In a lot of organizations, you're finding tools are not necessarily the problem. Culture is a big challenge. Do the leaders actually use the technology? Do they understand where there's opportunity and where there's not? You know, a lot of folks are getting pressures from their board, pressure from the CEO to say, 'Hey, we're going to go be AI first.' But they don't actually know what that means. You know, what does it mean to be AI first? What exactly are the use cases? Where should we be deploying this stuff first and foremost? How do we need to change our operating practices, etc., etc.
Yo, it's the future. Don't hit the snooze. AI is here the boost. Team humans rolling. Brains in THE GAME WITH DAN TURIN. Y'ALL REMEMBER THE NAME. PEOPLE in the no stopping the float. Better work, better life. Yeah, we're ready to go.
Robots and humans, a collab, you know. Okay, let's start the show.
Good morning, good afternoon, or good evening, depending on where you're listening. Welcome back to AI in the Future of Work, a very special episode of the podcast. This is episode 400. We've reached the four century mark and we got a great guest for today. I'm your host Dan Turchin, CEO of People, the AI platform for IT and HR employee service. If you haven't yet joined our weekly newsletter, do it. Mash that button in the show notes.
If you like this episode, and you better, or any other that we've published, you owe it to humanity to pay it forward. Share that episode or the podcast with just one friend. They'll know that you're invested in making humans better with AI. If you leave a comment, I may share it in an upcoming episode. Like this one from Tina, who hails from Munich, Germany, and is a startup marketer. Tina's favorite episode is number 295 with Thomas Otter, one of the all-time greats, general partner at Acadian Ventures about how one good decision made him an accidental HR tech pioneer. We will link to that one in the show notes.
We learn from AI thought leaders weekly. And of course, the added bonus, you get one AI fun fact. Here it is for today. Personal friend Isaac Sackelik writes at CIO Online about whether or not Agentic AI is the end of SaaS tools. In it, he says, 'While a total collapse of the SaaS model is unlikely, AI is fundamentally changing how software is developed and consumed. Recent benchmarks showing AI's ability to write the majority of its own code, and the rise of vibe coding have led some to believe CIOs might build rather than buy.' Key considerations for CIOs: Building makes sense for proprietary workflows, but only if the organization can maintain rigorous DevSecOps and governance standards. Software may become more composable and invisible, where AI agents interact with back-end systems via natural language. CIOs should push for outcome-based pricing rather than traditional seat-based licenses, especially as AI agents begin to perform work previously done by humans. The value of established SaaS providers is shifting from their code to their domain expertise and their ability to integrate with an enterprise's proprietary data.
Of course, we'll share a link to that full article in the show notes. My commentary on this one: Even when conversational AI can update a customer record, generate call notes, book travel, or review an expense report, there will still be a need for analytics, governance, and visual storytelling that go far beyond simple prompts. The decades of data and domain expertise owned by SaaS vendors in domains like CRM and ITSM and ERP and security would become increasingly important as AI homogenizes all the content. We have more analysts than ever thanks to spreadsheets even though Excel eliminated the data entry function. We have more radiologists than ever thanks to AI reading X-rays. In the former case, better tools led to increased demand for business analyses. It also led to better decisions. In the latter case, more people now get MRIs. Now that AI adds capacity to support higher patient throughput and lower cost. We all benefit from the disinflationary effect of technology. AI improves access and reduces the cost to deliver the same services. SaaS vendors need to adapt to outcome-based pricing and new conversational interfaces. All the hyperventilating about the death of SaaS misses the real point, which is that AI is disrupting how humans interact with machines, which you know what, it's net positive for everyone.
Now, shifting to this week's very relevant conversation. I've been looking forward to this one. Wade Foster is a rare repeat guest on the podcast. His first episode was number 217 back in January 2024, which seems like eons ago. Wade's the CEO and co-founder of Zapier, the workflow automation tool used by over 4 million people to connect more than 7,000 work apps. To date, Zapier has automated more than 81 billion tasks and more than 25 million Zaps have been created. Wade and his co-founder Brian launched the company in 2012, part of the YC S12 batch, and have built one of most impressive companies of our generation. They bootstrapped Zapier to a valuation north of $5 billion. They've done it by living their core values and humbly building products people love. Wade holds degrees in industrial engineering and business administration from the University of Missouri Columbia. Go Tigers. And with all that said, it's my pleasure to welcome Wade back to the podcast. Wade, good to see you, man.
Give us an update. What's new in your world?
Yeah, good to be back, Dan, for episode 400, no less. That's pretty cool. You know, we're just shipping as much stuff as we can over here in Zapier land. That's what's new.
More than two years ago, I said seems like eons ago, but oh my gosh, just think about how the world has changed. What stands out to you talking in, you know, April 2026 versus January 2024? What are the highlights in Wade World?
Well, I think the thing that is most obvious is what happened at the end of 2025. We had a new launch of frontier models. So, this would have been Opus 4.5, GPT 5.3, and Gemini 3.0. And that set of models was a step forward increase in capability. It went from, you know, coding agents that kind of worked to coding agents that actually work in practice. And, you know, obviously that might seem very valuable if your interests are in building software, but it actually has really big implications for all knowledge workers. Because it turns out if every knowledge worker has access to a coding engine, basically every knowledge worker has an engineer. They can help them do their job. And so much of the work we do day-to-day can be boiled down to like these little workflows that a coding agent can assist with, can automate, and can help you do your job way, way, way better. And so, you know, it's still early, but you certainly see engineers benefiting massively from this. But inside Zapier, we're starting to see every function — people, marketing, sales, finance, you name it — are all accelerating because they have access to these frontier models.
Wade, for the three people on the planet who aren't familiar with Zapier, and I assume they're all listening to this podcast, just a little bit about your amazing origin story and just share what Zapier does.
So, Zapier is the most connected AI orchestration platform alive. We started as an integration company. We helped you connect the tools you were using. So, Salesforce, QuickBooks, Slack, Trello, you name it. And quickly added workflow to the mix. And then nowadays you've got agents, you've got workflows, you've got our Zapier MCP that helps you bring those capabilities into your agent harness, whether that's Claude or ChatGPT or Codex or Claude Code, you name it. And so, you know, basically we've spent a decade plus trying to help you automate your work and make it so that your tools work for you and not against you.
One of the things that is interesting about where Zapier got its start versus the current world is that Zapier is all about automating deterministic workflows. So I know exactly what I needed to do, I'm going to automate the task, it's going to do it the same way every time. And along comes AI and it's all about non-deterministic workflows. So, I know you live it every minute of your life, but to others it might seem confusing about where the non-deterministic portion of AI complements the deterministic portion of the workflow that Zapier owns.
Yeah. Well, it turns out both of these are really valuable. In fact, most of us at work, we probably value deterministic workflows more than non-deterministic ones, especially in enterprise examples, because we want things that are reliable, cost effective, and do the same thing every single time. What you get from a non-deterministic workflow and what you get from AI is the ability to open up use cases, use cases that were really tough to be deterministic. And these often involve unstructured data. So think like call transcripts. You got a big old bucket of call data. This is common in the people world because you interview candidates all the time and you've got these call transcripts coming in. Well, how are you going to plug that into a deterministic workflow? And the answer is like you can't really. But with an LLM, they can help you do interesting things with it. So you might say, 'Hey, I've got a rubric to assess these candidates.' And so yes, the interviewer is going to like assess the candidate against the rubric, but also I want to have an AI take a stab at trying to assess against this rubric, too. And so they have a set of examples of like, this is a good answer, this is a bad answer, this is what we're scoring on. How did the candidate answer those things? And it can reason over the top of those answers and start to turn that unstructured data into something that's more structured. And that then I think lands you in this case of like these hybrid workflows which is actually where I think the world is going is you're going to see these sort of mix and match of deterministic and non-deterministic workflows where you're going to sort of reach for the capabilities that AI can provide in places where you can't get it from a deterministic workflow but you're still going to want to lock it together in a way that allows you to get the most reliability and cost efficiency over time. And so, you know, the people that I see that are most sophisticated with this are often doing those types of blended workflows.
One of the challenges introducing non-deterministic behavior, of course, is it adds risk because it's based on statistical probabilities and it's probably going to do something that you'd expect, but it might not. And we talk a lot, you know, in the industry about the importance of context and grounded data and tool calling, you know, using appropriate guardrails. Kind of talk me through how you and the team have approached that as Zapier.
Yeah, I think so. Our approach has been that we spend a lot of time building these like hybrid use cases where we see the most value out of using the LLM is actually to go build the workflow. Hey, I want to go describe this in natural language. I want to take these interview transcripts, compare them to a rubric, generate this output, log it in our ATS, right? I want to describe what that workflow looks like. But then when I go to run the workflow, I actually want that to look much more like code, like a program. And that's where Zapier helps convert that to a structured workflow that runs in a very deterministic way. And we can assert for you which parts might need to run through AI. So, for example, the part where we're going to go look at the interview transcript and compare it to a rubric. Okay, that's an AI step right there. But then the part where we take the answers, we feed them into an applicant tracking system, we notify this person. Like those things follow very rigorous steps that don't actually need an LLM to do them. And those are the things you want to happen in a structured way over time. And so, you know, that is the pattern that works the best is you can build with AI because it's so good at understanding human language and human context and you can kind of have that back and forth where you can say, 'Oh, yes, like that.' Or, 'That's close but not quite. Can you sort of adjust it to be more like this?' And then once you've got it locked the way you like, you say, 'Great, run that every time.' And that becomes a program versus an agent that is actually like making its own decisions every single time. Instead, you've got a program that runs in a very specific way every single time.
Very satisfying answer. Thank you. Those who have not used Zapier, which again is a dwindling few on the planet, but imagine there are a few, may not or may learn about MCP, model context protocol, and say, I no longer need Zapier because my agents are going to talk to other agents via MCP and that protocol was designed so that I don't need an orchestration layer. You mentioned that there is obviously an MCP server available via Zapier. Talk to us about orchestration platform, orchestration layer, how that complements or is complemented by the MCP protocol.
Yes. So this I think is really important. Increasingly the way folks like to build automation is they like to do it in an agent harness. So, you know, you might be wanting to talk to your tools inside of Claude or ChatGPT or maybe for the more sophisticated folks, they're already using something like Claude Code or Codex or Cursor or something like this that is like a more pure play coding agent out of the box. And often times you want this agent to go work with the tools and data that you have. You want it to work with your calendar, your email, your documents, your CRM, your HRIS system, like you name it. You have all sorts of data that are logged in disparate sources. And your day-to-day often involves going into these tools and doing work and compiling and pulling them together in different sources and then creating something new and pushing that out to a new world. So that's the workflow that most of us sort of live in. Now it turns out a lot of times when you go ask these tools to go work with another tool, you have a couple options. The first is you can say hey can you go connect for me? Now if you're in a coding agent like Claude Code or Cursor or Codex or something like that, the agent will often say sure I can help you. Can you give me the API key? Now on the surface this seems great. Oh, cool, I just have to give you my API key and you'll go do this for me magically. And for someone who has maybe not been exposed to engineering best practices, this is going to feel like an amazing experience. But quickly, what you realize is that copying and pasting API keys, plain text into files is not something that you should be doing. Engineers learned that long ago. But now we have so many more white collar workers exposed to these tools and are now able to build things in interesting ways. They're often learning the hard way some of these best practices that you should do.
Now what you get by using Zapier MCP or soon we're launching an SDK is you're able to come in and connect your apps to Zapier. So you can paste those API keys, you can OAuth into those tools through Zapier and now Zapier can help you connect and pull all these things together. So, you can say, 'Hey, I want to go...' like I'll give you an example for me. At the end of my day, I have my agent run a daily recap. And what it does is it goes and reviews all of my meeting notes. I use Granola. It goes and looks at my outstanding to-do items, and it looks at the ones that I completed from the day. It goes and looks at all the Slack messages I sent and the threads that happened from the day. It reviews my calendar for the day, and it reviews all my email for the day. And then it packages that all up and says, 'Here's what happened today. This, this, this, this, this. Here's what action items you have for yourself based on what I saw in Granola and all these other things. These are the important things you need to do next. By the way, Wade, do you want to add any extra context to this?' And, you know, usually I'll do a quick voice to text and say, 'Yeah, here's two or three things I think are really important we need to remember.' And then it compiles that all into an end-of-day recap every step of the way. Zapier is the tool facilitating interaction with all those data. So it's reaching out to Granola and pulling those notes. It's reaching out to Slack, pulling in the context there. It's reaching out to Google Calendar, pulling from there. It's reaching out to Gmail and pulling from there and helping me put together this tool that then follows up on all those action items. You know, when the agent sort of says, 'Here's the action items you need to do.' Often it'll say, 'You need to write this email. You need to do these Slack status updates.' Then I'm saying, 'Great. Can you go ahead and build those for me?' 'Sure, got it.' And then Zapier is reaching out and drafting the emails in Gmail. It's sending the Slack messages out that way. And so it's so valuable when you as a knowledge worker have access to a tool like Zapier or MCP or the SDK that can bring all those tools together in a governed, trusted way, combine them with an agent like Claude or Codex or you name it that can now go out and interact with the world on your behalf. And that I think is where there's going to be an increasing number of knowledge workers that are going to realize like this is so powerful. It's a super way to work for everyone, not just engineers, but for everyone. I've been working this way now for four or five months. I'm not going back. It is so excellent to work this way.
And you also, it's auditable, right? It's one system of record that you have where you can see all of the tool calls, all the workflows that have been run on your behalf.
Yeah. And it benefits me in a couple key ways. So like this daily recap, the fact that it's drafting those emails, the fact that it's following up on these Slack items, all these things at the end of the day, it used to take me an hour to close out my day just to follow up on all these threads every which direction. And then often times I didn't even get to all of them. Like some of them I just had to be like, well, that's not happening today. And now in 5 minutes at the end, all this stuff is teed up for me and I just have to do quick approvals. Press send, go, go, go, go, go, go. And so I'm able to stay on top of all of this stuff. Then what happens is I start to find that I have more blocks of time in my day. And then it gets really interesting because how do I want to go spend that time? Often times I'm thinking, hm, you know, do I want to come be on another Dan's podcast, you know, for episode 400? Do I want to go work on a new marketing campaign? Is there a new product that I want to go launch? It sort of opens up this time to like really go invest in something that's going to propel the company forward rather than me having to go chase down loose threads all over the place that don't uniquely require my expertise and skill sets. You know, I think those are the places where AI is so magnificent is it can do those things and it can lead me to do things that I think I'm uniquely valuable at versus just being sort of a glorified errand boy for certain things.
