Hi everybody. Good afternoon, good evening and good morning wherever you are. Thank you very much for joining the bottom club today to our builder series featuring executives from top tech companies revolutionizing the world as we speak. My name is Yos Madia and I'm joined also by a good friend of mine, Vargas.
Hi everyone, thank you for joining us today. So today we are fortunate enough to interview none other than Wade Foster, co-founder and CEO of Zapier, a company that's revolutionizing how teams automate workflows and connect the tools they use every day. Zapier's story is incredible as they've reached several billion dollars of valuation while raising only something like $1 million, which is remarkable, and they've done it also fully remote, but we'll get to it later on. So Wade, thank you so much for joining us today.
Yeah, thanks for having me folks.
The pleasure is ours. So Wade, you started building Zapier in Columbia, Missouri out of all places and you actually mentioned that that pushed you to build a profitable real business without the famous Bay Area capital. But in today's AI gold rush with companies raising hundreds of millions of seed or Series A, is seed strapping still a competitive edge or is AI simply too expensive to do it the Zapier way?
I think it's even easier to do it today. And so, yeah, we do see some of these AI researcher, AI foundation labs, neurolabs, things like that that are getting huge amounts of seed round funding. But what I see a lot more of is companies that are either bootstrapping or seed strapping, kind of the Zapier style, raise one round and then try not to go to raise anymore. And I think it's a lot easier because of how good AI coding tools are. You can build a lot more a lot faster. And so if you're an entrepreneur with an idea and a good understanding of your customers and know how to sell, know how to get some distribution, I feel like it has gotten easier. I don't want to say it's easy, but I think it's gotten easier to do this. I actually heard about the story of a guy that was the only employee on his own company, Base 44, and then he sold it two weeks after like eight or six months for $80 million. So perfect example of being bootstrapped and it's only because of AI. So it's amazing to see and I guess maybe one day we will even see a unicorn out of one guy's bedroom or basement.
I think it's quite possible. Yeah, maybe it's even happening today and we'll hear about it in a couple of months but let's see. And indeed in the early days how did you balance the customer development with the product building especially when you were still figuring out who are your user ICP?
Yeah, I think this is where having co-founders is so helpful. I'm fortunate to have two co-founders, Brian and Mike. Exceptional engineers, exceptional product thinkers. And so the way we kind of divvied up the task is I was out there, you know, posting in forums, trying to drum up interest, trying to find people who could be potential customers. I'd be showing off the product over Skype. I'd be taking copious notes saying, 'This is working, this isn't working.' Things like that. And then relaying those back to Brian and Mike and saying like, 'Hey, how do we go do this?' And so they're out there building away trying to make the product better while I'm out there soliciting all this feedback and trying to sell and get that going. And so we kind of had these two parallel loops running. And as a result, I think it helped us go pretty quick.
I agree. And something very interesting that you actually worked fully remote from day one. Even now when you have hundreds of employees. So now that we see companies coming back to the office, has this trend affected somehow the strategy in any meaningful way or you still going all the Zapier way all the way?
We're bought into the distributed way of working. Yeah, I understand why people like being in office. I think for us being distributed is an advantage and I don't mind all the return to work. It makes it easier for us to recruit all over the world. So we salute you the return to office crew. Thank you.
Zapier was never the story about following the herd, right? And something very interesting that the choice to remain remote even after joining Y Combinator, but what organizational habits or systems were the most important to make that work from day one? I mean how do you manage hundreds of employees in every different places of the world? If you need to write us a playbook what's going to be the main chapters there?
Well, you know, I think the benefit we had was that we didn't start with a hundred employees fully remote. We started with three founders. And so we had a couple mechanisms that helped us. One, it was three people who deeply cared about the customer and the business and the product who were communicating all the time over chat. So we were just sharing here's what we're learning, here's what we saw today, here's a problem I ran into, can you help, etc. So, we're getting used to documenting all this stuff out in public in chat so that we're all learning from each other. The second thing, we had a philosophy. Don't hire till it hurts. This is really valuable because then when you go hire someone, you know exactly what you want them to go help you with. You know, this is the problem. This is where the pain is. Come help us do a great job at this. And so, when they come in, you can really tell, are they solving that problem? Because the pain goes away if they're doing a great job of it. If they're not, the pain is still there. And so you can tell very visibly what's going on. Then the third thing was as you're onboarding those folks, indoctrinate them in the culture. Make sure that they're getting used to communicating in public, sharing what they're learning, sharing what they're having a tough time with. And as a result, you build all this documentation, all this transparent work about how you know your customers' problems, how you think about the product, all that sort of thing, which then helps other people that are now coming along behind you. They can start to understand the decisions. They can see the way of working and you kind of rinse, wash and repeat. Now there's one other benefit to this too which was not there when we started but today the big benefit you get of doing this this way as well too is it's not just humans that benefit from seeing all this documentation and this discourse that's happening inside of your documentation tools whether it's Slack or Google Docs or Notion or Coda or things like that. It's also AI agents that are now slurping up all this information that's really valuable again for them to go do work for you or for humans to ask answers of those AI answer engines. So you get a lot of benefits from working in this way and that you know if I was to like reduce it down to a very small set of things that it worked I suspect it's those tools.
I agree and I guess it was even more beneficial in the COVID days when you see the craziness of companies suddenly hiring tons of people and then they need to fire them. So it's not also very good for the culture of the company itself. Okay. So Wade I want to shift the gears a little bit to your products of Zapier because you launched very sophisticated AI-driven tools like the Zapier agents and MCP but when did you realize that Zapier needed to evolve from automation to intelligent automation?
We were starting to dabble with AI in our products a few months before ChatGPT launched. My two co-founders, Brian and Mike, had this inclination, especially Mike, I think he gets a lot of credit for this, that AI was going to be a really important technology and have wide ranging implications on what Zapier could do and deliver for our customers. And so Mike and Brian moved into individual contributor roles in the company. They stepped out of leadership jobs and said, 'Hey, we're going to personally put our hands on the keyboard, start building stuff, figure this out.' Wade, your job is to go, you know, run the business and manage the Titanic, so to speak. And, you know, I think for most people at the time, they looked at that and was like, 'Ah, that's just sort of weird co-founder curiosities. They're just, you know, off in the garage doing mad scientist stuff. Pay no attention.' Right? And I remember they built a text bot where, you know, it was a phone number. You could text the phone number and ask it questions. And you know, I remember asking it. It was around my wife's birthday. I was like, 'Hey, what do you think I should get her for my birthday?' And it was giving me some suggestions back. This is GPT, I think three at the time. And it was like kind of a cool experience. You're like, 'Ah, this is pretty fun.' Of course, a few months later, ChatGPT launches. My reaction was like, 'Oh yeah, this makes total sense. Of course, this should exist.'
Well, it's funny you say code red. We didn't quite actually go straight to Code Red at that moment. Instead, you know, we were just sharing it with the company. We shared it in general channel in Slack and saying, 'Hey, here's this cool new product. We think it's pretty cool. We think this might have interesting ramifications for Zapier. We also like new products. We think you like new products, too. Go play around with it. Have some fun with it.' So, it was very suggestive. I don't think we had really gotten the conviction yet to be like, 'Code Red, all hands on deck. Let's go.' But then what really changed was over the course of the next 6 months, a few things happened. One, we started to see a few internal experiments and customers start to dabble with putting an LLM step in the middle of a Zap and getting really interesting results. It was starting to be pretty powerful the types of use cases you could do. Two, we had invited a customer to come out to our all-hands summits that are all in person. We invited a customer that was using a lot of AI and he was sharing a lot of what he was seeing and what he was benefiting from and how he felt Zapier could adapt to this world and you know I think he was both positive and optimistic about what Zapier could do but he was also challenging us. He was like hey I think if you all are not paying attention to this stuff it could potentially disrupt you. And so you know that was planting a seed in our head and then the final thing was the GPT-4 launch. When GPT-4 came out, it happened 6 months after ChatGPT came out and that speed was one thing we noted, but two, GPT-4 was a lot better than GPT-3.5 and you could feel the power of that model. And as a result, we started to say, well, that's interesting. What happens when 4.5 comes out 6 months from now and then five, you know, six months after that? That was a code red moment where we were like, we are not taking this nearly seriously enough. And so we paused the whole company for a week. We said, 'Okay, we're going to run a companywide hackathon, not just engineers. We want everyone to be building with AI, using AI, playing with the technology to really build up our understanding of how this impacts how we work and the types of products that we build for our customers.' And so that became like a really pivotal moment in the company to get us more anchored on where to head. And you see the results of that today. You see the power of tools like you know you can put AI steps anywhere inside of a Zap. We have Zapier agents that are almost these purely autonomous tools that you can give it goals and instructions and directions and it will go attempt to solve those for you. You see that with Zapier MCP which allows you to hook up tools to an LLM whether it's in Claude or Cursor or Claude Code or ChatGPT or things like that. So all these product experiences, I can rewind the clock and look back at that moment and say we set a lot of that stuff in motion that week.
Yeah, 100%. And it's funny because I remember when ChatGPT came out, people just like said, 'Oh, it's cool. We can write funny songs with rhymes and that' and then suddenly it changed the whole world few months afterward. So it's incredible. And you did mention the MCP which now users can access Zapier's 8,000 plus apps directly inside tools like Claude and ChatGPT making Zapier feel much more like a connecting tissue of all the AI ecosystem but can you walk us through specifically the process what were some of the biggest maybe technical challenges or key insights that you learned along this way initiating the MCP?
Well, it's interesting you ask, you know, what are the big technical challenges? There definitely are some, but I think the technical challenges are far dwarfed by cultural challenges. The big thing you have to do is, you know, folks are sort of heads down working in a particular way. They have goals. They have OKRs or KPIs, road maps that are there to be delivered on. And those goals, those road maps, those have likely been set in motion by you. So you said these are important. We need to go achieve these things. These things matter a lot. And now you're waking up and saying, 'Oh, and also this other thing matters.' And now you got to figure out how to get the culture to go from one world that you've prior been championing. And honestly, probably some of that still matters quite a bit to this new world that also matters a lot. And it's like, well, what if the old world moves to the new world? What if the old world needs to disappear? And so there's a whole sense-finding exercise that you have to go through. And you know, I'll be honest, like six months, a year, some very confusing, chaotic times inside many of these companies going through. In fact, a lot of folks are still going through it. And I wouldn't say that we are, you know, totally through the other side on this. So, the cultural challenges, I think, are first and foremost. It's getting people to be comfortable using the tools. Getting them to understand how to use it as a daily part of their workflow. Understanding what are the capabilities and how do you build products that are not deterministic that have a probabilistic outcome. So there's all that sort of stuff that's going on. Now if I was to point to the technical challenges, I think the real ones are, you know, around big legacy monolith code bases. There's a lot of excitement for vibe coding and these tools are exceptionally good at green field projects, but if you have a, you know, 12, 14-year-old codebase, they're not as good at working inside those tools. Now, the models are getting better, but it still is a big effort to make it possible to get really good, really valuable results out of those endeavors. And so, you know, that's certainly not a solved problem yet, but I do think it's getting better every day.
Absolutely. What I will take is don't fall in love with your product because things are moving blazingly fast. So, you need to adapt as fast as possible. And specifically, you mentioned vibe coding. I know that it's all about the intent of the agent. So whatever you build because otherwise it's build something really fast but as you mentioned to take it to a 14-year-old database it's something else and also everyone talks about AI so I guess we need to talk about trust right because AI is now embedded across internal workflows and customers' products but how do you ensure that automations remain trustworthy and doesn't create let's say fragile or hard to maintain systems for example even avoiding hallucinations and...
Yeah so I'll give kind of two answers to this I think again there's a cultural answer and then there's a technical answer so on the cultural side I like what Brian my co-founder talks about he says hey you can delegate the work but you can't delegate the responsibility and I think this is really important for how humans interact with these AI systems. Humans I suspect are going to have really valuable roles in the future on two sides of AI. One, it's what do you want the AI to do? What ideas do you have? What problems do you want to solve? What experiences do you want to exist? And so, you're often kicking that problem off. The second place that a human is really valuable is on the other side of that. Okay, the AI has now generated an outcome. Well, is it good? Is it correct? And so humans have to take responsibility for having high taste, high judgment, high evaluation of those outputs. So those are like cultural tenets that I think are key to instill inside your organization. Now on the technical side, I think there's a lot of things that people are working on to help with this. One, we try and move as much stuff to deterministic workflows as possible. I think there's obviously initially a lot of excitement around these probabilistic outcomes, these agents to do these things, but many workflows especially inside of an enterprise setting that you want a lot of reliability and you want low cost. If it can be done in a deterministic way, you probably want to do it that way because it's going to increase the overall reliability of the system. The second tool that you see everyone using these days is eval. You cannot run a production AI system at scale without really good eval. Otherwise, you're just sort of sticking your finger in the wind and just evaluating, you know, feels good. Feels like the wind's blowing that way. I think we're on the right track. But these evals are you getting consistent, high reliable output from your systems that you've designed. And so I think that is especially valuable as well too. And ultimately, I think we're going to see the emergence of the agent inboxes, human inboxes, where humans are now almost like, you know, in the past, we had an email inbox where, you know, we're replying to humans and humans are sending us tasks and we're going back and forth with them. I think in the future we're going to have an agent inbox where agents are sending us work and asking for feedback, asking for us to evaluate it and we're going to just be turning through that inbox and saying, 'That's good. Do more of that, that's not so good, try and make this better, etc.' And behind the scenes, there's going to be a bunch of statistical models helping us improve the overall reliabilities of these systems over time.
The last thing that you mentioned that we will talk with the email rather than with someone else. It's very revolutionary. Maybe another idea for a startup. So, who knows, right? I do have one last question before I pass the mic to Amato. So, for decades, Zapier is known as the poster child for product-led growth, right? And it did work for a very long period of time. But now I do know also that you moved sort of enterprise-led growth. You still have the product-led growth but at the same time also a dedicated sales team from a cultural perspective. If you can share with us some of the adjustments that the company needed to be doing in order to not be 100% self-served.
Yeah, I think this is one area where I've probably had the biggest learnings and probably changed my mind the most over almost 15 years of running the company. I love self-served businesses. They're so much fun. You know, the fact that you can put a product up on the internet, get good distribution for it, a customer can adopt it and pay you without a human in the loop, it honestly feels like magic. I still remember the first time like it happened for us and I was like, this is the coolest thing ever. And so it's really easy to get very addicted to that type of business model. Now the challenge that we found with Zapier is that that works really well up to a point. Zapier is this product that is easy to learn but hard to master. It has a lot of depth to it. And so there's a lot of powerful things you can do with the product, but those things are hard to uncover. And so in a self-served model, you get a lot of people coming in and adopting the easy to do stuff. But the value both that we can deliver and the user can get out of that easy to use stuff, it's somewhat capped. And so the learning that we had is we can really help organizations, particularly larger organizations because those tend to be the ones with budgets to spend and have these bigger problems. We can actually help them solve bigger and bigger things if we put a sales motion in the loop. And I think I had to learn over time that sales isn't a four-letter word. It's not a used car dealership and things like that. That's not great sales. A great salesperson is someone who deeply understands the customer's problems. They understand how your product can solve those problems and they act as a guide. They say, 'Hey, let me show you what is possible here.' And you know, as a result, you're able to go into these organizations and solve bigger and bigger problems, which is great for the business, great for the customer. And if you have a really strong PLG engine, I guess you don't have to do this, but shoot, I felt like we were leaving so much opportunity on the table by not using all these customers that already know and love the product and saying, 'Hey, let's just go help you solve even bigger problems with this product.' And so that's the learning how to do that has been I think one of the more important things I've learned especially in the last 5 years.
