Zapier CEO: Individual AI gains won't make a company more productive (Wade Foster)

I had another CEO friend say 10 out of 10 fun and 10 out of 10 anxiety right now. I was like, "Yeah, that kind of resonates." It is not hard to find individuals that have, quote unquote, 10xed their productivity with AI. It is much harder to go look around and find the companies that are 10x productive because of this. Why is that? We have the technology. We don't know how to build the organizations around it quite yet. There's going to have to be a lot of changes, not just with adopting the technology, but also rethinking how these organizations work from the ground up.

Yeah, thanks for having me.

I did not. I grew up in central Missouri, Jefferson City. It's the capital of the state. And my exposure to jobs were you could either work for the state, you could be a doctor, a teacher, police officer, right? As a young kid, I'm like, "Man, the number of jobs that exist is very, very small." My worldview was quite, quite small. And so I didn't really get exposed to entrepreneurship until much later. And what really kind of was the wake-up call for me was the tail end of university when the financial crisis hit.

It didn't matter. No one was hiring. No one was taking jobs. And I think that was when I realized like no one's waiting to give a handout to me. Even if you do a good job, you kind of have to make your own path. And I was fortunate enough to find a local entrepreneur, had a small software shop in Columbia, Missouri, and got hired as an intern there. And that's where I fell in love. I was like, "Oh, building software is really fun." And started to have that itch where I'm like, "Man, I think I would really like to start my own company someday." So, this is 2011. SaaS is on the rise. This is the year that Stripe launched, Twilio launched, things like that. And if you would go to the forums of these companies, you would see a very common type of customer request. And the customer request would always be, "When will you integrate with X?" right? And it was followed by a bunch of customers saying plus one, "Me too. Yes, I want that." And eventually a product manager saying, "Hey everybody, thank you for the feedback. Seems like an interesting idea. We'll take a look."

Right. And so Brian sees this and is like, "Hmm, I think we can make sort of like a nice, simple UI so you don't have to know how to use the APIs and connect this stuff." My day job, I am working with a Marketo API doing email marketing, and I'm a bad engineer, and so I'm having a tough time. So anyhow, we've got this local hackathon coming up, Startup Weekend. We go, we team up with Mike, we holed up into a garage in the back of this building all weekend, and came out and had a prototype. It worked. It demoed well. So ended up winning the hackathon. And then, after the weekend, it was like, "Well, what do we want to do with this thing?" We're in central Missouri. It's not like there's a lot of investors in the area that are just jonesing to put dollars towards a bunch of just recently college kids. And so we're like, "Okay, well, we want to keep working on this. We're having fun." So we said we'll do it nights and weekends. Kept our day jobs and we'd meet up after work. We'd stack some Hy-Vee Chinese, which is the local grocery store, so just get some takeout, and then we would just get to work.

I had this gig at this local software company which was doing some natural language processing. In fact, it was a professor trying to automatically grade student papers, which now seems quaint. It's like ChatGPT could do that. But in 2011, it was like, how do you do this? And I remember trying to sell this product, and even though the product was slick, well-designed, good engineering, I just couldn't get anyone to buy this thing. I just thought I was bad at my job. That was kind of like my recent history is like, "Man, selling software is hard." Then with Zapier, I remember the first customer we had, this guy Andrew Warner. I had emailed him. He ran a podcast, Mixergy, at the time. And I come across his name in a forum where he was looking for a PayPal-Highrise integration. Just emailed him and said, "Hey, you still looking for one of these?" And he said, "No, but did you build a plugin for it or something?" And I said, "Well, sort of. You can hook up PayPal and Highrise." And then I did some snooping and found what other tools he uses. Realized he used Wufoo, AWeber, and a few other things and said, "You could hook that stuff up, too." And he pretty much bit hook, line, and sinker. I was like, "Oh, I need something to take my Wufoo contacts and put them in my newsletter that runs on AWeber." And so I was like, "Okay, great. We do that." And I was like, "Brian, Mike, build the thing, right? We got to go." So give him access to the thing. And the next thing I remember is he's like, "Wade, this looks cool, but I can't get anything set up. Can you get on Skype and show me how to do it?" I'm dating myself with all these products. I'm coaching him through setting all this stuff up, and every step of the way the product just barely works. It is rough. We've not been working on it but for like a few weeks at this point in time, so it is not very good yet. But we get to the end, the zap is turned on, and we go to test it. So he fills out his form on Wufoo, hits submit, and we go refresh the AWeber page and see like, hey, is there an email address in there? Sure enough, there's the email address. And his reaction is like, "Oh, this is incredible, Wade. How much money do I owe you? This is going to save me hours every single day." And that was when I knew. I was like, "Holy cow." This product kind of stinks.

It barely works.

Whereas in the past, I was working on this quote unquote slick software, the software that's good, well-designed, all this other stuff, and couldn't get a taker. And so I was like, "Man, if we could just get this product to not totally stink, I think we're on to something." It was like early December that this happened, and by January, I'd quit the day job and was like, "All right, let's go figure out how to go faster on this thing."

Yeah, we basically had noticed a couple things with GPT-4. First, it came out roughly six months, I think, after 3.5. Two, the capability improvement was obvious. Just talking to GPT-4 was just a lot smarter than 3.5.

And the cost had come down meaningfully.

And so it was one of those things where it felt like, you know, what happens when 4.5 comes out? What happens when 5 comes out? Does that happen six months from now? What happens if it happens four months from now? Are they going to launch these things faster and faster? Are they going to get better and better? Are they going to get cheaper and cheaper? Even if one of those things is true, this is very impactful to the whole industry and to what we work on. If all three of those are true, then this is the most important technology that has maybe ever been invented, certainly since the internet, certainly in my active career. We weren't exactly sure what to do with that yet, but we were like, we need people to pay attention internally. And we never called a code red. We didn't even know what a code red—no one even had any sort of like, "Oh, when Wade calls a code red, here are the steps that you take." It was just like, "Code red!" And people were like, "Okay, I hear that. That's important. What do we do?" And so we were figuring out what to do. I remember teeing up a handful of things that we were like, "Okay, this is what I want to change in terms of how we operate." A lot of it was like, here's how the product needs to change, here's how operations needs to change, here's a list of stuff that's really important. When I look back at that time, though, I would say there was really one thing that stood out as being more effective than anything else: we paused the company for an entire week and ran a hackathon. Didn't matter if you were engineering or not. It's like everyone's going to put your hand on a keyboard and you're going to build something with AI. And at the end of that week, we went from about 10% of people using AI as part of their daily job to over 50%.

And as a result, more and more of the company started to build an intuition around what is this technology, what is it capable of? How might we build in tandem with the technology for our customers? How can we use it ourselves to improve our own operations? And I think that just company pause where it was like, "Hey, we need to just go do some sense-making together," I have since seen be productive for almost any other company that does one of these. They all cite the hands-on learning time being way more impactful than the CEO memo or the AI committee or any of that kind of stuff.

Yeah. Well, I think some of this was a little bit fortuitous, right? In 2011, we set out to just integrate stuff. We're going to make integrations really easy for people. And so our integration story is 15 years old. We've just gotten really good at integrating any tool that you have on the internet, far better than anyone else. Pretty quickly, we started to notice, as did many others, that these AI models are incredibly smart on a bunch of textbook information. They're trained on the internet. And so if you want to ask it like, "Hey, how should I think about a marketing campaign?" or "How should I think about this strategy?" or "How should I think about building a product like this?" it will give you a very generally good answer. But it kind of leaves you wanting. You're like, "Well, I don't want a generally good answer. I want something that's great for the environment that I operate in." What a lot of folks started to notice is if you gave it that context, if you let it see your CRM data, if you let it listen to your closed-won and closed-lost Gong calls, if you gave it access to your email, your Slack, just all this stuff, and then started asking questions and then started building automations, you got very specific and very good answers—stuff that you start to go, "That's as good as I'd get from a good teammate internally, or maybe better in some cases because it's able to read over all this stuff that most humans can't do and hold in context." And so that was where it was like, "Oh, Zapier really has a role here because we can bring the integrations into these model capabilities and help you orchestrate the workflows, the agents, however you want to sort of go build those out." And so we sort of just doubled down on that. In the past, the integrations were all done through APIs. Well, now there's this new thing, MCP you called it, stands for Model Context Protocol. But effectively, it's just a way for agents to talk to other agents. For the most part folks, it doesn't matter to know much more than that. But the nice thing you can do about it is it allows you to talk in natural language to an agent or to something else, and then behind the scenes the AI knows how to convert that into structured requests, aka an API call or something like that.