Zapier Co-Founder & CEO on Building the Agentic Enterprise | Wade Foster

Yeah, thanks. So Zapier, for those of you who are not familiar, we're the most connected AI orchestration platform. We really help you connect all your tools into your AI systems, which is really important because these AI models, if they don't have access to your data, they don't have access to your tools, you kind of get a vanilla experience. But if it has access to the content that you use, so your CRM data, your ticketing data, things like that, it's a totally different experience. The value of these models gets immense when you use it in that way. And for us, the transformation was really important when we saw ChatGPT come out, really actually GPT-4, which came out six months later. It was a real sort of wake-up moment, in part because GPT-4 was so much better than 3.5, the models were cheaper, and it came out just six months later. We felt like, wow, there's a slope here that if this continues is going to be really important for what we are working on. We're helping people automate their work, and it feels like AI is going to be an enormous ingredient in that. And so we called a code red internally and said we got to go figure this out, both in terms of the products we sell but also how we operate internally.

And you know, I think the thing that we've learned in the three years since is that there's a couple ingredients that really do help organizations change. The most important ingredient, the one that I see most common both that worked at Zapier and has worked at many other companies, is just workshops, hackathons, hands-on keyboard. So much of what we talk about with AI stays way up here, stays at the 30,000-foot level. But when you see the leaders all the way down to the front lines putting their hands on the keyboard saying, 'I'm going to build an agent' or 'I'm going to build a workflow' or 'I'm going to build some sort of automation, however basic,' all of a sudden the discourse starts to get a lot more practical and tactical, and you start to see some of the fear fade into the background. And you know, humans' natural curiosity starts to lean in. I don't want to claim a hackathon is a panacea for solving all your problems, but once you get more of the organization just a little lean forward, a little bit more engaged, all that other friction and resistance you start to have, it just gets a little easier.

I mean, I think the alternative sounds way more expensive to me. To not do this feels like it would potentially mean, you know, that you're going to get out-innovated, out-competed by others. And so, you know, the marginal cost of using these tools, it's not that much. Yeah, it hardly, it almost feels like a non-decision to me.

Well, I think the real opportunity that we're thinking about is how do you give access to tools to these AI agents. You know, this week we launched an SDK, which is a powerful way where you can authenticate any of your tools. So CRM data, help desk data, email, calendar, things like that, to Zapier in a governed way. And now an AI agent can go interact with it in a powerful way. So for example, I have a tool that after every meeting goes and fetches the transcript out of my meeting note tool and then it takes all those action items. It drafts emails that needed to go out. So if I said, hey, I'm going to follow up on this, this, and that. I'm going to introduce you to this person. Those things are automated, automatically drafted. The context to update the team that happens in Slack. Great, it's created an update for me that goes out into the right channel, into the right thread. All those things are work that I would have historically had to have done myself. I'd spend hours at the end of the day just triaging little bits in here. Now I've got an agent that just goes and handles all that stuff for me. And that leaves me more time to work on product development, work on, you know, new marketing campaigns, things like that. And so that's the power of what happens when you give these AI systems access to your tools in a governed way.

I think the key place where Zapier operates is that action layer. You know, if the AI is the brain, Zapier is the arms and legs. It's the thing that's out there interacting with your tools, doing it in a governed way. I think that governed part is really important. You don't want every agent to have just like unlimited access to these tools. So you want something that can say this group of people has access to Salesforce read access, this group can have write access. So you need that governed piece that sort of allows it to happen in a way that represents the organizational trust system. But we're there to sort of help facilitate all that reliably.

Yeah, I think the most important skill set that is going to endure is human curiosity, like entrepreneurial spirit. The one thing that these AI systems don't do is they don't think for themselves. They're still inert systems. And so you do need to prompt it and kick it off. And that requires, you know, a human to have judgment and to push on those things. The thing that's going away is all that like munging work in the middle. So many organizations as they get bigger and bigger, there's a big internal coordination task. You know, takes this bit of context, re-ranslates it for this team, re-ranslates it another way for this team, re-ranslates it for another team. Turns out AI is very good at that work. It can take written context and put them in different formats for different groups of people. So that, you know, the human labor that sort of sits in the middle of those systems, I think those jobs, which candidly a lot of people don't love that work. They don't love writing all the status updates. They don't love re-ranslating all that stuff, increasingly that can get delegated to an agent and now the humans can get pushed to the more creative endeavors, product development or sales and marketing, the things that like really actually do matter for the future.

I mean, definitely the latter. You know, the organizations that are moving the fastest right now have builders in all disciplines and domains. And that's what's really powerful is that you now can have your accounting function has someone who can build inside of it. Your HR function has someone who can build inside of it. Your marketing function has something that can build inside of it. And that is very powerful because most of us have been in organizations where you get rate-limited by engineering or IT, so you sort of end up, you know, adopting these third-rate solutions to these problems. But when someone in that function who understands the discipline, who understands the domain, who understands the workflows can now say, 'I want to go make this better myself,' that is a massive unlock for those disciplines to go solve their own problems. So I do think we're going to see the rise of just more people building in all these functions and it may not be every person, it may not be every job, but you know, even if there's just a 5 or 10% increase in those functions, that is a huge benefit to how various functions in an organization work, especially the non-technical disciplines.

Well, I think this is a thing that everyone is struggling with. You know, most organizations are caught in a really difficult choice. You can be scared of giving access to these tools. And if that's the choice you make to say, 'Hey, we're going to restrict access,' oftentimes, you're now trying to compete with your arms tied behind your back because somebody else is going to be willing to do this and they're going to take more risk and they're going to be able to innovate faster than you. The alternative is you give access to all these tools, you turn these systems loose, and we're going to find ourselves in, you know, a big security vulnerability. We're going to find ourselves failing a compliance check. And that also feels very risky. So, there does need to be a third way. And I think one of the most important things is permissions to these systems needs to get far more granular. You don't want to give all-or-nothing access to the CRM to every agent. You don't want to give all-or-nothing access to take actions in these tools. But you want to get a lot more granular. Let me give you an example. I mentioned that workflow at the end of the day where I have my agent draft follow-ups for me. The key thing is I'm having it draft them. I'm not having it send those in an automated way.

Yes. Humans still in the loop for that workflow. They land as drafts in my Gmail inbox. I pop it open real quick. I take a look at it. I press send. Now over time as I build confidence in that system, maybe I will let it go eventually send those in an automated way. But I'm not going to do that on day one. I'm going to build the right system that allows me to monitor that and feel trust in that. And that is where there's a lot of investment both at Zapier but across vendors across the space is trying to figure out how do we make sure to provide that type of governed access to those tools because if you do it well that is the real unlock. That is the third way that I think organizations need to be able to really feel like they can invest fully in these AI systems. You know, if you don't have like a proper governed setup, it's pretty terrifying. You know, I think I've heard a lot of OpenClaw mentions this week. OpenClaw has shown the way but I think one of the big failures is that you have to give all this access to a tool and it doesn't provide some of those governance mechanisms out of the gate. Which makes it great for personal use but to use inside of an enterprise, it fails that test.

Yeah. I think one of the biggest challenges that humans face with automation is we often just don't know where to start. I think this was true in 2011 when we started the company. It's true now in the age of AI. When I go talk to our customers, when I go talk to folks that are getting started with AI, it's not that people don't want to use this stuff. People are excited about this. But there's a cold start problem. I heard the prior speaker talking about this where he said, 'Hey, every UI is a chat box.' That doesn't help you know what use cases should I do, where should I start? And I think more people in the space, whether they're vendors like us, they're educators, the biggest thing we can do is to put practical examples in front of people and say, 'This is something valuable that you can start with. This is how you can kick that learning curve off.' And once you get that learning loop going, that's when I think humans' creativity starts to kick in and you start to get a feel for this where you start to go, oh, if I can draft follow-ups to my emails, what else could I do? Maybe I could also take that same transcript and maybe I could have it generate a case study. Maybe I could repurpose it for social in this way. And once you get that creative loop happening, it makes it a lot easier. But I think the biggest thing that is often holding companies back, individuals back, is not that they aren't excited about this stuff. It's not that they don't have access to tools. It's just simply what should I do when I sit down in front of my laptop and I want to get started. What is the first thing that I'm going to go build? And I think that's where there needs to be a lot more investment is to just get people understanding the art of the possible.

Yeah, we have an AI fluency framework and we use this to assess candidates who are looking to get a job inside of Zapier. And we shared this with the world in part because we felt like a lot of companies were imposing these AI mandates. We must be AI first. We must do this. But when we went and talked to people and said, 'Hey, you know, what do you think it means to be AI first?' You get a lot of stares back. Generally you'd get a lot of people that again were excited to do it. Say, 'Hey, we actually do want to be AI first. I'm not opposed to this effort. I just don't know what it means. Like what are you talking about here?' And so we thought, well, hey, that's reasonable feedback. Like this should be more concrete than just, hey, we should be AI first. And so the exercise we went about is we found the power users in every single function. And we just sat down and talked to them and said, hey, what do you think makes somebody, what's a non-starter in your discipline? And if someone's not doing these things, you're telling they're just not AI first at all. What's sort of the baseline? What's like what you would expect anyone in this discipline to be doing? What's advanced look like? And then what does cutting edge look like? And so we have a simple framework. We broke it down into every discipline inside the company. So engineering, product, sales, marketing, finance, you name it. And we released this AI fluency framework. And now we use it as part of our hiring test. So when we go assess people, we assess them on all these dimensions as part of performance reviews that are part of this as well too. But I've heard from, I shoot, I probably heard from hundreds of companies now that have adopted this as well. I'm sure it's maybe even into the thousands. And you know, it's just a helpful framework to take it from this high concept of like we must be AI first to something just much more practical and tactical to say okay, this is a good starting block for us. Now, one challenge with it is it quickly goes out of date. This stuff moves fast. So, this week we launched our V2 or I think two weeks ago, we launched our V2 of it. You know, I suspect, you know, six months from now, a year from now, we'll have a V3, a V4. Because the space just moves so fast.

Well, so at Zapier, we run hackathons probably two, three times a year across the whole company where we just invite people to build and you know, it's protected time where they can set aside the daily work that they're doing and come back and say let's relook at old problems. Let's look at new technology and let's see if we can solve problems that we couldn't six months ago. And that keeps everybody's skill sets fresh. It keeps everybody up to date on what the state-of-the-art looks like. Now, I see that pattern reflected in a lot of enterprises as well where they'll do these hackathons every so often. And these hackathons are not just for the engineering, they're for all functions. Maybe once every six months is probably good. Yeah, I think for companies that are just getting started, they often will do a workshop with a smaller group of people, they'll maybe choose a slice of leadership, you know, the people who are going to be responsible for rolling this out to the whole company and they want to get them up to speed because these are the people that are going to have to bring it to the masses. And so they'll usually start in a pocket and then try and get it to go from there. But the best companies that I see are generally thinking how do we get this from a small group of people to everyone in the company. I think there is a mindset in some companies that it's going to just be a centralized group of people that go figure this out. And I understand that impulse but for whatever reason the companies that are at the cutting edge are pushing it out and having more success pushing out to most and those are the ones that seem to be going the fastest.

I think the daily brief is like just a real basic starting spot. You wake up in the morning, you should just get a well-drafted brief as if you had the world's best assistant.

Well, gosh, I mean there is so much hype right now. You know, it's hard to point to any one thing but there's like so much hype out there. So you know, in the short term it definitely feels like things are probably across the board overhyped. What is underhyped is I do think that we're going to look out five years from now, 10 years from now and all the stuff that feels overhyped today we're probably going to be far surpassing that.