of doing any kind of seat-based SaaS revenue, You have this marketplace thing and we just launched this app platform where third-party developers, as well as us, could build apps that sit on top of Airtable. And he was kind of like, what if you just went all in on getting scale and your eventual revenue model could just be like taking a skim on the, uh, the apps built on top of Airtable? Zuck is a really, really smart guy. So I deeply thought about it. And I mean, we actually didn't end up doing that because, A, I felt we could get to some hybrid alternative with a combination of freemium and other pricing factors and Are we like more of a SaaS business or are we more of like a consumer product? It's a valid question. We could have gone either route and aim more for a consumer-esque tool. And we obviously picked the road of more SaaSy. And we were also moving upmarket into enterprise and able to command really large ASPs in enterprise where we have multimillion-dollar paying customers who are building mission-critical operations on Airtable, where the value to them is higher than maybe they even get from a Palantir solution in some cases, or A solution built on a ServiceNow or Salesforce. So there is some strategic direction and vision that is embedded into the pricing model as well. Are we aiming more for the mass scale, mass everybody using it, or are we aiming to go and serve the highest end use cases? Can we do both with some creative combination of pricing? I mean, to your point, maybe there is an option where we do have a more restricted tier of seats so that you can just add everybody in the company to that workspace. They still can collaborate. But if they're a very limited or infrequent collaborator, you charge a lot less. So I think those are all valid questions.

I want you to help me get off Google Sheets, even though I love Airtable more than Google Sheets. I'm in Google Sheets 20 times more time. And every time I'm there, I just curse you. I curse you, Howie. I'm like, Howie, just let me get off Google Sheets.

All right. I'm going to record a Loom of you cursing me and just autoplay it until I finally figure out the fix to this. But I think it's a valid push. And now, especially in the agentic era, we have even more reason to rethink pricing. We just did this big, like, AI-native relaunch, and one of the things we did was we actually bundled— before, we had AI as an add-on to Airtable. So you have the seat price of Airtable proper, and then you basically had to be on a paid plan of Airtable to then also pay for an add-on of AI, which also then came with some basically inference credits. To run AI use cases, and we basically now said no. Everything needs to be agentic by default. Even the free users should come in and get a vibe coding like app building experience that we're going to eat the cost on. Inferences cost money, and we're just going to eat that cost even on our free plan. And so we've already made some changes to our pricing thinking and gross margin thinking to be able to support the type of new product paradigm. It wouldn't be that radical, or it would be very natural. for us to then continue to think about, okay, what are new disruptive ways that we can just get more people to get value out of Airtable?

Roll of DaaS isn't just a podcast. Roll of DaaS membership is a private invite-only community for founders and executives of important data and AI businesses. It's a place for off-the-record, high-trust conversations focused on the biggest questions data leaders are facing. The community is about peer learning, curated intros, and access to private events. Everyone in the group is building or leading something meaningful in data. If that sounds like you, apply to join at worldofdaas.com. That's worldofdaas.com. How do you route the AI requests to the best model, and it's both the model that both perform the best on that specific topic, but also there's a lot of models which are super low cost. You can even run them locally. So is there some sort of layer that sits on top of it and is a routing system?

For sure. So we have basically 2 very different complementary types of AI in Airtable. One I call AI building and AI at runtime. AI building is basically where you come in and this is the vibe coding experience. We have Omni, this agent that you talk to and can build or analyze any app in Airtable. You can tell it, build me this dashboard in my CRM, go create me this new app that is a podcast tracker, go add this new object or field to my existing CRM. So it's the app building agent and that one We basically do have an underlying system where we can choose different models for different types of requests. And some we optimize for latency so we can very, very quickly do a simple request. Some we optimize for quality where actually it turns out the reasoning models don't always do the best at text processing type tasks. So if you give me a bunch of notes and you say, I just want you to extract out the key people referenced in these freeform notes and then go and update those notes into each of those people records in my CRM. Actually, we found a non-reasoning model like a GPT-4.1 actually can outperform the reasoning models in a lot of these use cases. So we're doing a lot of our own selection of what is the best model to use for each of these different types of requests or layers of requests.

But how does it, you just know based on what the request, how do you even know what the request is and how to route it?

That's an LLM step itself. Eval are super important. I think this is something that is probably underappreciated, which is Building an agentic experience is so unlike traditional deterministic coding. When you think about your algorithms class from CS, it was very mathematical. There's a closed-form answer to like the optimal way to do a sort of a list. Whereas with these agents, you do have to like do some trial and error. And by definition, having a really good eval setup is how you can very efficiently navigate the somewhat stochastic nature of, of LLMs. You don't know predictably how well they're going to do for any given prompt until you try it. And so you just have to try it. And evals are the way to try them in a consistent, quantitatively measured way. So we have a really good eval stack where we effectively can go and just try lots of these different combinations. We can A/B test. Okay, what if we use this more efficient, smaller reasoning model to decide whether to call on this bigger, slower reasoning model For this type of request, I have an internal power user config mode where I can actually go in and hand override the different models we use at different layers or for different types of requests.

All of a sudden you can be like, wow, we could save $100 grand a day if we just do this thing. And it's as good. Sometimes it may even be better.

I mean, I think we're still at the point where we're not really cost optimizing. I would rather eat as much cost as we can in this moment. To deliver the best quality experience. I have high confidence that like cost will come down naturally for any given quality level. Sam Altman literally tweeted a couple months ago, O3, we just decreased the price right now by 5x. That's insane. Literally as of this tweet, we dropped the pricing by 5x. And so there's massive race to the bottom, software optimizations to inference. There's obviously the hardware costs going down for the same amount of effectively flops of compute.

He's taking the Zuck pricing strategy. He's, he's trying to get mass adoption, not trying to optimize.

Totally. And I think that's probably the right thing to do in their race, which is, I think there is a huge advantage to the provider that has the cognitive referent or default set of models that not only people use through the consumer product ChatGPT, but then even programmatically through API. If everybody just learns the ins and outs and the nuances of OpenAI models and the texture of O3 over, let's say, 4.0 Sonnet from Anthropic, then there's huge long-term strategic advantages to that.

Especially if there's memory.

Exactly. The memory part for sure, on like a per-customer level, makes it much more sticky. But I also just think it becomes a standard. It's Excel, going back to our earlier discussion, like Excel becoming a standard. I remember in the Windows 95 era, like, you know, my family was poor and we actually didn't want to spend the $300 on Microsoft Office at the time. And so we bought some like knockoff version of it. I forget what it was called. It was WordPerfect actually, I think, and the correlated and the associated products. And it was just not as good. And I remember like going into school in our school, we have the computer lab and we learned Excel and Word and all the Microsoft products there. And I learned all the shortcuts and the advanced features. And I went, oh, the thing that we have doesn't even have all that. So having the standard where like people just learn, developers and consumers alike learn.

This is why I want everyone using Airtable as well.

Okay. All right. You convinced me.

Okay.

Mark my words. We will do some radical improvements to our pricing model to expand the funnel by far of people using the product. Um, I'm very aligned on that. On this note of drawing the parallel back to ChatGPT, I do think in the AI era, There's clearly some products that have been successful doing an enterprise sold motion. Some that come to mind are like the Harveys of the world, very much enterprise sold. Palantir's like custom AI deployments model, right? AIP. But I think by and large, the vast majority of real AI value creation and adoption and even revenue has come through consumer-led, end-user-led adoption. I mean, ChatGPT alone, I think OpenAI is now like a $10 billion revenue run rate business. And I assume the vast majority of that is ChatGPT. I mean, ChatGPT alone probably is the lion's share of total AI revenue out there today, true gen AI revenue, at least. Not to mention you have the cursors of the world, you have the Manises, et cetera, that are all PLG led. And so I think there is a strong favoring of, for AI experiences, it's better shown, just self-served rather than sold or told.

Yeah. And no long-term contract, easy to just try it out. So some of these do have much higher churn than others, but it's good.

I'll take a 10x bigger funnel with 2x higher drop-off, right? You know, like that's still, you still net out way better.

Yeah. Yeah. I mean, no salespeople to pay and all this other stuff. So your CACs are much lower. Like if your CACs go down by 20x, who cares if your churn goes up by 3x?

Yeah. I mean, your CAC could be zero if you're ChatGPT. It's just massively viral. People just talk about it.

Well, actually, I don't know. I bet you their CACs are insanely high because most of their CACs are like Sam tweeting and Sam's time is probably worth, I don't know, $1 million an hour.

That's fair. Okay.

Well, so it's probably like their CACs are probably like insanely high if you put his hourly rate on it.

Okay. Yeah. Although you could say he probably derives some amount of personal entertainment value from the tweeting. So is that actually just like a moonlight job or his actual like paid job? And at one point, at least, he said he had zero equity in OpenAI. So actually, maybe his time is free at that time.

From OpenAI's perspective. Yeah, exactly.

I do think within the era of AI, the importance of getting just more people into the top of funnel to just actually use and try your product is more critical than ever. So to your credit on the pricing model stuff, it's like, I think Palantir, short-term bullish on, long-term uncertain on. I would never bet against them, but short term, they've clearly found this repeatable motion of selling a turnkey way to just deploy AI into the enterprise. Longer term, does all of the AI deployment just become self-serve and so easy to spin up that companies don't need like the current Palantir very heavyweight sold model? They send an FDE, you're paying like millions of dollars for a custom build.

I do think when you're dealing with security type of information, information that is national security related or something like that, you're probably still going to need to be sold to for quite a while.

Oh, for sure. Definitely for that. But Palantir today is actually now already trying to sell downmarket as well. Even like smaller companies that are not dealing with like high security issues, doing like much more banal deployments. Here's how we're deriving insights from customer feedback.

If I ever saw a random company doing that, I was like, okay, that's not a good deal. Like Palantir is— you're paying Palantir at least $5 million for your implementation. If they're going to sell like a $100 grand thing, that's probably not the right thing for them.

I think there's this near-term windfall that Palantir is getting around like CEO-led purchases of Palantir as an easy button to show that they have some AI implementations. It's like the CYA, cover your butt to the board and to the public market, maybe. Even if you're a smaller company and not in security, not in high data sensitivity, to just be able to say like, I have an AI deployment. And I think as the self-serve ways to do that and do that in a meaningful way, not just gimmicky or toy way, like early on, people were all using ChatGPT for superfluous, fun, gimmicky things. Whereas now I think they're actually figuring out how to use it for more meaningful use cases and better yet, using the underlying models in a tool like an Airtable to then power a more recurring workflow that is actually high value. You shouldn't be paying hundreds of thousands or millions of dollars to Palantir to do a bespoke build for.

A ton of ad hoc hires that are going on in our industry. Back in the day, an acquihire was a pejorative word. And today acquihire could mean you're getting a billion dollars or more than a billion dollars. What's your thought about where we are in this weird moment and how does it continue? Does it stop? Is it just like this odd moment in history?

I mean, I was acquihired in my first company by Salesforce. And at the time that meant maybe a low 7-figure payout and some retention plan or whatever. When you look at the recent acquihires of, I mean, very, very recently, Windsurf or Scale, et cetera, I mean, these are monetary sums that look like full business buyouts. The Scale $14 billion price tag, I mean, that's like bigger than a lot of public companies. And the crazy thing is, I mean, it is truly an acquihire because partly maybe to avoid antitrust scrutiny, they're not actually buying the business. They're not buying the company, the entity, the business, the IP. They're intentionally leaving that behind. And in fact, Scale, the business, is still an independently operating company. To me, it's a little bit of like the key person theory of human civilization. Like, how much of technological progress is driven by a couple very, very important key actors who have the unique insight or wherewithal to just really move things forward? And it seems like, and I would generally agree with this, that right now in this moment, given the ingredients, Of all of the potential things you can do with AI, and especially if you have massive resources like a Meta that you can gain leverage with. If you win at AI, that is worth trillions of dollars to you. But if you lose at it, you've basically become irrelevant. We are at this moment where it's not just about hiring up an army of 100,000 people to work on a problem. It's about finding the critical people who really have the combination of insight, ingenuity, But also just extreme motivation and intensity to make big, big things happen. So I think it's actually somewhat rational for these acquihires to happen, not just because they have to be done this way to avoid the antitrust issues, but because Meta is basically saying, look, I value Alex and the team that's coming along so much to really accelerate my efforts on AI that it's literally worth, I mean, $14 billion. is not even that big of a percent of their total market cap.

But Meta values Alex, who is going to be an executive at Meta, an important person, but not the CEO, not anywhere close. They value him way more than Apple values Tim Cook in terms of compensation. So these CEOs of public companies are less valuable than often 100th AI engineer at Meta.

It's a very valid point. And I actually think it's a good system if you can value talent rationally based on how much leverage they provide the business rather than having some artificial internal reasons or hierarchical reasons for why you can or cannot pay talent a certain amount. This becomes like a whole soliloquy about capitalism, but at least in this specific regard, Ben and Jerry's tried the experiment at one point of like the best paid employee, which happened to be the CEO for them, could not be paid more than 10x more than the lowest paid employee who might be someone literally manufacturing the ice cream in the factory. And of course I believe in fair wage, like, you know, people should be paid well and so on. But you also need to just hire the best person that creates the most leverage for the most important jobs or the roles. And so maybe an interesting comparable here would be like in professional sports, You have some players that probably make far more than the manager of the teams. And that's rational because if the player is that good, the team actually wins and then the whole franchise becomes more valuable.

Yep. The analogy may hold all the way through because it's very possible that this employee is only important in a very brief moment of time. For 3 years, they're going to add value and then they're not going to add as much value. Just like a great professional athlete may only be good in a very short moment of time. 20 years from now, they're almost certainly not going to be playing on the team anymore.

Totally. This is one of those things that it's like a known best practice or piece of, uh, managerial theory, which is like, it's a team, not a family. You gotta think of each team member as being a part of the team in this moment, and you can appreciate them for that and always think highly of them in retrospect. But not every team member needs to be part of the team forever. And in fact, like that includes the CEO and sometimes the founder as well. But it is always about Solving for what we need at this moment in time. Because if we don't put one foot in front of the next and really execute well in this moment, we're not going to have the chance to play a good hand in the next moment.

Yeah, that makes a lot of sense. Now you've been doing Airtable for 12, 13 years now. How do you think about the next 12 to 13 years? How do you think about your future and what you're doing and obviously built a great company, but how do you think about this like going forward?

I remember reading this HBS article a while back called Your Strategy Needs a Strategy. And it effectively, it said, look, the right way to think about strategy is actually market dynamic dependent. Meaning if you're in a market or a moment with a very low rate of change, it's a stable environment. And you can maybe call a certain era of mature SaaS as being an example of this. It's kind of deterministic. You can execute in a pretty known way. And it's really about predictability and how do you build a really durable business. And a lot of the PE buyouts are examples of this where, okay, we know what we're getting. We're not going to expect some crazy rabbit out of the hat innovation from this business that we bought, but we're just going to chug in day in, day out. We can probably, you know, continue to scale this. Whereas if you're in an environment where there's an extremely high rate of change or like a lot of uncertainty, which I would say we're definitely in across all of software now because of AI, I think you have to just be completely adaptable. Agility is much more important than having a clear known long-term pathing of we're going to do this and this and this and this, and this is exactly what's going to happen. And so to me, it's about playing in the right space. First of all, picking the right vision and the right category to disrupt. And maybe the category doesn't fully look baked yet, but no-code to me is absolutely like a category that holds even more relevance and is even more exciting in the agentic era. By def— like, literally, if you interpret the words no-code, what can people who don't know how to code do with software or app building? That only gets more magical, and the market only radically expands in the AI era. So are we picking the right space? Are we playing the right space? I feel very confident the answer is yes. And then it's less about what's our predict 10 chess moves out what we're going to play, and instead say, like, do we have a good hand right now? Do we have the right pieces?

Okay. But like, just asking like more of a personal question. At some point, Zuck's still there, but he hasn't been there as long as Jensen Huang, who's been there forever. He's been doing this crazy, but then Larry Page is no longer CEO. Jeff Bezos is no longer CEO. Bill Gates left a long, long, long time ago. And at some point, people want to go do other things. How do you think about it personally? Like they, I'm sure they still have those challenges and those other types of things. Gates left right when the internet was happening. It could have been. Maybe he should have stayed, but he decided he wanted to do other things.

So there's 2 questions. One is, do I have the energy and the motivation to keep doing this or doing anything versus some people go into semi-retirement, maybe they do an easier job or whatever. There's nothing wrong with that. It's like the hobby projects or, and no offense to anyone who goes in permanent board role.

They become a venture capitalist. It's the typical retirement, right?

You know, VC can be hard as well in very different ways. Do you just not have the energy? And it's fine to admit it, to not be an operator anymore, full-time operator in intense mode. And then even if you are motivated to be an operator, is this no longer a space that interests you? On the number one front, I am more fired up than ever before. Look, maybe partly it's a function of age. I'm thankful that like I got my start very early. You cited the math early, but I'm only 36 right now. And so I'm just beginning and I've just learned the basics of how to be A somewhat effective operator and founder CEO and whatever. So I'm nowhere near ready to retire. And then on the 2 front, is this the right space? If we were reaching this kind of mature, that starts to look more like a PE type of business, you just do the same thing year in, year out, and the margin of difference between me and some other, I would no longer be that fired up about it, nor would I probably be like Six Sigma better than the median person at it. And so there's probably somebody else who could do that equally well or better. Whereas now this is really a refounding moment. I get to do something that I'm really excited about, which is start with technology-first innovation. In the mature market environment, it's no longer about finding disruptive tech innovation. It's more about saying like, SaaS is kind of mature. Anybody can build a database or a hard-coded database with application logic on top and create a CRM for X. And then there were so many different permutations of X. Here's the CRM for this kind of doctor or this kind of dentist or this kind of plumber, et cetera. You can build a good business that way. But what fires me up is like, I think I'm a technologist at heart. I've always found that really understanding the cutting edge of what's possible technologically, and maybe in the beginning of Airtable, that was building our own robust single-page app framework before there was good prior art. React didn't exist. Backbone even didn't exist. Like we had to roll our own rich frontend experience. We also had to build our own backend and create a real-time database engine and a real-time collaboration layer, et cetera. All these things were actually hard technical problems that we also combined with a creative sense of taste of how do we not just build tech for tech's sake, but make this an awesome and delightful product experience. And we're going to learn a lot while doing it. The new Airtable has every bit of that and more that we had in the beginning. And so it's like, I get to go and play with all the cutting edge tech. What, how do you build agents? How do you build new things with models? How do you go and use existing products off the shelf? And then how do we like creatively think about reproductizing that into a new form?

Now, a couple of personal questions. What is a conspiracy theory that you believe?

I think I hold the world as a, I hold a very stochastic view of the world, which is I don't think I know how everything works perfectly. I don't think anyone knows how everything works perfectly. And I think there's almost this certain level of randomness, just known unknowns. So therefore, it's never about, was the lunar landing definitely real or was it fake? Probably given my priors, like P99.9% likely real, but 0.1% maybe fake. So I just watched that comedy movie, Fly Me to the Moon with Channing Tatum and ScarJo. It's literally about this. And obviously the whole thing was meant to be comedy, but I just love to think about the what-ifs or the possibilities. Not because I then come away like thinking, oh, it's 100% faked, but it tickles my fancy to imagine alternate realities and then hold both those things. But I don't know if there's a big conspiracy theory I hold.

Last question we ask all of our guests, what conventional wisdom or advice do you think is generally bad advice?

I definitely think the stuff that Brian Chesky talked about in that now famous founder mode talk resonated very deeply with me, which is there was an era of Silicon Valley, and this was true for Airtable, the conventional wisdom told to us was, hey, you got to go and apply all these traditional ways of running this company, hire traditional leaders who have a traditional skillset and so on. And there's probably some truth to that. Like, you can't run fast and loose a 2-person company, even as you reach maturity and need to be public ready and so on. But I also think if you lean too much in that direction, which a lot of companies including us did, then you lose the ability to have that agility, which is especially important now. And also you lose the unique alpha that I think these highly innovative companies that by definition, if you innovated well enough to get amazing product market fit to begin with, that's your alpha. That was your alpha at that time. And if you can keep doing that instead of just maturing and milking the original innovation, a really long time, you can go create new innovation, reinvent yourself. That's both profoundly valuable, but also requires a radically different approach to running your business, running your company. And Brian would call it founder mode. I call it just high agency, being in the details, being radical, being age— high agility, and disrupting yourself every at least couple years.

What's the biggest conventional one that you took out or you stopped doing, and it just had the most impact in the company?

Was it like, okay, we don't need all these weird HR rules or something, or Some HR roles you still definitely need, but I think it's like one, how I hire. So we touched on this a little bit earlier, but when I think about hiring, execs are not just about managing people and process. They really need to be experts of their own craft. So if you're a marketer, if you're a product person, you actually have to be good at product. You have to be good at marketing. And especially now, you have to be highly fluent in getting hands-on with real AI use cases and building things. And first of all, it takes fluency with tech and products to be able to talk about those products and to be able to make decisions about them. And now more so than ever, it's all so nuanced and so fast moving, you have to be in it. So detail-oriented and in-it experts of their craft as executives, I think is one big shift for me. 2 is like, I just accepted as common Silicon Valley wisdom that you have to have all of this process and structure in general in how you do things. We used to use a lot of the RACI model, which is how you define different multi-party responsibility and accountability. Literally, that's what it is. And anytime you have a really complicated matrix of, well, these 5 people need to be informed about this thing, and this person needs to be accountable to this, but that person's responsible for this, you're probably overcomplicating it. And what you probably should be doing is picking one person to be the DRI with a small lean team to just go spike on this, rapidly move, and do as much in a zero-dependency way as possible. Try stuff, ship it faster.

The HR innovation that Amazon had was the 2-pizza rule. We're going to do pizza, and then somehow you just API it to everybody else, whether you're the HR team or whether you're a product team.

It sounds simple, but the implications are quite profound because it means that sometimes structurally, it actually changes the nature of work that you take on. There are certain types of projects that you can't do this way. And in fact, if you prioritize this style of execution, it actually shifts how you execute and therefore even how your product or business model starts to look over time. You become an accumulation of lots of rapid learnings and quick hits rather than trying to make these huge, slow, single swings that either have to be right or you fail.

Oh, this has been amazing. Thank you, Howie Liu, for joining us at World of DaaS. I follow you at Howie TL. What does the T stand for, by the way?

Just my middle name, Thomas.

Okay, great. I follow you at Howie TL on X. I definitely encourage our listeners to engage you there. This has been really awesome.

Awesome. This was really, really fun. Thanks for having me, Oren. And yeah, happy to chat.

Thanks for listening. If you haven't already done so, please subscribe to World of DaaS wherever you consume your podcasts. YouTube, Apple, Spotify, and more. And please help us get discovered by leaving a review. And check out worldofdas.com. That's worldofdas.com. And of course, connect with me on Twitter @Auren. That's A-U-R-E-N. Would absolutely love to hear from you.