Has AI been an accelerant or a threat or both for your company?
I think it's both for most companies. There's this incredible technology that has emerged and everyone is figuring out what are the use cases that are super valuable in the future. And so the companies that can leverage their advantages today to build those capabilities are going to benefit massively in the future. And we think we have some of the really important assets. Like the fact that we are connected into over 8,000 different tools is a huge asset for us. But any company that has existed for any mere amount of time that says, 'Oh, don't worry about AI. It's not going to do anything to us,' I think is making a foolish choice. And so we've been very aggressive at trying to reposition our company in the age of AI to take advantage of the opportunity and not let others displace us.
This is AI First with Adam and Andy, the show that takes you straight to the front lines of AI innovation in business. I'm Andy Sack and alongside my co-host Adam Broman. Each episode we bring you candid conversations with business leaders transforming their businesses with AI. No fluff, just real talk, actionable use cases and insights for you.
Greetings everyone. I'm so excited for today's episode. We have Wade Foster from Zapier. Zapier makes you happier, the automation company. And today we're going to get into agents and the future of work. And really two things we want you to get from this conversation. One is we want you to have a clear understanding of the evolution from automation to agents which have just occurred and make sure that you understand that. And then also a simple mental model for the future of work which we think really involves managing agents. So those are the two things we want you to get from this congress. With that allow me to introduce you to Wade Foster. Wade, welcome.
Yeah. Howdy everyone. Thanks for having me.
And of course my co-host Adam. Good to see you again, buddy.
Yeah, you too. Really excited to talk to Wade.
Wade, before we get started, if you wouldn't mind just give a brief introduction to you and like what did you do before Zapier and when Zapier started and what is Zapier?
Yeah, sure. So, I'm Wade. I'm the co-founder and CEO here at Zapier. We founded the company in 2011. Before that, I worked in marketing in a mortgage company for about 10 months. And before that, I was in school. So, just doing what school kids do.
Out of the mortgage industry.
Yeah, there you go. Something like that. But, you know, I think the reason we chose to work on Zapier is my co-founder Brian noticed that when you would go to the help forums of all of the emerging SaaS products at the time. So, think like Salesforce, Zendesk, Dropbox, you would find this common pattern in the help forums. Customers would be asking when are you going to integrate with XYZ product? And then in the help thread, there would be all these comments from customers saying plus one, me too, I need that, I need that, I need that. And then eventually a product manager would chime in and say, 'Thanks for the feedback, everybody. We'll take a look.' And if you ever worked at these companies, you know, that's more or less code for not going to happen. It's going to get dumped in a backlog somewhere. And so our kind of thinking was, huh, maybe we can make it really easy for a non-technical person to build these integrations themselves. They don't have to know what an API is. They don't have to know what code was. And this was really appealing to me because I was working in email marketing at the mortgage company and we were trying to do a Marketo implementation and Marketo had this old school SOAP API integration with our CRM. And so Zapier started with that. It's like how do we just connect as many tools as possible for folks? And of course over the years, you know, we've added workflows and workflows became really popular. And now, of course, we're in the age of agents and AI workflows and all this sort of stuff. And those are increasingly becoming the most popular thing on Zapier.
Wait, like let's make sure we get our definitions correct. So, can you for our audience, can you define automation and agent distinctly?
Sure. Well, you know, I think I would probably make the difference between a workflow and an agent. Like to me, automation is kind of like the superset of those two things. But a workflow is deterministic. It's a human, it's code that is deciding what happens next. You're saying, I want to start here. I want to do this next. I want to do this next. I want to do this next. And then at the end, I want to have this output. Now, you can have AI in the mix there, but you still have a very concrete set of steps that are going 1, 2, 3, 4, 5. Whereas with an agent, you're saying, 'I'm going to give this agent, the LLM, a goal. I'm going to give it tools that it can go access, and now I'm going to let the LLM decide what does it do? 1, 2, 3, 4, 5.' And then it's supposed to deliver the output. And that's the key difference between the two. And neither one is better or worse than the other. You get different outputs and different tradeoffs depending on what you get. But at the end of the day, you kind of want to pick the best tool for the job depending on what you're trying to solve. And both of those things are a subset of automation.
With AI. I mean, you certainly started the company well in advance of ChatGPT 3.5. Workflow, you know, your tool has exploded in use in the last 3 years alongside. What's the last year been like at Zapier? Bring us current with what you're wrestling with and how things have changed for you at your company.
There's a couple key innovations that we've been working on. One is a Zapier Agents product. This wasn't possible, you know, 3, 4 years ago, but now you could come in and you can create an agent where you give it a goal, you give it access to tools, and the LLM goes and operates on it. It's a really simple experience to set these things up. It's actually much easier than kind of like traditional workflow configuration to build these things because all you have to do is think through well what is it that I want to achieve. You write it down like steps 1, 2, 3, 4, 5 and then you authenticate your tools. So that has been one really popular addition to the platform. The second thing that we've spent a lot of time working on is our co-pilot experience which exists to build both agents and workflows. This I think has probably been the most important unlock. LLMs are fantastic.
And Wade, let me just interrupt for a moment. When you say co-pilot, do you mean Microsoft Copilot or this is Zapier's co-pilot?
Yeah. This is a tool that you can, is a chatbot that you talk to that builds automations for you, either an agent or a workflow. And this is so powerful because now you don't actually have to know what exactly am I trying to configure. You can just describe what you want to have happen. And then on our side, we figure out, do you want an agent? Do you want a workflow? Is it better to be more deterministic, less deterministic? And that I think is one of the most beneficial tools that we've built. I'd say the third thing that we built that's really important in the last couple years is our Zapier MCP server. And the reason this is powerful is you can bring the underlying capabilities of Zapier into your agent tool of choice. So this might be Claude or ChatGPT. It could be Cursor or Claude Code. And now you have all of the action inventory that Zapier has. So you can say while you're chatting back and forth with your AI friend, you can say, 'Hey, can you go fetch my emails from today? Can you go look up what calendar events are coming? Can you go do research on these things? Can you go generate this for me?' And you get that sort of very fluid experience of your AI having arms and legs to go do stuff for you. That is really, really powerful. And so I think those three innovations have been really important for us in the last couple years.
Has AI been an accelerant or a threat or both for your company?
I think it's both for most companies. There's this incredible technology that has emerged and everyone is figuring out what are the use cases that are super valuable in the future. And so the companies that can leverage their advantages today to build those capabilities are going to benefit massively in the future. We think we have some of the really important assets. Like the fact that we are connected into over 8,000 different tools is a huge asset for us. But any company that has existed for any mere amount of time that says, 'Oh, don't worry about AI. It's not going to do anything to us,' I think is making a foolish choice. And so we've been very aggressive at trying to reposition our company in the age of AI to take advantage of the opportunity and not let others displace us.
Adam, let's get you in here. What, you've been listening to Wade for the opening here and unusually quiet this Tuesday morning.
Yeah, I'm actually fascinated by the way you're describing it, Wade. Like it actually I've learned a lot just listening to you in the first, you know, 10 minutes here. So the question I have for you is, first of all, do you have an integration already with, I'll make up like Claude or ChatGPT, I'll pick on Claude because they're sort of in a weird way the most sort of advanced as it relates to this conversations, Claude Code or Claude co-work, like is there already on Claude co-work if I went to Claude co-work and I said hey I've got this idea for you know this thing I want to do to automate a workflow in some way? Does it have an integration with Zapier already where like it would say fine I'll go grab Zapier and Zapier's already got the integration down and I'll sort of stitch this stuff together and Zapier becomes an important part of the glue. Is that sort of in place right now?
So there's two ways that you can integrate Claude with Zapier. First is through Claude connectors and so there you can turn on the Zapier MCP setting and now you have access to the entire library of Zapier connectors right within Claude, Claude co-work, Claude Code, you name it. So that's way one. The second way you can integrate Claude with Zapier is in zapier.com itself. So if you're building a workflow and you want to have an AI step you can specifically select any of Anthropic's models so Opus 4.6, Sonnet 4.6 or some of the older models as well and you can use that to power AI steps within your workflows. So those are the two key ways you can work with Claude or Anthropic's ecosystem inside of Zapier.
Right? And I assume that's the same in theory with ChatGPT or Codex or Gemini, you know, or whatever, right? That's the same idea.
Yeah. But to your point, Claude is definitely I think a little more forward facing on this particular type of work today than the other two.
Yeah, it seems like that right now. Like I mean it sort of came out of nowhere it feels like but if you're paying attention to Claude Code it didn't come out of nowhere but for those of us like me that are not coders like it was like oh my god all of a sudden I just want to live in co-work. I never saw that coming.
Mhm. Let's talk a little bit about go back to something you said a second ago, which is like workflows versus agent. Do you feel like we're pretty quickly going to be in a world of agents where we as individuals, whether personally or at work, that we're just sort of working with our agents and our agents are helping us figure a bunch of stuff out and including figuring out how to automate a workflow or how to define a workflow, how to automate a workflow. Like does and is it become like this giant black hole like it's just like we all just end up using agents for everything and then agents figure out what are their best tools and the best everything like can you kind of comment on you know because you've been living in this world and you've been building this stuff and you're kind of in the center of it like help us think about how you Wade think about agents era.
So I do think that we are going to be in a world where our agents perform increasingly more and more tasks for us and we are interacting with them. The way developers have experienced this is their first experience was with like tab completion inside GitHub's Copilot tool. But increasingly as these tools have gotten more and more sophisticated, the developer's role has become more abstract to the point where with the latest batch of models that have been released, you're hearing more and more very senior developers say, 'I rarely look at the code anymore.' Which is a remarkable statement. They are prompting the code. They are doing systems design with the code. They're having the code do review, its agent review the code itself. And so you're seeing this developer take on more abstract tasks and delegating more of the code creation and code review to the agent itself. I think this will happen across all of white collar knowledge work. And I'm seeing it even for myself with my day-to-day work. You know I'll have an agent that will prep me for a meeting. Instead of me going and doing the research directly, I'll have an agent that goes out and drafts status updates or intro emails after I meet with a customer for the AE team or updates the CRM rather than me going and updating the CRM manually. So you know I think there is a huge amount of work that is going to be delegated to these agents in the future and our job as humans is going to move up that abstraction layer.
Wade, I'm curious if you'd comment on what you've seen from your customers in terms of their willingness and ability to trust agents.
Well, I think this is one of the most crucial challenges today is that agents have really two tricky challenges compared to workflows. They are generally less reliable. They're probabilistic. They have agency to choose how they go and act and behave on these things. And so if you really care about a very specific output, your agent is probably going to underdeliver compared to a workflow. The second thing an agent struggles with is it's more costly. They're burning lots of tokens often to go figure out what that task is, especially if your task is more complex, whereas a pure workflow is going to be deterministic code, raw API requests that are just not very costly at the end of the day. And so my general advice for folks is that they probably should be running more things as workflows if they can. That does not mean that your workflow may not have AI as a component of it. But if you care about cost, if you care about reliability, you're probably going to want to run the thing closer to a workflow than to an agent. Now, that does not mean that when you go build the workflow or build the agent, you're not having an...
Well, because you get the reliability and you get the cost advantages from a workflow. Like generally most those two things, I think it's hard to imagine for how many use cases you're not going to care about those two things in an enterprise setting. Now, I find that where agents are very helpful and where LLMs are very helpful is in the creation of those tools. You're often trying to prototype or figure the thing out. You're going back and forth. You're not exactly sure if this is what you want or what you don't want. And so in that world, agents are so valuable because you're still trying to figure it out. But once you figured it out, then you're like, 'Okay, I want to lock this in.'
Yeah. Sorry, Andy. I just to jump in. Like that's fascinating Wade because you're right. Like it's my brain is tracking this and it's sort of exploding a little bit because it's like whoa I could use an agent to help me figure out a workflow and to actually build a workflow, an automated workflow in a way that I, you know, my particular if I'm relying more and more on an agent to sort of do things for me and I'm sort of used to like checking their work because they're not totally reliable and they can make mistakes and they've got some agency so you start to learn to like rely on an agent but trust but verify but it gets kind of wild because what you're describing is agents can help you build these workflows and then so that they're different but you're using one to help you with the other and so that's why my head's kind of going like whoa.
