Code Red: How Zapier hit 50% AI adoption in one week, with Wade Foster

In moments like these, there's a lot of incumbents that have gotten fat and happy off of these big fat margins and they're used to operating a different way and that's what kills them. We just like jumping in the deep end and saying, 'Let's go figure out what this thing can do for us.' We had never done a code red before. To me, all it meant was like, 'Holy crap, this is important. We need to do some different stuff.'

I'm good. How are you, Greg?

I am joining from my home in central Missouri.

I lived in California for 10 years.

I came back for family and started having little kiddos and aging parents and all that stuff.

Yeah, founded the company in 2011 with two good friends, Brian and Mike. We were working at this mortgage company in Columbia, Missouri. Brian and I played in a jazz and blues quartet and we'd pitch ideas back and forth. And one day Brian had spotted that if you would go to like the popular forums of these SaaS products at the time. Think like Salesforce, QuickBooks, Evernote, Dropbox, there was a very common pattern which was when will you build an integration with X? And there would always be plus one, me too, I want that as well. And then a product manager would chime in and say, hey, thanks for sharing the idea. We'll take a look. Which if you've been inside these companies, code for probably not going to happen.

And so Brian saw that and said, 'Hey, I think we could make a really simple tool that would make it easy for people to integrate these products without having to know how to use the APIs.' And so that was the very original nugget of Zapier.

Yeah. So all three of us still on the board, all still working very hard to make Zapier a success.

I think that's a pretty darn good bet. If I was to try and attempt to take the counter position, it would probably be something like in the future everyone is just so equally good at doing this that it ceases to be a competitive advantage. But I have a very tough time seeing how that is the case in the foreseeable future. Based on what I see on the playing field, there is a big disparity between the people who are being successful with it and the people who aren't right now.

We're certainly proud. I think running a company can be pretty humbling though.

We have five values internally that we spend a lot of time advocating for. Zapier has been at its best when we are living those out thoroughly. So one is default to action. I think you see this amongst all successful companies. They're just very action-oriented. They see a problem, they try and fix it. One of the counterintuitive things that I always see a lot of folks fail is they often think that the best way to improve is to think hard about the problem and then take action. But often I have found that you take action in order to see what happens and then you learn from that and then you take action again and that's how that learning loop kicks off. I think the second thing that's been really important for us is transparency. We've run this like big distributed team since the beginning and we have this value of transparency and this is really helpful for knowledge sharing and it's enabled anyone new to the company to quickly search and find out how do we do this, why do we make this decision, things like that. It turns out that value is very fortuitous because agents, AI agents, massively benefit from that level of documentation being easily consumed and out there and so we're starting to get even more tailwinds from investing in that transparency.

Sure. All of our finances are public to everyone internal to the company. So, we share the P&L, the balance sheet, the cash flow. Our CFO goes over them in detail once a month to explain here's what's working, here's what's not. I'm pretty honest about here are the three, here's some things I think we're doing great at, here's some things I think we got to be better at. And I try my best to also call out not just my own personal successes, but also places where I think I've made a mistake and hey, we're going to make a course correction here. This is my fault. We're going to go fix it. And I try and encourage that throughout the entire company as well, too. Where I think one of the hardest things I see new Zapians when they join the company that they have to course correct to is getting used to working in public and getting used to sharing their work before they're ready. There's a personality type that you know is a bit more perfectionist and they'll think oh it's not ready to share with Wade yet, it's not ready to share, it's not ready for someone to see this thing yet.

The people that I've seen be the most successful are the ones that are fastest learners. And usually the fastest learners are getting lots of feedback early and often. And so they're willing to put their work out there when it's rough and it still needs molding. And they're willing to go talk to smart other people in the company and say, 'Here's a first pass. Here's the napkin math. Tell me what I'm missing. Tell me where I'm wrong.' And that learning loop just fires so quickly. And so they're able to scale the mountain much faster than the person who goes off into the corner, shows up two weeks later, and they've never gotten a bit of feedback. And you're like, 'It's not bad, but like where have you been?'

Exactly. Sometimes that works, but oftentimes you find that you've walked off in a direction and now you're not in the same direction as everybody else. That feedback loop and that iteration loop is so critical. That ties back into another value we have, which is growth through feedback. And I think this is about being a learning machine. Part of the longevity of Zapier at 14 years and part of what success we have had has been because we've tried to just keep learning from our mistakes and get better over time. And this takes some humility. It takes some willingness to say, 'Hey, we don't get it right all the time.' With folks that are not successful, you often see them sticking on a thing where you're like, 'You're just wrong. You need to change your mind if you want to be successful.' And so I think that's a big piece, too.

There's the Bezos maxim of one-way doors and two-way doors. I think that's a pretty reasonable maxim. I think the challenge that most companies have is that many decisions become one-way doors even when they're not really. And so that is the hardest thing is you just have to have the courage to just keep pushing and reminding people that things are two-way doors. I was in a meeting yesterday where the proposal was pretty good. There was probably like two or three things that I probably thought if I was doing it I don't know if I'd do it quite that way. Like it's probably not so good. But my feedback to the team was like if the thing you just have right here if this ships next week I'll be doing somersaults. Figure out how to make this thing real. That's because I think the action is way more important than if I'm right or wrong about those two or three items that I wanted to nitpick on.

What I've observed is as companies get bigger, especially of companies that are good learning environments, is when something goes wrong, they sit down and they say, 'Okay, why did this go wrong? And what can we do to fix it the next time?' In any one instance, that is a very healthy practice. But what happens over time is this accumulation of rules that are there to minimize the risk of this mistake happening again. And so if you do that for a long period of time, the organization ceases to take chances, ceases to take risk. Everything starts to look like a one-way door, not a two-way door, because you've had all these situations that have learned behaviors over the team where you said, 'Oh man, this bad thing happens. Let's not have that happen again.' And so it's really hard, I think, organizationally to not do that exercise because it's a helpful exercise to go through that. And I think it takes usually some bold thinking internally for folks to say, you know what, we had this bad outcome, but it wasn't our fault. We're going to keep doing it the way we were doing it. And sometimes that's going to happen, and we're okay with that. We're willing to take the chance that we might have the same mistake again because we think that there's also this upside if we keep doing it the way we're doing it. That is a difficult thing I think for humans to say because all mistakes are bad. No, some mistakes are just cost of doing business like you're going to make mistakes. I often describe startups as one giant experiment and inside the giant experiment there's thousands or tens of thousands of small experiments. Some people are just not well suited to work in that kind of environment. Some people just don't like the idea of experimenting all the time. They want to experiment. I know my job, right?

Yeah, we have this empathy over ego. I think this has been really important for the culture we've built where we're trying to make sure that again it connects that learning culture. We want to empathize with our customers. We want to empathize with each other. We want to understand what that point of view is. If we let our ego get in the way of that, it really can prevent you from learning. Made Zapier be like a fairly like polite but rigorous company when we think about decision-making.

The culture tends to have organ rejection on that kind of thing.

And then the last one has actually just been grown in importance for us is don't be a robot, build the robot. We're an automation company through and through. By having that value, that's partly why we have been quicker than some others to jump on the AI train because we said, 'Hey, there is an opportunity here to build the robot.' Like we can use this tool as a leverage for our own internal operations, but we can also embed it into our products and give those capabilities to our customers as well, too. We just like jumping in the deep end and saying, 'Let's go figure out what this thing can do for us.'

I can give you the progression. We thought it might be important a little before ChatGPT. So my two co-founders, Brian and Mike, had gone off, stepped out of their day-to-day jobs, started working as ICs in the company again, and were like, 'We're going to go experiment with AI because we think this might be important.' My job was to keep managing the existing business. And they had built this text bot where you could just text a phone number and ask it questions. And so like I'd ask, 'Hey, do you have any ideas for my like wife's birthday? I got this weird like pain on my toe. What could that be?' And you're basically just talking to GPT. Very simplistic precursor to ChatGPT. And it was a cool experience. And so of course ChatGPT comes out later that year in November. That was exciting because we were like, 'Oh yeah, this makes sense.' But you know, I don't know that it was the like flashing red light moment for us quite yet. Instead, our approach was we shared it in Slack in general. We said, 'Hey, here's this really cool product. We like cool products. You like cool products. Check it out.' But again, very suggestive, very bottoms up. Then over the course of the next six months, a few things happened. One, we started to see more and more customers doing interesting things with AI inside of Zapier. Things that in the past would have been impossible or very difficult. I remember seeing somebody build an invoice app for his spouse who was doing gardening and that was the type of experiment that would have raised like a $5 million seed round five years ago and I was like holy crap. And so you're starting to feel this like momentum building where you're like something's going on with this, right? Then GPT-4 launches in March of that year. That was the moment. And for us, the reason that was the like pull the lever code red moment was that it was only six months between 3.5 and 4. There was a huge capability improvement between 3.5 and 4. And the costs improvements were also massive. And so we just saw that and said if you project that trend line forward another six months, this is the only technology that matters right now. And so we pulled the code red there.

We had never done a code red before. So it was not this like known definition internally to me. All it meant was like holy crap this is important. We need to do some different stuff.

It was clearly both.

Totally.

Exactly. We sat down and just started to sketch out like what are the things that need to change here? We looked at all the roadmaps and said okay instead of being very suggestive we started to say here are like three very specific places that we can embed AI into the product experience to be better. We started to look at the operations of the business internally and we started to say okay here are a handful of places that need to be rethought in the age of AI. Thing that really worked was to say how do we actually just get the entire employee base to be much more aware of what this technology is and why it is important for our products and why is it important for them to use. And so we paused the company for a week and we did a hackathon with everybody and that took us from around 10% of people using AI after that week it turned into just over 50%. So in one week we got half the company basically addicted to this stuff.

Sure. So basically we said hey all these roadmaps you're working on, all the like marketing campaigns you're doing, all the stuff that you've got going on, pause and we want you to go build with AI. Now Zapier has a long history of doing hackathons. So we know how a Zapier style hackathon works and to do that there was a few things that had to happen. First you had to get your procurement team like dialed in because you can't do a hackathon if you don't have tools. So my co-founder Brian did a bunch of like quick little Loom tutorials. You have to remember there wasn't like a lot of education on this stuff in 2023.

He was just saying here's how I did it. Here's how you hit the API. Here's how this works. Here's why it's like a normal API. Here's why it's not like a normal API. And just coaching people, especially developers, on how to do this stuff. And so we told folks just go build something. If you're an engineering, build a thing. If you're not in engineering, go play with the tools and figure out like how it could help you with a task that you have to do on a day-to-day basis. The other side of the hackathon, we always do a demo day. And the demo day part is always really helpful because one, it creates some accountability. You got to show off what you did. So, you can't just like sandbag the week. But more so than the accountability, it's the learning because now everybody gets to see, oh, you did that. How did you do that? I didn't even think to do that. Why did you do it that way? And so that learning loop kicks off because now you're starting to learn from each other. I've talked to a number of companies who have been going through their own AI transformation. And one of the most common patterns with the successful ones is they do some form of hackathon. It just is a very simple way to get people excited about this stuff.

Our chief people officer took that head of AI transformation role. We held out on like naming that role for quite a while. And the reason why is I never felt like the way to solve this problem was to make it one person's job. This is a technology that everyone needs to learn how to do. Just in the same way everyone's got to learn how to type. Everyone's got to learn how to use email. Everybody's got to learn how to use a document processor. You're going to have to learn how to wield AI in whatever way that looks like in your role. And I felt like if you named a person that all of a sudden people are like, 'Oh, that's that person's job.' So we held off on it. So all this stuff is like raising the floor internally. The next thing we observed was, 'Okay, we've got the floor raised quite a bit. There's a mile long list of ideas at this point in time of things we could go do to improve the product, to improve our internal operations, and they're not always the type of projects that a motivated individual could just show up and snap their fingers and be like, I've solved it. Tada.' They often involved change management and org dynamics. And I wanted to raise the ceiling. I wanted to see this change more differently. And it turned out that Brandon who was our chief people officer, his team, they had been some of the fastest to adopt these tools and they were already running experiments in recruiting, in performance management.

But here inside of Zapier, it was happening. And so I thought, huh, I really would just like to see what Brandon has done in HR happen more broadly across all these other areas. And because the problems we were having were not necessarily ones of technical expertise, makes sense to put the person who has been doing that successfully in charge of those things. And so that's what led me to say, 'Hey, Brandon, I want you to go figure this out for the whole company.'

There is a central team inside of his crew that's like very small that are kind of like AI engineers that can go in and help people out. But by and large, most of it is him going to different teams and figuring out who's the DRI, who's the person that's like really responsible for this stuff, and then bringing that group of people together to trade notes, share learnings, push on systemic gaps that are causing everybody to have problems on a certain problem internally, that kind of thing.

It's basically like corralling the troops around the high priority problems and making sure that those things happen.