How Airtable is Ushering 500,000 Organizations Into The Era of AI | Howie Liu, CEO of Airtable

Good. How are you, Matt?

Yeah, you too.

Sounds about right. Yeah.

And I think I knew Snowflake at the time, but it was early. I mean, it was early for both companies.

What's changed and what's not changed over almost a decade.

Automations.

I think, yeah, I think that's a very fair depiction.

Yeah. So today we're really positioning as a complete app builder that's the fastest and easiest to use out there in the market. So it's really a rapid application development platform. We are really focused on being scalable enough to serve the largest enterprises, although we also still get used by a lot of either individuals, bottoms-up teams, companies, et cetera. It's very self-serviceable still, but we've really focused on raising the ceiling. And so if you want to build a useful business application—so data, some workflows, and now with AI as a part of it—Airtable is the fastest, the best way to do it.

The stats are 500,000 organizations use Airtable. Fifty percent of the Fortune 500 are paid customers. And really, the largest enterprises have been the biggest focus for us in terms of revenue growth and go-to-market, and even, like, the product roadmap, being able to scale up to support those customers. If we were a public company—so we're not—but we're in the hundreds of millions of revenue, and we would have ended last year as a top-decile grower amongst public SaaS companies.

And we also are forecasting a next 12-month revenue growth rate of also top-decile growth. Yeah.

And as of Q4 of this past year, and now we just completed Q1, we are cash flow positive as a business.

Yeah, I mean, it was a very unfamiliar thing in the realm of VC for a decade, right?

Yeah, I think from a metrics perspective, exactly. I mean, we can certainly talk about the public versus private question later, but it's really important for us to be a great company. And I think great companies tend to be public-capable, whether or not they choose to be public. So we're really focused on being a high-quality business, high-quality revenue as expressed by NDR terms, obviously growth rate, efficiency, Rule of 40-type things. And then also having a very clear positioning around our category.

What do we stand for? What do we do better than anybody else? And a very, very clear, large, and hopefully growing TAM for ourselves.

I'll take my credit for that. Or 20%, like a true VC.

Yeah.

So, if it's okay, I want to almost zoom out for more than a decade into how I see my interest in AI now kind of merging with our ability to play in this space. Back in college, actually, I got really fascinated by neural networks. This was back '05 through '09. Neural nets as a concept have been around for many decades, right? I think it was first articulated in maybe the '60s.

Computing power was very meager at the time. But I found the idea of instead of having to deterministically program every single thing I wanted a computer to do—and I loved programming—but it was like, oh man, it's just so tedious to have to come in and program every single step, right? Like, if this, then that. What if you could just teach a computer through exposure to, here's what I want, and have it learn, right?

And if you learn it advanced enough patterns, maybe you could learn how to behave in more than just rote ways, right? And so the idea, the promise, I think, of ML and AI was always very alluring to me. I experimented a little bit with the Netflix data science dataset, did some stuff around what are all the things you could apply neural nets to? Eventually, though, I reached this kind of point where I made the call that, at least at that time, what you could do with supervised machine learning was powerful but fairly narrow, right?

As in, I actually thought that human-computer interaction, like figuring out how to build products with really great UX and that solved a real business problem in a very intuitive way, was a bigger potential opportunity because the constraint to be able to use software effectively was very UX-oriented, right? And I literally left college with this premise in mind. And it was the kind of founding basis for my first startup, short-lived, was acquired by Salesforce, and then came out of that with the vision for Airtable, which was very much around using UX as well as really clever technical architecture to build a real-time relational database backend.

To ultimately power a product experience that made something very potent—application building, right? Building B2B apps just like you could on Salesforce or ServiceNow or other very powerful but very complicated tools—extremely easy, extremely fast, right? So UX was kind of the real philosophy, the lens. And I think what's really cool is that AI has reached this point where the large language models are now capable of doing things that are very broad but also very deep. Meaning, it's not just very specific use cases like ImageNet and classification of images, right?

It's not even just sentiment analysis or simple text-based analysis. You can literally go to the best models out there. You try ChatGPT with GPT-4, you try Claude 3 Opus, whatever you want, and you can ask it human-like questions, right? And the more strategic or creative the question, the seemingly more surprisingly well it does, right? And so, to me, this is the ultimate moment where you can take these models that have reached incredible potential.

And I actually think now the bottleneck to their application is UX. It's actually giving people the tools to very easily deploy those models into not just a few very central core use cases within a large company, but actually into the fractal of every single function, every group within an enterprise. It's that final mile, and all million miles of it in the enterprise, where even current state-of-the-art models could be creating 1,000 times more, 10,000 times more value than what they're being exploited for today.

So that's what really excites me. And coming full circle to Airtable, we've had a beta program for our AI capabilities for a year. We just launched into full GA about a couple months ago. But in the beta, I think what we learned was exactly this, right? On our roadmap, we had considered things like, do we need to give our customers different specialized fine-tuned models for different purposes? Like, hey, here's a more marketing fine-tuned model, here's one for customer feedback, et cetera.

And what we actually found is that the models are already capable out of the box, the pre-trained models. Take GPT-4, take Claude 3, or even Claude 2 before that, take Llama 2 and now Llama 3. They're already capable of being prompted in either zero-shot, few-shot, chain of thought, but with very simple prompting techniques. And we've tried to develop some UX around this, like prompt templates, like a prompt builder, to make it easier.

But they can already be embedded into the data and workflows of Airtable to be useful. And the key here is you can't automate end-to-end things with AI, right? And that's why I think the agentic AutoGPT-like approach is really challenged because it's just too hard for AI to kind of reason about every single step, break it down recursively, go in and come up with everything it needs to come up with to do something very complicated and end-to-end, especially the more advanced or creative the type of knowledge work you're dealing with.

However, it's already capable of creating immense leverage if the human is the one designing the process and creating different intermediate steps where you can create an AI output, they can see the output, they can edit it, they can approve it, they can do something with it and chain that together with either other human steps or other AI steps or other automation steps, right? And it turns out Airtable is effectively a giant chainable no-code platform of business logic and integration components and data and kind of UI components.

Right.

Exactly.

Yeah, so we certainly have a lot of the simpler use cases. I'll give one tangible example. Actually, multiple very large retailers are excited about using Airtable for translation of product SKUs, right? And there's one particular one that's very, very big and that we're working with very closely to do this. But I think the value here is, what does that mean, translation of product SKUs? Traditionally, you have all these product SKUs as a retailer, right?

So you're selling everything from guitars to toothbrushes and so on. And if you want, on your website, for instance, to show off the product and the description and the name of it in different languages—Spanish, French, et cetera, but also many of the other languages—traditionally you would pay millions of dollars per year to a translation agency, right? Humans would go out and translate every single one of these products, and it could be millions or tens of millions.

And so that's an example of one where the human-designed workflow and the visibility of the process. So, in our case, we allow the customer to actually design, like, hey, here's the human approval workflow. So AI can first of all generate a prompt for AI to run. So you can have a customized translation prompt. And the customization can be based on the specific product category. So you can give the prompt an input that's like, hey, if it's a musical instrument, let me tell the prompt that.

So that the translation prompt knows, okay, this is a musical instrument. Here's upright bass guitar. Now translate that into Spanish, right? And it knows bass does not mean bass the fish, right? That would be very weird to publish on your website. Like, oh, we sell upright fish, smallmouth bass guitars, right? Very confusing. But in this case, being able to chain it into different steps, right, and say, like, hey, we're actually going to generate a prompt step, and then that prompt step is going to feed into another prompt, right?

And then that's going to give the output. And then here's the coolest part: this is a general design pattern we've seen with a lot of our customers, then having AI critique the AI output, right? So have a different AI prompt that takes that translation and says, score this on accuracy, right? And you would think that AI would be biased to be soft on AI output, but it actually isn't. Like, it's actually very good at critiquing its own output.

And then, of course, you can have a human look at every step of the way and either put that into a really nice interface workflow where you just look at one product at a time, go down the list of, okay, here's the original AI output, here's the AI critique of the output. Now I can manually override that or edit it or hit approve if I need to. Or you can filter and report on how many of these outputs are scored by the AI as good enough, right?

Okay, let's check the box on those and then only look at the ones that require manual review. So that's a very simple workflow.

But I think, I'll share some personal examples. I use Airtable, for instance, to track a lot of earnings calls of our customers. So I kind of have my own super CRM of customers that I'm directly engaging with, either as an executive sponsor or I'm in a meeting with somebody there. And I want to have AI-driven intelligence around a customer if I go onsite with them, right? So actually, just a month ago, I was in New York and did a whole customer tour.

And for every one of the customers that I went to, I had basically a composite of internal context. So if the AE left notes about this account and use cases in Salesforce, we have lots of Gong calls with these customers that's all fed into Airtable. And then there's a prompt to basically summarize and extract key insights out of it. So it's not just a bland summary. It might be a prompt that says, hey, tell me if we've ever talked about AI with this customer, right?

Or tell me about the key digital transformation initiatives being driven by the CIO at this company, if that's something that our reps have talked to them about or discovered. And then combine that with external resources like earnings calls, 10-Ks, news articles, et cetera. Have there been management changes? If there's a new CIO, have they talked about how they think about low-code or RAD or whatever? And it basically all combines into this very, very tailored dashboard of all the things I care about from this customer that would have taken me hours to go and manually research through our internal and external context, and can even suggest, like, hey, here's how you might position Airtable to this specific person based on what we know about them and this company and what we know about Airtable.

Like, we can feed in some context about, here's how I like to position Airtable for different personas or as a kind of app platform or for specific use cases. And it can come up with a tailored suggestion of a positioning statement, right?

So that's a killer use case for me. We've actually found a lot of customers wanting to build similar things. Some of them have built similar things. Like, there's a PE firm that has basically been using Airtable for a lot of this kind of portfolio intelligence type use case. Could be a good one for you. But really, the sky's the limit. I think the open-endedness of what you can do with these LLMs, right?

Again, they're not narrow, they're broadly capable. And in some ways they act like experts of many different domains, right? They understand business process within this company. They understand strategy. It's been trained on every HBS article and every great business book. So you can ask it very sophisticated questions, and you can do it with a lot of context that you've put into Airtable, and you can have a really slick workflow interface to read that all.

So we've just seen a lot of open-ended use cases where the customer is pairing the flexibility of Airtable as a platform with the open-endedness of what these LLMs are capable of. And some common patterns emerge. A lot of marketing orgs. But we actually just had somebody from Amazon Web Services present on stage at a conference we did today about how they're transforming every part of marketing operations with Airtable plus AI. In this case, they used our Bedrock offering of AI for themselves, so kind of very meta.

But they're able to go and automate a lot of parts of campaign planning and how they execute on their marketing campaigns with AI because there's a structured workflow and dataset in Airtable, but now they can introduce the AI steps.

So, the very abbreviated version is, first off, from a data security and sensitivity standpoint, any AI option that we offer—we allow the customer the choice of what model provider they want to use. You can use OpenAI's models, you can use Anthropic's models. We have offerings to use open-source models like Llama 2 and 3. We're about to launch our IBM watsonx integration. And that's really great for a lot of customers that especially care about data governance.

There's also some great price-performance ratio stuff there. But in all of these cases, we have the guarantee that data is not trained on or retained by any of these providers. And in the case of Bedrock, because we as a service are hosted on AWS, Bedrock itself is an AWS offering. So there's also kind of a great story around, well, it's all already in our private cloud on Amazon Web Services, right?

But I think there's going to be a lot of different customers who have, as I've gone out and talked to large enterprises, everybody has a different appetite or set of constraints, right? Some companies are very publicly partnered with OpenAI, like Morgan Stanley, for instance. They've publicly talked about working closely with OpenAI. There are other companies that really would rather partner with, or use, models with other characteristics or partner with a company like IBM.

Because of the enterprise safety and data governance emphasis for that platform. So first off, on the model layer, we want to give our customers choice. And frankly, I think it's very difficult, especially for any upstart—or even though we've been around for a while, we have nearly a billion of capital on the balance sheet, and we're now profitable, so that's accumulating every month.

We're still a relatively small player in terms of the resources that we have. If we were to try to go and train our own models compared to now the arms race between really the leading people, like Meta with Llama, Google, OpenAI, Anthropic, and of course there's other ones as well. IBM, very much in the game with both their own models, but also hosting other open-source models. I think it's very difficult to create anything relevant if we were to train our own foundation model.

So our name of the game is really to take the latest and greatest models being developed out there and give it to our customers embedded into the data and workflows that they're already able to shape in Airtable. And it's that proximity to data and the proximity to be able to chain it together with the human interfaces, the workflows, the automations, that really makes it come to life with value. I mean, I think of, like, you've got chat experiences that are very kind of one-off.

Like every time you open a chat session, you're kind of starting a new prompt, and any data context, et cetera, you want to pass into it, you're doing it manually, right? And then on the flip side, you can build a very custom app with AI, but then you've hardcoded the use case and the prompt and how it works. And so no-code and Airtable specifically is kind of occupying this wide realm in between, which is you can customize the shape of the process and where you introduce AI, how you prompt the AI, which models you use in between.

But it's very fluid, it's very rapid to deploy, and you can fill in all of these use cases that are not being served by the other two extremes.

So, yes, in a way. I mean, I think it's almost like if we could have worked for the past 10 years on building the perfect platform to be the deployment vehicle of operational AI workflows, I don't think we could have done much better than build exactly the Airtable product that we have today, right? It is the fastest, easiest way to stand up a dataset or to integrate with existing datasets and to build a human workflow around it, like a human-in-the-loop workflow that's very intuitive.

And now to plug in AI steps, right? Whether it's an AI field output that you can drag onto an interface layout, whether it's an AI automation step, you can kind of make calls to AI services through our embedded serverless hosted code, or use our API to do more integrations. And I think all of that results in kind of an experience where I do think the positioning of Airtable increasingly will be about not just, hey, you can use Airtable to build data-driven apps or have a business app where you've got data, you've got the workflow, but actually the motivator for it is going to be increasingly, by default, hey, I want to automate my workflow.

I want to systematize and then automate my workflow with AI, right? And so, yes, is the short answer, because I think why do you want to improve your operations with Airtable? Well, the fact that you can use it as the fastest way to get AI into that workflow is going to increasingly become the driver to use Airtable in the first place. And my hope would be every customer of ours is using AI. And we actually have two different areas of AI investment.

One is everything I've talked about up to now, is what I call AI at runtime. So, AI to automate steps in a recurring workflow in Airtable. You're basically building an app with AI capabilities as a no-code builder. The other one is AI to build the apps in the first place, right?

And we're also beta testing those capabilities with some of our customers now. The runtime capabilities are now fully GA, but the builder capabilities are like, hey, come into Airtable and just tell it what you want it to build. And it's not going to give you the perfect app, especially if it's very complex, but it can be a really great starting point that inspires you to then riff on it, iterate, and you can use our no-code primitives to then very easily edit the output, right?

So I think increasingly, yes, AI is going to be a very, very central part of the Airtable experience. And I wouldn't want any of our customers to not get the value of AI in Airtable.

Yeah. So this question has come up multiple times, and I have a very clear point of view on it, which is no. And that's not just because I'm biased, but because here's why. I think Gartner and Forrester have talked about this. In particular, I think they've done a good job of describing why no-code actually becomes more relevant in the age of, for some reason, they're calling it TuringBots, which is what they refer to for AI to generate apps, whether in code form or no-code.

It's a little nerdy, but I'll take it. But TuringBots, that's more of the AI-for-building-apps experience. It's going to be a big thing. But the problem is, if you have AI that generates code output, if you're not a coder, it's very hard to inspect that output and either take that and edit it. I don't know if you've tried, but as a programmer myself, I've tried using AI both in the Copilot form, where it's very integrated into your IDE, but also just going to ChatGPT or Claude 3 Opus.

And prompting it with an ask to generate me a chunk of code for something. And obviously the newest models are getting better and better at generating bigger and more complex chunks of code. But it's still extremely frustrating if the code comes out, you have to run the code to see if it works. And then if it doesn't, if you didn't actually understand how the code worked, to try to only use natural language to give it feedback to get to the desired output.

It's an extremely frustrating experience. And so, first of all, I think the fact that no-code by definition means that the output of the TuringBot, the AI builder, is actually in a form that a non-technical user can understand. So the very person who's asking the AI to generate the app can then fully understand its outputs. I mean, I think that's something we take for granted with language: when it generates a language output, of course you understand it, of course you can edit and tweak it.

If it's code and if you're not a coder, then you're kind of out of luck, right? And in fact, even worse, it could be generating output that is secretly doing bad things, right? Like the data is actually corrupted, or it's leaking your data, or whatever the app use case is, it's performing in a very hidden way, which any programmer knows. It's quite common. Those are the worst kinds of bugs and also probably the most common types of bugs, the bugs that escape very immediate visible detection.

But actually are kind of quite pernicious, right? And those happen all the time, and AI is just as susceptible to generating those types of bugs. And I think even if you try stuff where you have a more agentic approach, where you have the AI, kind of Auto-GPT style, come up with a plan to build out a more complex app, and it builds each part independently, tests them, combines them, I think we're a very, very long ways away from getting to the point where the AI can do that well, can do it thoroughly, and not have those hidden behaviors, and also be able to do it sufficiently well enough for any meaningful app.

Right. And frankly, once I think we get there, I think we've actually developed, by definition, we will have developed the building blocks of AGI. So as long as we haven't had AGI, I don't think we will have TuringBots or AI code writers that are able to fully automate, in a satisfactory way, the building of a complex app. The B2B apps that we're building for our customers or enabling our customers to build get increasingly sophisticated and complex over time, right?

And we'll power the end-to-end content production supply chain, really. I mean, it's almost like ERP software, but for a digital output instead of the traditional physical supply chain. It's that level of sophistication as SAP, but for a much more agile digital workflow that we're able to power. And there's no way that AI is going to be able to do that by generating code without human inspectability, editability, and ultimately iteration with the AI.

Yeah, I think it's funny. It's one of those things where there's that quote: we often overestimate what's possible in the next few years, and we radically underestimate what is possible in 10 years. What is that, like Amara's Law, right? Yeah, yeah, who knows? But whoever said it, they sound smart. But I think what's happened is, first off, obviously ChatGPT went mainstream, and I think that's what called a lot of attention to what even that generation of models were capable of.

And it's funny because I talked to people who worked on that model, and I think the wide consensus amongst the model research community, not just at OpenAI but other companies, is, like, I don't think it was anticipated by anyone that it would blow up in such a mainstream way. If anything, it was like, maybe this is like the precursor to the model. Maybe GPT-4 or GPT-5 will be good enough to really be the one that kind of breaks out. But even GPT-3, wrapped around in a pretty simple chat interface, it turns out.

And obviously there was some tuning to work well with instructions and chat commands. But I think the fact that it took such mainstream attention is because it is so broad and so deeply capable, and it was just enough of a glimmer of what was possible. Like, people could try things like upload a board memo, right?

And have it give you feedback. People could have it do fun things like generate me a Dr. Seuss-style story about my kid, right? Like, doing this, teaching them this value, right? Or it could be things like take this chunk of a call, right? And obviously, the earlier models had some challenges around smaller context windows, so that you couldn't upload the whole call. That's expanded a lot over the past couple of years.

But I think what really kind of captured mainstream attention and created all of this hype and excitement was a very real and very justified recognition of, holy crap, these models. I mean, some people have been talking about AI for a while, right? There's always been the pundits. There's been some investors that have been very all-in on AI for a while. But I think this is the first time where anyone, if you're the CEO of a very large company, if you're the board member of that company, if you're a public market investor, if you're just a layperson, you can go and experience the power and the breadth of how it can be useful and see that this has approached something that starts to feel like human intelligence again in bite-sized chunks, five-second chunks at a time, maybe 50-second chunks of human intelligence at a time, but real human intelligence.

As an approximation versus the much more narrow and rote or behind-the-scenes applications of ML from before. Right. And so I think that hype was for a good reason. I think what the world and what enterprises and maybe what the markets underestimated was, okay, there's all this top-down pressure from every CEO of every Fortune 500. I mean, there's not a single CEO I've talked to or encountered who does not care about AI, right? Like, they all know it's real.

Even the ones who were like, oh, this whole blockchain thing, there's all this hype. Like, I think it's BS. And they were kind of not entirely wrong about that. But I think with AI, the wide consensus is like, this is real. It's going to change everything. Now, we don't know how it's going to change things, but we know we need to try, right? And I think the gap between, okay, we know it's real, to like, how do we actually translate this into real impact in our company, right?

There's obviously so many different vectors. There's the customer-facing vector. So if you're a retailer, how do we design AI-powered consumer shopping experiences, right? Like a personalized AI shopper, et cetera. But then internally, there's just, like, tens of percent potential of margin gain or revenue impact from figuring out how to use AI to accelerate internal operations and produce better outputs, whether it's more impactful, faster-to-market marketing campaigns, product development, whatever. And that gap, I think, has not been fulfilled still by either point solution providers who offer AI in a very narrow, like, kind of turn this on and now you get summarization of this call in this feature, in this product as a feature, or separately as a chat product, or separately as a centrally technology-built set of custom applications.

So I think we'll eventually get there. In fact, my thesis is you could freeze all model development in its current state and say, you know what, GPT-5 is never going to come out. Like, there's not going to be any better models. The models we have are all we're going to get for the next 10 years. I think we would still have 10,000 times more value creation possible with bringing those models to the final mile of every enterprise and enabling every group, every team, every function to fully exploit their potential in their actual workflows and their actual first-party data.

So it's kind of a blurry line of, like, when did we first become really an enterprise platform, right? Or an enterprise-grade platform? And you could go all the way back to 2015, when we launched publicly, and I came and did the Data Driven NYC event. And even then, even early on, we were getting product-led growth into large companies, right? And I think I assumed that you would get more virality amongst individual users or SMBs, but actually the opposite was true.

And I think Slack probably saw this as well. We got more virality driven by expansion within enterprises, real large companies, right? And seeing that initial user go to five to 10 to 50 to hundreds to thousands, that actually drove a lot of our PLG. So PLG actually for us happened as much in true enterprise, like really large multi-thousand, multi-deca-thousand-person companies, as it did across SMB and kind of prosumer. And maybe that is also a function of our type of platform, right?

Like, our platform is really meant for groups to build shared applications on versus—you could use it as an individual, but it gets better and better as you deploy it to more people, right? So I think we got kind of pulled into enterprise very early on. And I would say the initial set of enterprise capabilities we built were much more defensive, meaning we'd literally have enterprises calling us and saying, hey, look, we've got hundreds or thousands of people using Airtable for pretty business-impactful or mission-critical use cases.

And we haven't had you go through InfoSec yet. And you haven't talked to one of our procurement people. We don't have an MSA. Maybe there's many different teams at XYZ company who are paying for you on a credit card. We got to get you in officially, right? And so there's the obvious stuff you build in that first phase, like SSO. You build the SOC compliance features, and you gain the capability to at least get approved as a fully sanctioned vendor, right?

A fully approved vendor. And you build just enough go-to-market motion to be able to go through the hoops that these enterprises want. But there's a big difference here because you're still getting pulled in from the early adopters of the product within the company. You're not going and offensively selling to the company. You're letting that bottoms-up expansion happen. And then at some point, you're almost forced to go talk to central procurement and IT, and then you get approved, hopefully, and then you kind of keep going from there.

So that was kind of the first era. When you depicted the PLG era of the company, we were playing in enterprise. We had a lot of enterprise penetration, but it was much more of this: we got pulled in by the users and by the kind of bottoms-up adoption. I think it's about accurate to say in, like, 2021 is when we started getting more serious about being able to actually tell a more strategic story and go on the offensive to enterprises.

And what that means is being able to go to a senior buyer within the company, and even if there is some organic adoption within their org, to not just make the pitch about, hey, well, you've got 500 people already using Airtable in the marketing org of XYZ large, let's say, apparel maker. Wouldn't you like to buy another 500? Right? And here's the trend line of your collaboration growth, blah, blah, blah. Like, that was kind of the very PLG-oriented enterprise sales model.

And what we've been doing, or trying to steer towards, over the past three, maybe even four years, has been about going to that senior buyer and saying, no, it's not just about seat growth. Like, yes, we have adoption within your org, or maybe we don't. But more importantly, we understand your business process, right? So if you're the CMO, we understand the process of global campaign planning, right? And the different phases of that, and how important it would be to consolidate a system of record that entails every part of that process, but still enables each of those groups to kind of act with agility, to have their own control over how they run that process.

They want the flexibility, right? Or it could be for content production, right? Or it could be for building digital products. We've even powered retail store openings for some of the largest, fastest-growing restaurant franchises out there, right? Or luxury franchises. So I think there's this powerful story that we can tell to a senior buyer. And we have to have also the platform capabilities to actually deliver on that, right?

More scalability of data capacity in the product because we're going to support these bigger, more mission-critical datasets, right? Better connectors to the important systems of record, right? So we actually have a Snowflake connector, right? Full circle back to that Data Driven NYC event. Who knew that? The guy presenting next to me, I would soon someday have a product integration story with, or to even more legacy systems. Like, our customers have built integrations between SAP data and Airtable, and we have a connector for Salesforce, et cetera.

But being able to tell that complete story of how we're not just a product that end users love and adopt, but we can actually be the platform that you implement a really important business process on, and you get all of these benefits of consolidation on that platform and end-to-end visibility, agility, the better handoffs between different groups that are working together, right?

So in marketing, you can imagine all the different stages of rolling out campaigns and all the different groups that are responsible for the handoffs and coordination there. It's really hard right now. A lot of enterprises are doing this with spreadsheets or point solutions that don't talk to each other.

I always find these very difficult because my mind goes to, well, there's actually 20 different dimensions. So let's draw out the full space of 20-dimensional kind of where everybody plays. But obviously, to simplify things, I think the easiest way to put it is there's this convergence between categories now, right? And traditionally, collaborative work management lived in kind of this low-end, shallow category. And I think a lot of the players there still are more tactical, team-level, project-management-oriented.

And then there's historically been a category of digital process automation. So think old-school companies like Pegasystems, right, that are really tackling very complex business processes, but in a very brittle way, right? Once you implement something on Pega, you're not iterating on that on a daily or even monthly or even probably yearly basis, right? And it's a very heavyweight thing. And so you're really only doing it for a very small number of high-importance processes or processes where you really have to have it standardized, right?

And then I think you've had a number of platforms that have played kind of in a low-code way. Salesforce and ServiceNow are probably the two best examples. But even Jira, right, from Atlassian, is kind of a low-code product because it's ultimately configurable, right? The data model, you can extend it, you can configure it. There's some extensibility options. So you've got these major players that started with one application. So Salesforce obviously started with sales CRM.

It's in the name. But they expanded to support CRM adjacent to that customer record. They have Marketing Cloud, and they've also built a platform that can be generalized to even more use cases that are not customer-facing, right? And so I think the ultimate endgame, the biggest opportunity, and everybody's trying to go after it, including us, is, well, what about all of these use cases? What about all these operations that are not cleanly defined in an existing vertical category, right?

You've got the SFA CRM category. Salesforce is clearly the dominant player there. ServiceNow has ITSM. That's a very well-understood category. It's existed since before ServiceNow. But it turns out if you actually go into a very large enterprise especially, there's this myriad of business operations that are actually very important to the value chain of this company that are being run in much more ad hoc ways. And so I think of us as being the canvas to build all of these other use cases.

And a lot of them where we've done really well is at the intersection of processes that are very strategically important to the business. So maybe as an extrapolation of that Marc Andreessen statement, software is eating the world, I think every industry is seeing some kind of existential disruption right now in one way or another, right? And it's often digital, right? So every traditional media company has to behave more like a streamer and, in fact, build their own streaming service and play into that in order to survive, right?

Every retailer has to become more of a digital retailer and go omnichannel and think about stores not as the primary driver of commerce, but like an experiential place where they can cross-pollinate buyers who then may transact online, right? And so on and so forth, right? Fintech is disrupting financial services and how you think about financial products, et cetera. And so I think every single industry, and some sooner than others, are getting really disrupted.

And as part of that disruption, they have to stand up a new way of operating, right? It's as if, like, when SAP first came out as an offering, it's because software and the advent of computers and the ability to literally digitize these operations for the first time ever provides so much efficiency and time benefit, right? You can actually start to manage your supply chain faster and with less errors.

And so just the initial software kind of digitization of a very traditional offline set of processes was a huge game changer. I think the same thing is happening now, but it's more around how these companies need to reinvent themselves. And when they stand up those new operations, Airtable is often the platform that they use because only Airtable is fast enough for them to stand up quickly, for them to keep updating and make it agile.

So digital product operations is a very common use case for us at retailers because they're trying to stand up digital product and build a product org that behaves more like a tech company or a tech startup. And how do you power that? Well, Airtable is a great digital product operations platform, right? Companies trying to rethink how they do marketing operations, big brands, big CPG companies, any org with a large marketing operation, you've got to upend how you do that because the way that you're doing marketing campaigns is very much changing.

Channels are changing, personalization and the level of regionalization needed for these campaigns is changing. So marketing operations turns out to be a really great use case. And I mentioned retail store operations and openings. How do you go and now accelerate the pace of how you open stores, especially in an era where there's more moving parts and you need a platform that really keeps up with the pace of modern competition, where it's almost an arms race where if you don't figure out how to digitize and do these things faster and more efficiently, your competitors do.

And the pace of the game has just been set at this new benchmark, right? The most aggressive, fastest-growing businesses in every industry are moving so much faster than any companies were 30 or 40 or 50 years ago. And so I think we really enable companies to move their most important, fastest-moving operations onto a platform that keeps up with them and, in fact, allows them to customize but also iterate on that process and increasingly add AI into it.

And by the way, just in the spirit of vulnerability, we've definitely had our hiccups, right? I mean, in full disclosure, we've done two RIFs. We're now profitable, we're growing. We actually never stopped hiring throughout. So it's been very important that we continue to grow, we continue to execute, and we continue to hire the right people to execute on the next phase of the company. We've been shipping products, we've been growing revenue every single month of our existence. But in no way would I say that we're omnipotent.

We've figured everything out. We've always—I think there have been, like—

There might be some hiccups along the way, right? Everybody's seen the Jensen video where he talks about, "If I knew how hard it was to start a startup, I would never do it." And it's kind of true, right? Not like—I love my job, and I think there are so many fun parts of it, and there have been different learnings at different parts of it. And that's what keeps it exciting for me, is, like, it's always a new job.

But, just very, very bluntly, there are parts where it's like we didn't figure things out super ahead of the curve, and in catching up to it, we either lost some precious time to execute or we had to make some difficult decisions to kind of change our course slightly, right? Oh, we thought we were pointed this way, and actually we need to go slightly this way, right? Whether it's rethinking how we execute, the way that we structure our teams, the priorities—the product org actually has gone through kind of an org structure change where now we have these pillars that I think are now really well aligned to the strategic priorities of the business and even aligned to revenue drivers for the business, while still able to have a high degree of autonomy to innovate.

So one pillar, as an example, is solutions. And this is responsible for thinking about specific use cases on the platform and increasingly productizing a solution SKU around those. So we're going to launch our first solutions this year. We've had templates, we've had blueprints of use cases before, but this is our first kind of equivalent of Sales Cloud CRM for Salesforce on the platform that we've always had. But we also have a pillar for AI. We have a pillar for teams, which is more about the onboarding, self-serve, and kind of the journey for bottoms-up.

And we have an enterprise pillar. And I think the point here is it's taken us—you have to not only figure out the strategy, be clear about that, but then also design an organization that is really well aligned and accountable to that strategy. And so I really feel like we've gotten to a very good place, and we're executing, like, we're humming along, it's working. Customers are really loving everything that we're building.

Like, when we talk about the product roadmap to them or show them our AI capabilities, we, for instance, just had, two days ago here in New York, an AI workshop where we had really incredible customers represented in the room, like, from Fortune 500s. And they literally got to build with our AI capabilities really exciting use cases that, honestly, they all went home so, so excited to see what else they could do with AI in Airtable. So I think what we're working on is now very aligned to what enterprises want and what they're actually able to start using.

But it wasn't overnight. I think we definitely got there through a series of different changes and clarification of what are we focused on, what's our position, what's our platform story? And ultimately, how do we execute as an organization against that?

I think it's a moment in time for every company, which is not to say you have to be PLG to start with, but I think every PLG company, and this is true whether you started in ZIRP or after, I don't think it's going away, but I think if you start with PLG, you will eventually run into an asymptote of PLG-driven growth alone. And I think the answer of why is sheer math, right? I mean, if you're Facebook, you can go PLG for a very long time because every single person in the world that has internet can be a user, right?

And so you can literally see exponential math work at the top of the funnel. Like if you had 100,000 people sign up in your first year, you can have a million sign up the next year, and then 10 million the year after that, and 100 million, and so on. Obviously, at some point you saturate the world's population. But I think it's very different for B2B companies, right? Because first of all, there's only so many companies out there, right?

And second of all, there's only so many companies that will hear about you, will have the proclivity to come and use your product. And I think the viral loops aren't quite the same in B2B as they are in a Facebook-type, very network-effect social graph company. But all that means is you might still be growing, but I think there's a natural linearization of every top of funnel. And this has been described by a few other people.

Oliver Jay, who was the CRO of Asana and before worked at Dropbox, has a post called "The PLG Trap." I think roughly, I agree with most of the points, which are PLG is an awesome way to start building a product, right? Especially in B2B. Not the least of which because not only is it a very cheap source of growth and revenue, you don't have to worry about, like, is my CAC/LTV break-even, right?

I mean, the CAC is zero effectively, right? But additionally, it's almost like it's a forcing function to prove that your product is actually great. You have to have true product-market fit. There are other types of B2B companies where you can kind of fake it for a while by aggressively selling a shoddy product to customers. And then the writing's on the wall with churn and so on of, like, this is not a sustainable thing.

But I think with a PLG-originated business, you have a certain amount of confidence that this is a good product and people like it and people are using it and it's not vaporware, right? And so I think it's a really powerful starting point, but I think you then have to connect the dots to ultimately, well, how do we make sure that this is also valuable to the real decision-makers within the company? And at some point you run out of steam with just individual points of light within a large company.

You have to go either because you're forced to, because the CIO is going to ask, like, wait, we have like 2,000 seats of this, but scattered across all these different places. How do I make sense of this? And why is this a valuable product to us? You better have a good answer to that. And the more that you can actually elevate as a platform to offer value-add, rather than just sell on the FUD of, well, if you don't pay for the enterprise license, you're not going to have control over all this usage.

If you can actually elevate to a value-add sale where it's like we give you end-to-end connectivity between all these different groups that were working in silos and they're able to move faster and it can be centrally managed but also locally iterated on, and execs get end-to-end visibility over this really important process, right? I can see all of my store openings as a restaurant franchise, and every single stage of that is managed in one place. And so I can see literally the end-to-end.

It's like the ultimate 10,000-foot view of my business. That's compelling, right? And that's CEO-sellable, not to mention CIO-, CFO-, COO-sellable. And so I think if you can make that transition to almost ride the wave of PLG, but then instead of waiting for the wave to flatten out, jump up to that bigger narrative of, and also we are strategically important to executives, and here's our story as an enterprise-level platform, then I think you can really win it all, right?

I mean, that's my belief. So PLG is fine. It's a great starting point, in fact. But I don't think it's just ZIRP, although ZIRP certainly fueled an extension of that PLG runway. But I think it would have flattened out for every PLG company regardless. It would have linearized at some point or another just because of sheer math. Like, you cannot keep doubling, tripling, 10X-ing, or even 30%-ing the top of funnel indefinitely in B2B.

Well, I'm going to answer that in a flashback style, which is to say, I'll start with the conclusion first and then go through the story. What I wish I knew at the time of each of these phases is to take advice from people who are as close to the next phase, the next immediate phase of the company and the problems that we need to solve, as possible, versus taking advice from people who are too far removed, either in the forward direction or in the back, right?

But there was a time where I tried to really learn from as many people as possible. And I think one thing I learned is, maybe like a neural net, if you feed too much noise into it, it actually gets confused, whether it's feeding it into a prompt or in the training data. It can actually create more confusion than it creates coherence. And so I think being very selective about who and what you learn from at every phase—and you do have to come in with an opinion about it—it's like shaping the source of learning for yourself.

And so you have to come in with some thesis of the type of organization, business, product that you want to build in the next phase. Meaning, if you went in and talked to like 50 different really smart advisors, and they could be other founders, they could be VCs, they could be operators, you could talk to 50 different best-in-their-own-class people, some from consumer. You could talk to the brilliant heads of growth at great consumer companies.

You could talk to CEOs of really large enterprises. You could talk to X and Y and Z. But ultimately, and let's say it was us, right, going into that PLG-to-enterprise transition, if you're not coming in—if I'm not going in—with a clear point of view of, I think we need to become enterprise-oriented, and I think what that means is actually that we have to have that strategic value prop to sell to the buyers, right?

Then what you end up getting is a scattershot of different perspectives that, if anything, just pulls you in a million different directions. Or even if you over-gravitate to one perspective that isn't aligned with the direction that you've set for the business. So it's a little bit of having opinionation going into asking for advice that I think is really important. And I think ultimately that's the only secret, right? Be very good at asking for advice, but also be very good at shaping a perspective on who you ask for that advice, where you ask, and allowing that to kind of create a positive reinforcement learning technique.

Or approach of your own brain, right, at every phase. So I think a lot of it is like, these problems have been solved for each of the phases that you described of Airtable, right? Every company initially has to just figure out product-market fit. And that's more about just figuring out what's a problem, a TAM, or a category, or just a specific type of customer problem that we can solve. And then, non-trivially, how do you build a product that solves that well, and in a way that isn't an overfit of one customer, of one instance of that problem, right?

Okay, you got that. Then how do you go and turn that into a business? How do you commercialize it? What's the revenue model? Have you thought about that? Maybe you thought about it upfront. And when you do have that revenue model, it could be self-serve, like we had initially. How do you optimize it? How do you make it keep working really well? How do you think about upgrade flows if it is self-serve?

And if you want to add sales touch, how do you do sales, right? And then when it goes to shifting from PLG to enterprise strategic platform and sales-led platform, how do you do that, right? And so I'll give very tangible examples. I think some of the best learnings and evolution I've had as a CEO in this current phase are from really great enterprise operators who very intentionally have very parallel experiences to what we're trying to do.

So David Schneider, who was the president of ServiceNow from, I think, a few hundred million in revenue to many billions in revenue, and really was the architect of a lot of their go-to-market systems and thinking, has been an advisor to us through Coatue for years now. And likewise, George Hu, who was the COO of Salesforce—and I've talked about Salesforce and ServiceNow as platform companies—it's no accident that George, as well as David, are both very close advisors to me.

I consider them mentors, and have really helped shape a perspective on how can we build this business into the next phase. And at the same time, though, neither of them expect me to literally take exactly what Salesforce or ServiceNow did verbatim and apply it to Airtable, because we're obviously a different product. We have PLG, which neither of those companies really had, especially at the later phases. And also, it's a different era. It's a different phase, right?

Like, what worked back in '99 or 2002 may not apply now, right? And so it's not a one-for-one translation, but it is like taking inspiration from the right and most analogous sources of learning.

Thank you very much. This was fun. Good to see you, Matt.