How Wade Foster Is Rebuilding Zapier for the Age of AI Agents

The big goal that we are gaming for right now is we want to shift from what I call individual AI to institutional AI. So when you look at most companies, it is not hard to find a number of people that are like 10x the transformative employees, whatever you want to call them. These people have just been massively accelerated with AI. All of us have these folks inside our organizations. That's sort of true. What is much more harder to find is to find the companies that are 10x because they've changed their way of operating. You can find companies that are growing really fast. I mean, you look at Anthropic and OpenAI and, you know, even some of these other AI companies, but even when you go look at their operations, their operations actually don't look that much different than what you or I were doing in 2015, like they're kind of running the same go to market playbooks, the same scale of playups with like some AI dabbled in. So, I don't think we've actually fundamentally changed how companies operate with AI. And I think that there is a real opportunity.

Sure. So yeah about a year ago we launched our AI fluency framework and it emerged from just the natural progression of us getting better at using AI internally. And so, you know, we had been slowly adopting. Well, slowly we felt like it was slowly. You know, I think everyone else it was probably... I think everyone else probably looked at Zapier and said, 'Wow, you all figured that out pretty quick.' But, you know, we've been at it since 2023. And we had gotten to a place where in about 12 to 18 months, we had some data that showed that 97% of Zapians were using AI daily as part of their job.

And so we were bumping up against this problem where we would go interview candidates and we would interview some folks that were smart, talented, etc., but were just really behind on how to use these tools.

Let's see. Well, when we first started this about a year ago, what really behind me was one, there was people who just hadn't used it at all still. Like there was people who just basically were not using it. I mean, look, a year ago, you got to remember like a year ago, there was still people who were, you know, we didn't have Opus 54 and we didn't have this new wave of models that were so good at everything, you know, last July. We were still using like the Oer model in ChatGPT that was a very good series of models, but it wasn't obvious yet. We didn't have Claude code. We didn't have these tools around. And so, most people were like, you know, what I was looking for was just like, if you're an engineer, you were using these things to help you write code. You were probably still in an IDE. If you were not an engineer, you were still probably using ChatGPT for deep research. You were using it to enrich leads. You were doing a bunch of things. But you would find people who were either not using it or using it in just very basic and unambitious ways. It was like, 'Oh, I want to rewrite this email or write this headline or something like that.' But you also found folks that were just not using it at all. Or just philosophically against it. They were just sort of like, 'I just don't think this is it. I just don't think it works.' And so we were just trying to screen out people that were like, 'Hey, look, it's no fault of yours. You're just not going to be successful here. This is like we're doing you a favor. If you are saying you don't want this, don't come work here.' And so we needed a way to just be like hey this is kind of what the bar looks like. And it was also helpful for internal employees because we had folks internally that were using AI but there was always this sense of 'Am I good enough? Am I actually pushing the envelope fast enough on these things?' And so it was both for externally and internally. And we just did what HR teams have done for forever which is create a rubric. We went around and we talked to every function and we talked to the people who are using AI the most in those functions and just said tell us how you're using it. What are you doing with it? What would impress you about a peer using this stuff? Who are the people that you think are even further ahead than you? And then we set up a four-level tier.

I mean, we're talking to our customers all the time. We're paying attention to what the labs are doing like, you know, so we're having a sense of like...

So we pay attention to what they do. A lot of startups are really interesting to talk to. People who are building from the ground up who have nothing. So just talk to the YC company who's going through the batch right now and they have a blank canvas and they have no baggage for how they go operate. And so there's a lot of crazy ways people are working there. Some of them you're like I don't know how that's going to play out over the long term but it's an interesting experiment to see oh that's different than how someone like us who's been at it for 15 years. We have all these things that we're doing a particular way that are battle tested and they're good but there's a question mark now is it the right way in the age of AI? Does it actually need to get rebuilt and be better? And so we have to tear that down before we build it back up right.

Yes. So, the four levels are unacceptable, acceptable, advanced, and then transformative, I think are the exact words. You can go look up the specifics, but basically unacceptable is like, hey, you're just not going to get you can't work at Zapier at this. The next one is acceptable. It's like, hey, you can keep your job, basically.

Okay. So, you're asking, do we have a sense of what percentage of the org should be at what? And are you optimizing for like do you have a view of how the mix should be?

Yeah. It's like, I guess I'm supposed to take care of myself, whatever. But really, it's like write code and talk to users. Funny enough, if you go read Peter Drucker, management consultant, his whole thing is like the purpose of a company is to create a customer and the way you do that is marketing innovation. And so it's sort of the same words to describe a different thing, the same thing, which is like, hey, you got to build stuff for customers and then you got to sell it, you got to market it, etc.

I'm a people person.

Yes. Those are the things that generate customers effectively. 100%. So everything else needs to be minimized.

I think Jack Dorsey is very much on the right track here. I'm not sure I buy everything he's pushing. Like he's got this opinion that everyone will report into him or this AI entity and so he has 6,000 direct reports or something like that. I don't know about that. I'm probably more prone to believe it than most but I think that's pretty farfetched. I still do think that most people want to have a human that they can sort of go to and no one human can have 6,000 direct reports. So I do think that there are some implementation details that are still going to get ironed out and experimented on.

I don't know. I think average was 900 some people. That's probably not the right size. You know, I paid is 20 people. That's probably the right size and we're still not doing everything perfectly.

All businesses are going to kind of settle into a margin profile at the end of the day. And if you're taking too big a margins, you're going to have competitors come in and out compete you. And if your margins are razor thin, you're going to struggle to operate. And so you're going to kind of settle into a spot that works for the business. At the end of the day, that probably dictates how big you are.

Well, I don't know that you get to work back from margin. I think it's kind of just dictated by the things you build. I never have sat down and said, 'Okay, I'm going to figure out how Zapier's margin profile should work.' I think it just sort of is an emergent property of the value we deliver to our customers and the way the business operates.

Yeah, it's definitely something we think about and talk about. It definitely feels like job families are squished together, have more overlap, because AI raises the bar or raises the floor for what anyone is capable of. So if that's the case, then that would suggest, okay, why isn't there just one job function in a family like doer or seller or something, right? The counterargument to that is that there still is a disproportionate advantage for people who are really good at one thing. Like you think of a designer who is truly exceptional, they are still able to do things that an engineer with AI trying to learn how to do design can't do. And same thing with engineering. A really great engineer equipped with these AI tools is still way better at me who's just vibe coding my way through this stuff. And so, the experts feel like they're getting accelerated in the thing that matters. And so, it does still feel like you're going to have a lot more generalist capabilities in your organization, but there's still going to be places where you want somebody who's like, 'This is our hardest problem, and we want the person who is best at this area in this thing working on it.'

That's right.

You're right. Yeah, it does change at the end of the day. And so, I don't know like at the end of the day, does this mean the teams that we're all assembling are just like the way to think about it isn't hard skills anymore? It's like, oh, we need a customer obsessive, we need a craft obsessive, we need a, you know, XYZ on the team and within the group of people, they just all using AI and using AI systems that have been built up and around that. Is that the way these teams get applied?

Well, I think the thing that really stands out and look, this always stood out. It just is much more important now is how curious are you and how fast do you learn. Those two things are basically the entire name of the game now. It just doesn't matter how much you actually know about a thing because the reality is we all have these LLMs that just know far more than you. So how intelligent you are kind of is not the thing that dictates whether you're successful or not. It's, you know, some people call it agency, some people call it self-starter. There's a lot of words people are throwing out there for this, but at the end of the day, I think of it just like, are you curious? Are you learning fast? Are you doing stuff? And if you're more oriented to go that direction, you're probably going to be massively accelerated in this world. Whereas if you're at a role where you've sort of been rewarded for your book smarts, I don't know that that's like you're not going to beat these LLMs, you're just not.

Yeah. I think a lot of us who built pre-AI are finding another gear. The good news is the demand for automation, the demand for this stuff is through the roof. Our sort of first decade, people paid lip service to automation but then they would just blit scale and throw people at problems and that's how they would go scale. Nowadays, there's tens of trillions in TAM that have been unlocked. Of course that's going to create a bunch of products trying to target automation generalized specialized etc. So a lot of what it means to lead internally is just making people aware of here is the truth of the playing field. The good news is this is happening. The challenge is that there's so many more ways in which you can fulfill these things and then here is how we are going to differentiate and here is how we view our opportunity to stand out in the market. And then some of that's just changing our product lines. And so we have a I call it a game one game two operating mechanism internally where game one is like hey this is the AI first playing field and game two is sort of navigating our existing customer base and how do we support them and how do we do a good job here while making sure that we're fully accelerating on the first thing. And so it just creates mental models for people to make decisions and figure out where to invest where not to invest. Otherwise it just constantly becomes a push and pull between the business you've got and the business you want to have.

Yeah.

Ours was in 2023 when the GPT-4 launch came. We called it code red and said, 'Hey, this is totally going to change our business.' We had seen the ChatGPT launch and were excited about it, but we hadn't yet gone, 'Oh, this is that code red moment.' I think it was a cool product, a lot of people were using it and excited about it. But what was different for us about GPT-4 was that it came just six months after 3.5, so the speed at which it got released was interesting. The quality of the model stood out—GPT-4 was a lot better than 3.5—and then the price changes, it just felt like the price curve was decreasing, so this stuff was getting cheaper and cheaper. You looked at that and said, 'Wow, if this is the start of a trend line that's going to continue for every generation of model releases, then we are not doing nearly enough to incorporate these capabilities into our product. We're not doing nearly enough to incorporate these into how we operate. This is a code red.' So we paused the company for an entire week and said, 'Okay, we have to go figure out what this actually means to compete in this new world.' And we've kind of been operating that way since.

It's kind of cliché, but there's the peacetime CEO, wartime CEO sort of labels that probably apply. I'm just much more in the details on certain things, much more opinionated, and much more decisive. We're placing bets that probably have a wider range of outcomes than before, where it's like, 'Hey, this might not work, but if it does, that trillion-dollar TAM or that ten-trillion-dollar TAM is there for the taking.' So there's a lot of that going on. I think you just kind of have to, if you want to work in tech these days, you have to be ready to sign up for that. I don't ask everybody in the company to work crazy hours or anything like that—I don't think you're going to win over the long haul that way—but I do think there is a level of intensity that has to get ratcheted up and a level of seriousness that has to get ratcheted up, which tech is definitely not known for. I look at my sister who's a nurse and my other sister who's a teacher, and they don't get paid nearly as much and they are way harder working than most of the tech industry.

That seems mostly performative rage bait to me. I'm sure it works for them, but I don't think that is... Look, when we started Zapier, I don't know what we worked, but it was a lot. I don't remember any waking hour that we didn't work on the company.

I think you want that level of skin in the game for your early employees and probably your executive team and things like that, but you can't ask that of a thousand people. There's just no way you can do that. But you can have a culture of winning, a culture of seriousness, a culture of work hard, go home. You can have really high standards and really excellent folks and not burn people to the crisp. We've been in business for 15 years. If you want to run for 15 years, you have to have a way to throttle that up and down.

Gosh, that is the question, isn't it?

Yeah. Look, I think the game one, game two thing works really well internally. It makes it easier to put things in a clear bucket and have goals associated with each. The thing we have optimized for with respect to game one—our new area—is we know our job is to maximally accelerate that. But what does it mean to actually maximally accelerate this?

Here's our point of view: agents are going to be the predominant builders of automation going forward. So we need to make sure that our trigger infrastructure, action infrastructure, authentication layer, governance and policy layer, and runtime environment are all well embedded into Zapier—your MCP, Zapier SDK, Zapier CLI—so that when agents go reach for this stuff, they can build on top of Zapier infrastructure. They can build deterministic workflows, agentic workflows, workflows optimized for tokens. They're not there to token max; they're there to run your workflow in a very specific way that helps you get the most out of it. We're trying to make sure we are growing that as fast as we possibly can. We're operating a lot like we did in the early days of YC, where PG says, 'Hey, you want to try and grow 10% week over week on this stuff.' And that's what we're doing with those products. What does it mean to go as fast as possible there? You generally have the smallest number of people with the most amount of ownership as possible. If you put too many cooks in the kitchen, small things usually get crushed; they don't actually go fast. So it's a delicate balance of trying to figure out the group of people that will actually make this thing go as fast as possible, while knowing that most of the org will be keen to help them, and that help can be well-intentioned but can actually crush the thing. You have a very high-judgment leader sitting over the top, watching astutely, making judgment calls to say, 'Okay, we actually do need help. We have shovel-ready work. Let's bring more people in.' Or saying, 'No, no, no, we're not ready. You'll actually slow us down.' That's the delicate balance we're trying to manage.

It's like a blending of both. I do feel a lot more like we're operating with many of those principles we had in the early years. What is different, though, is that often I'm espousing these values and pushing the team to go operate in these ways and teaching them what success looks like. But I'm not doing every single one of those tasks anymore. There are groups of people doing a lot of the things I was doing at the time, and now I'm overseeing it and occasionally helping in some ways. What's different from the scale-up periods is I am a lot more opinionated and a lot more decisive about what we should be doing and what we shouldn't be doing. In the scale-up, one of the challenges I faced—and I wonder if you have this, I know a lot of CEOs struggle with this—is you hire your executive team, and the impression you get is, 'Oh, I need this person to figure out sales, marketing, engineering.' And you end up delegating too much to them, and everything becomes a committee. I've learned better how to partner with the executive team and create a better team environment. Where is my unique value? Where do I push and where do I push back? Versus, 'No, this is your area of expertise. You cook, you go make this happen. This is yours to own and deliver upon, and I'm here to support and partner with you.' Getting that balance right, I really struggled with in the scale-up. I was probably either way too much in the weeds or way too much on the other side.

I think so. I wouldn't say I'm uniquely great at this, but after 15 years of doing it, you can't help but get a little bit better, I would hope.

I think this is where Zapier has always been a little different. Our aspirations were never to go public and have that define our success. We raised a million bucks and then bootstrapped the rest of the way.

Sure, but our focus has always been more customer-oriented: how do we actually build something that people truly love? How do we make it so that more people love this, or love it even more deeply than before? That's always been the focus. How do we continue to grow that impact? And then the revenue follows. The revenue is great because it allows you to do more innovation, more unique go-to-market programs, and try all these things. Of course, we're competitive and want to grow that, but it's important to figure out what actually drives here. You don't want the tail to wag the dog.

Most are still in. I'm sure there's been a little bit sold in secondaries along the way, but most of that $1.8 million is still there.

At some point, they'll get some return. We'll figure it out. That's probably the harder question to answer. It was much more clear a few years back. Now it's a much more dynamic situation.

I don't know that... Of the folks I do talk to, the thing I'm most contrary on is the fundraising path. I sort of feel like more people can go much further with much less money than they think. And I think it's even easier than it was before because these AI tools make it so much easier to build. So that's probably the one place where I'm most contrarian to the advice folks are getting. I usually say, 'Hey, I feel an obligation to tell you this because most people are not going to. I'm probably the only person who will give you this point of view.'

I feel like I have to give you this because if you don't, you won't hear it anywhere else. The other thing that is not novel but certainly true is that distribution is the name of the game. Building stuff is trivially easy. How do you get attention? How do you get customers? How do you build something that's differentiated and matters? That stuff is harder than it ever has been, and it requires a lot of creativity. You really have to understand your customers well to make that part happen.

A couple things come to mind. First, the existence of OpenClaw is pretty compelling. This was built by one guy in Vienna, and he beat all the labs to a personal...

But it's still a guy. A smart kid in college could do that too, just spending all their time living off ramen in a cheap old apartment somewhere. You could make that happen. I do think there are ways that single people can use these tools in such extreme ways to get disproportionate outcomes. Even the compound startup idea you're talking about can still be done with a fairly small team. Look at Every team—they have four or five different products, and one person owns each one end to end inside the organization, which is fascinating. I just think you can go so much farther with so much less than before. Even if you're trying to raise your ambitions in terms of the scope of the platform you're building, it is still that much easier to build that it doesn't even out. In my head, the real hard part is the distribution part. I would find that argument more compelling if it were, 'Hey, we know we can spend this amount of money to really unlock a wedge on the distribution side of the house.'

It depends on where you're at, but that's where I would more consider it. On the R&D side, it also depends a little bit on what you're doing. If you're going to build models and have big capex expenses, okay, that's something you have to do. But if you're playing at the application layer, it just doesn't seem as important. Look, I'm not saying don't raise. We raised a million in change. But I think about what you can do with that initial injection and how far you can get. You can get so much further than you could before.

No, no, no. This is just a part of our go-to-market strategy. We're sharing more and more of the way we operate. One of the things I've noticed is that a way to get marketing attention is to share what I call 'bookmarkable assets.' Things like the AI fluency framework, the go-to-market skills, and our automation bench, which is a model benchmarking tool. These are all things people are really interested in. In the case of the GTM stack, those things are all using Zapier MCP, Zapier SDK, etc. The skills and all that stuff are open source and tap into the capabilities Zapier has. It's a way to educate the market on concrete, specific things you can actually do. I think it's really helpful right now when most of the AI narrative is around the hype cycle, funding, and personalities, rather than the more pragmatic 'How do I actually do this? How do I go from individual AI to institutional AI? How do I run an engineering team, marketing team, or sales team with these tools?'

Zapier has always been beloved by operations leaders in various functions. We've had tons of marketing ops, rev ops, sales ops, HR ops, finance ops. These are the people who have been core to Zapier's usage in the past. We've always kind of catered to the operations folks with a slight functional bent.

I think there are some advantages. The fact that people know of Zapier is obviously a good thing in most situations. The fact that we've been doing automation for 15 years, that we've connected to everything, and that we're really reliable at doing that—there's a certain trust that comes with that, which is really helpful. I think the challenge is that we're not the cool kids on the block, and we're not the young kids on the block.

A lot of it is coming on podcasts, telling the story, making people aware that we are as AI-pilled as any company. We've been doing this stuff for a long time, and we have invested a lot in our products. The product you use on Zapier today looks a lot different than it did three years ago. We're just making people aware that we have all the same connective tissue we've always had, but why that's valuable in the age of AI is that you can use it as the context layer for all your agents. These agents get so much better when they hook into these different tools and applications, and Zapier can power all of that. A big part of it is refreshing people's mental model of what Zapier is, why you use it, and why it's as relevant as ever in this age.

Maybe we should have Zapier 2.0, or do the Intercom thing—Intercom is now Finn. Maybe we should do something like that.

100%. These are the decisions I spend a lot of time sweating because there's no one else in the company who is going to say, 'Let's rebrand, let's change the name,' and can do that with some authority.

What's easier: to build a brand totally from scratch, or to pivot a brand that is known and beloved but needs to be known for a slightly different thing? It's not obvious to me which one is easier.

Yeah. Your point is right. Let's see how the Finn rebrand works—that literally just happened a month ago. It's interesting to look at people who are really good at branding. I think of that brand champion that does t-shirts and shoes. Growing up, that was the Walmart brand—the cheap brand you would get everywhere. Nowadays, it's what celebrities wear—it's a classy brand. They totally flipped that thing. So I do think we're in this territory of, 'Huh, how do you actually do some of this stuff?'

I know, we might be happier. Who knows?

I love it.