FLASHBACK: The future of remote work, juggling APIs, and dream integrations with Wade Foster of Zapier | E2221

23:41Like if you talk to an apartment, they ask what ads are running. They're spending$10 ,000. They literally don't even have any idea. They don't know what's running. They don't understand. They get shocked when we show them these are some of the images that are running. They don't understand, right? And now with certain pieces of infrastructure, we are showing that in them. We're teaching them that they can make these changes. All right. Listen, continued success. Where can people find out more? Please visit usetour.com or more. We love to talk. And if you're feeling inspired, you can go to gamma.app and start working on your idea and make an incredible pitch using this incredible software.

24:18All my startups seem to be moving to Gamma.app. It makes chef's kiss beautiful presentations that engage investors like me. And maybe we'll see you here on a future episode of This Week in Startups. We're talking to a company that I've been waiting for for a long time. If you've ever worked next to marketers, you've heard the term attribution followed by a forehead smashing into desks and a string of profane words. It's a tricky, tricky thing, but it's very important. That's why Olympic is interesting to me. It's a company that thinks they have a new way to help marketers attribute information to spend to results, and they're gonna take it in a broader context later on.

24:59So please join me in welcoming to the show. It's Tomas Puig, founder and CEO of Olympic. Tomas, how you doing? I'm good. Attribution, in my experience, working with people in marketing, it has been the never-ending problem, the impossible black box. Can you just tell people out there who aren't as familiar with the issue, why it matters so much and why it hasn't been solved yet? You know, I think the answer is it's complicated, one. Two, the reason why it's so important is that you really have a couple different types of marketing that occur. One they hear about a lot is like your Google ads, your Facebook ads.

25:28We call that programmatic marketing, right? It's very technical, very driven. It's very good at finding people who are interested in a thing, reactivating people, stuff like you would expect. But when you want to get net new people who've never heard about you at all, you want to broaden the market, you got to go big. You're going to do like a Super Bowl commercial, or you're going to sponsor the World Cup if you're international. We had a customer we worked with who spent$100 million over seven years to sponsor one major league team. These are incredibly large brand expenses as we talk about them.

25:58This is called brand advertising. It's called brand advertising, but it's just basically the first time you've ever heard a company. You usually go with that. And there's been a lot of amazing movements in being able to understand how spending a dollar on Facebook for an ad might get you money, but for the big stuff, right? Whether it's influencer work, whether it's a Super Bowl commercial, whether it's the Olympics, we know it works, but it's very, very hard to prove. So how do we know that it works? Because I know return on ad spend, every marketer in digital context talks about how much money they're putting into Facebook and getting out of Facebook and huzzah, raw ass.

26:30I've always been skeptical of the crypto.com arena or, you know, Gemini sponsoring the McLaren F1 team, because it seems like such a kind of an incongruous connection that it shouldn't seem to work, but people do it. So it must, but how have we known that it does before we cracked this attribution problem? You want to know one of the most interesting things we hear is especially about like, say the F1 side of the house, right? That you just brought up. There's a different thing that comes with a lot of these brand spends that's hospitality. And so like, remember marketing is not just like being a net new brand, but it's also having the best box in the world at the F1 race.

27:03So you can make the CEO of Fortune 500 show up. So it's essentially a big, a big lure to get people to come hang out with you. I mean, isn't all advertising to a certain extent that? True, but I was, I think about it more as a business to customer versus business to potential partner conversation, but maybe I'm thinking too narrowly. We work across the board. We work with B2B, B2C, right? So B2C is business to consumer, right? So selling, you know, whether that's a plane ticket or your favorite soap, whatever. And then there's B2B, which is like a company selling to another a company. You know what we always say the hardest thing about being a marketer is everybody thinks they're a marketer.

27:35So it's actually, you've got a very sophisticated thought process to it where you think you go, okay, marketing has a number of different purposes. One would be just get a new person in the door. The other would be, how do you sell something for more because it's more valuable? How do you make a deal sell faster because people have heard about you? You have to do less validation. So all of these things, when you're talking about like stuff, you have to actually think about the world in a much more sophisticated fashion. You can't be like, I have a ROAS for$1. Well, what if it's terrible at that, but it doubles every deal size?

28:01So if you don't think about just marketing as a digital advertising event, where you put in a dollar to get out two, in theory, you can have disparate impacts over a longer time horizon. And that's probably why it's difficult to sort out what's actually driving impacts. And this brings us to Alembic, your company. I think you're kind of on the edge of science fiction because I was reading about spiking neural networks, which was a new term to me. You guys use those. And you talk about causal AI as a way to infer connections between cause and effect in the marketing context. So tell me how this all works.

28:35We started the company quite a while ago, right? Really got, I think we did our incorporation papers, like just playing around in like late 2018, but we really got started in 2019. When we first started the company, we thought that what we wanted to do is we wanted to build Datadog, right? Like, which is a signal processing company for unstructured data. We were like, okay, we want to be able to understand all things coming in marketing, whether it's brand, whether it's this, whether it's anything, right? In that space, be able to look at and actually see what happened because that was the biggest problem at the time.

29:04And then as we started doing that, and we brought in these enormous data sets from some of the largest companies in the world, we had a really intense technical staff. The reason we thought that this was such an interesting problem is you are basically dealing with one of the most complicated data sets in the world. It crosses every medium. Think about what we just talked about an F1 race to TV, to a digital ad, to foot traffic, literally every medium and every modality. And you have to understand a huge amount of the world. So we started doing it that way. And then what we discovered was there's really two fronts to kind of the tech that we talk about a lot.

29:37One is just how do you even see the thing, right? The signal process. And the other is the causal connection. And they're different. You know, anybody who's worked in business for any amount of time has had this moment happen where they're talking to somebody on the technical side and they We go, I would like to predict something or I would like to know about something. And someone goes, well, we don't have enough data for that. One of the hardest things in marketing is we have, we did a famous case study with Delta Airlines for the Olympics. The Olympics happen for two weeks every two years. If you have daily data, which is actually pretty aggressive for a marketer, daily data, you would have 14 steps of time, 14 days.

30:12So how do you tell whether something's different or something changed when you have zero historical data? And Delta is the Olympic sponsor, right? Of the US team, which I recall from watching the Olympics. Shout out Delta's marketing team. Your startup needs custom software, but building your own secure production ready apps is hard, right? And not every founder has the expertise or time to get started. Well, now you don't have to just use ZITE, Z-I-T-E, make whatever you need from idea to a working app in just minutes, easily all on your own. Plus you can fill it in with forums, apps, databases, and other automations that you need as part of your product.

30:46So many of the no code and vibe coding tools are just flashy demos where they work on legacy systems that take so much time and effort to learn, let alone master. But Zyte is both easy to use and powerful enough to build out the apps you want. And it quickly connects all the other tools you're already using. For example, earlier this month, they added the ability to bulk create and update records. So it's easier and faster than ever to design data heavy apps that are constantly syncing between systems. That's the kind of attention to detail that sets Zite apart. Start using the number one AI powered business software generator on the market.

31:20I'm going to give you 50 % off your first project. Go to Zite.com slash twist to get started. That's Zite.com slash twist for 50 % off your first project. Yeah, you know, the most interesting thing that occurred with Delta in the Olympics is they did this huge sponsorship and it did well for them, right? It was fine. But the thing that actually was the most effective advertising for them. The thing that actually moved people the most to both consider, purchase, et cetera, was actually not the 30 second commercial or the 60 second commercial. At the time, Delta had worked with NBC. And one of the things that happens when you buy an enormous amount of media, right?

31:57In some cases, you could be eight, nine figures, right? You can buy a lot of media. They start giving you in programming stuff. So if you watch the Olympics, you saw that Delta had the medal presentation ceremony. So every single time an American athlete won a gold medal or a medal. It would be the medal presentation sponsored by Delta. Well, imagine you see a super emotional moment on TV. You see the Eiffel Tower in the background. One would assume you sell a lot of tickets to Paris. And so that most effective ad was actually not 30 or 60 second spot. It was that emotional moment that was actually in the program.

32:26Wouldn't even be considered an additional ad. In fact, you don't even necessarily know where to account for it because it usually gets brought as part of the package. Okay. So this is where I get really curious, because if we are going to ingest all this data and then use your particular form of neural networks to deal with it, how do you know if someone saw the emotional nettle ceremony? Because I presume that me, Alex, on my couch is not being counted. I'm a YouTube TV subscriber. I don't know if you know. So talk me through the links there. So where the big inspiration for all of this came from was a lot of the mathematics that came around during COVID.

33:05A lot of people think mRNA, the vaccines were the only big technology to come out at the time. The other thing that came out of it was a lot of incredible mathematics. Do you remember contact tracing? Yeah. So contact tracing, they had these ways where they could fairly anonymously see whether you touch somebody else. They had all this big indulgence at the time. But it was also the first time we had both the compute and the causal mathematics at scale to be able to do these computations, these huge simulations and see in real time and actual with it to see whether we're correct. And then I woke up one morning, I was like, wait a minute, what is the difference between a contact trace and a marketing touch mathematically?

33:38I'm going to guess very little or nothing. Nothing. Right. Okay. So for every other industry, nuclear, weather disease, drug discovery, we run these huge simulation models on super compute to try to figure out what's going on. We spend a trillion dollars a year worldwide on brand marketing, and we all use math from the 1970s. And so we brought those models down. So you asked, you know, think about like this, we actually have no PII in the entirety of Alembic. Personally identifying information. Yep. No personality identifying information. We don't even have the fields to store it. So everything we do is anonymous and aggregated and privacy respecting, but you won't think about like a singular human.

34:13You would think about it like a cohort, like a group of people. Perfectly fine for when you're trying to figure out what's going on. When we talk about causal AI, and I would love it, Tomas, if you would define that term for us. We're not talking about Alex saw X, Alex bought Y, the generated X, Z revenue for company Q. It's more like this event or thing led to a group of people doing something in particular. Yeah. So you're going to think about it like to respect both the privacy of the individual and also just keeping it sane. You look at things in what we call cohorts and groups of people. Okay.

34:46Got it. And so like, say you have a bunch of users who sign up on a week and you see how quick they quit, right? That's a classic cohort. You know that there's a group of people, et cetera. What you care about in marketing is, it's interesting, people often think that folks of marketers are like the stalkers who want to like obsess over the singular human. That's often not the case. Marketers just don't want to get fired. And so they just want to be able to know what's making money. And for them, when you're holding up, say,$75 million,$200 million on the P &L line, right, the profits and loss sheets, you better have an answer.

35:18And so that's where all of this comes from, right? It doesn't come from a place of anywhere else. And so when we were doing causal, right, we need to be like, look, there's no world in which somebody just watches a TV ad. And within 24 to 40 hours goes and buys something. Are you that person? Depends how late at night it is and how much Taco Bell is on the television. But yeah, keep going. You know what I'm saying, right? It's very rare that you get that tight. And so usually it's like, you see something on TV, you Google for it on your phone, you then search all the threads on Reddit, seeing whether this thing sucks or not.

35:49And then you buy it, right? It's much more complicated. And so that's why we do this type of thing where what happens at the moment is anything that you can see physically gets the credit, but everything outside of that might not. This also has a huge amount of negative effects when we talk about people doing this. So think about like rage bait and click bait. That's because you don't know if anything else works, right? You have a feeling that other stuff would. To go back to your early question, because I do want to answer it because We got one on two separate topics. First, what is a spiky neural network?

36:18When we have very little data, like those 14 steps of time, you can't use what we call a predictive set of technologies. You can't look at something and be like, I've seen a stock trend for the last 30 years and project what it is next week. So we had to find an outlier detector, something that could detect changes in things that required no historical data. Spiky neural networks are actually a digital twin of the human brain. You got kids, right? Yep. The first time they saw a dog or a cat, did they always know it was a dog or a cat after that? It's like there are certain things that happen in a kid's life where they're like, like it's very tight.

36:50It's a very short period of time. It doesn't take that long for them to go woof woof or meow meow. Exactly. And they go like, it's very thing. So when they see an object, like human sees an object, what it does is it can implant that pattern into its brain. So just like I'm looking at this table or I'm, you know, I touch the screen in front of me. All of that gets encoded into a neuron spike in my head. And a neuron spike is essentially a signal. It's a signal. I don't want to get too much into it on this, but the best way to think about it is that like the instant you see a signal, you imprint the shape of that signal onto that neuron as a threshold.

37:22So like a thing that would make it fire. And then when the next piece of data comes in, it looks at it and goes, hey, am I the same? Am I different? It's actually doing a comparison. And so when you're looking at data and you're like, nothing has existed, suddenly the Olympics come along. That pattern, the shape of the data is literally different, right? We know something new is occurring. This obviously has much stricter mathematical things, but it's not all like an LLM or anything else. It actually happens at the time you receive data. And because an LLM makes essentially predictions, it needs a lot more data.

37:49And so with your 14-day Olympic window, an LLM is just not a good fit for that circumstance. Yeah, it's not that those aren't awesome. It's just that it's bad for this. And so the way to think about what the reason SNNs are great, one, they're incredibly efficient. But two, they let us be able to look at data in a way of which if you have stuff that's random or you've never experienced before to be able to understand it. Now, you can combine this with other methods to be able to see things. So when we are able to basically predict these outliers from no data at all, we get from the day we see them.

38:18That gives us a deep advantage in being like, how does our model understand things? When something matches that shape that fires, we can be like, look, that neuron firing, did this neuron firing cause this neuron to fire? This is also how you solve what we call the normalization problem. Think about it like this. So you've got a whole bunch of signals. We talked about earlier, foot traffic, Nielsen ratings from TV, right? Digital signals from Facebook. How are those all the same? They're different mediums. One's physical, one's digital. They're different modalities. How do you actually put them into a same space to even make a decision?

38:49That's because everything becomes neuron spiking, so they normalize, right? You're looking at the same thing. So we can be like, output is the same even if the inputs are different. Exactly. And so it puts it into a common space, a common embedding space, as we often call it. This common space means that we all speak the same language and you can compare things and be like, did you cause you, etc. Now, the causal connections are actually completely different. So we have to first find these spikes. Then we can run a whole different set of massive mathematics to figure out whether they caused each other.

39:16They're actually different models. Okay, so they're actually distinct technologies that work together. That makes a lot of sense. So one, you have to figure out what are the things you want to connect to even be able to see them. And then the thing that makes the connections is a different piece of tech. Now, the thing that we do that's a bit more unique as opposed to others is most of the time when people are looking for causal connections, they're like, I have one thing I'm tracking, like sales or instances of disease or whatever. And you do everything looking backwards. You want to reverse the butterfly effect from that one object.

39:43What we do is we connect when we connect data, we actually connect everything, every single connection that possibly exists. That means sometimes we're calculating trillions of connections for a single customer. What that does is that creates a basically space that we can search for any connection to anything. And then you can be like, in general, does this cause this? Is the information that you get out of that sufficiently detailed, accurate? I'm not quite sure what the word there is to help marketers go, okay, now that we know this, we're going to reduce spend here and put more spend here because we have an 80 % confidence interval.

40:20Is it 100 %? Well, it depends. We actually give our customers their confidence intervals. So we're actually very transparent on it. Are you just a mathematics project pretending to be a startup? And I mean that as a compliment for what's worth. Can I note something? One of the things we often joke about is we're an applied science company. What we would say is so are all the LLM companies, right? These labs and everything else. I have a bit of a different take on all this. And what it is is this. I like, look, I actually adore the LLM tech and use it all the time. We even use it for UX. But the interesting thing to note about it is I was talking to a Fortune 500 CEO literally in the last couple of weeks and they were like, OK, CPG, here's my problem.

40:56So say I get the best model in the world, whether I decide it's, you know, Claude or ChatGPT or whatever. And say I give it to my staff and my guy out of New York asks it, what's the best tactic to go get more sales here? And then my competitor, who has the same model, asks it the same question. And then we both do the same thing. How in trouble are we? Well, it's not a Nash equilibrium. I'll tell you that. So what I view the world as, as I say, if we don't want the world to become a giant LLM oligarchy, the way that you think about it is you think about it like the only thing that will separate people is private data.

41:31That's where all the improvements will come from. So I many years ago said that, look, I want to build the best private evolving model in the world for our customers. And then as we get bigger and bigger, we'll obviously want to service lower levels. We start big on the enterprise. And then what that does is because we can see over long swaths of time, because we can find these things that nobody else would know that's unique to them. Then if you layer it in with your agents and everything else, you still have individuality. You still have a different view of competing in the market. Because you have your own data feeding it and that's private and therefore it's only yours.

42:03So no one else will have the exact same view of the world that you do. Exactly. And so you can make better decisions. There was a great guy who found a company called Renaissance Technologies. Oh, the hedge fund. Yeah, super famous. The first real high frequency trading crazy hedge fund. I've had friends that worked there. Oh man, brilliant folks. He would say that, look, all edges you can get in the market, all advantages have capacity. So if you have a big edge and you start making huge amounts of money, the market will notice. Everybody jumps in and that edge disappears. And so what I'm saying is, as you have private data, you can generate your own edges that can give you advantages over that period of time.

42:40But it's not so simple as just being like, how do I do it? And so what I would say is that like, from my point of view is I'm like, look, I think agents are very cool. I think they're another word for API often. I don't care whether an agent is a human or a machine. I want to be the agent, a controller. And I think that we have an online evolving model that provides that individuality. And that's how you end up with more Star Trek than Blade Runner. Good, which is what we want. And if you don't get that reference, read a book. So we're talking about a lot of compute here, Tomas. And this feeds into your last fundraise,$145 million.

43:11And you guys talked about building a second kind of cluster, I think out in San Diego. You had one in Virginia. San Jose, actually. San Jose. So sorry. Yeah. It'll be ready January, I believe. Fantastic. But an expensive proposition. I'm curious, for the other founders out there who are purchasers of cloud computing and other GPU access points, why did you guys decide to build your own clusters versus going the rental route? Oh, boy. A lot of reasons. One, just we have the skills for it. So my COO was one of the heads of cloud and infrastructure at Google during the IPO. And then he built LinkedIn with Reed.

43:46My co-founder, I'm kind of like the math nerd, but my co-founder on the infraside, John, he was the 13th employee at Twitter and scaled it from a Mac mini under his desk through IPM. Right. So we know how to stand up machines. The second thing is, is that we actually code our own GPU kernels, as we call them, our own low-level GPU software here. Because what we're doing is not the same as everybody else. We can't just go and use a package that everybody's already built to do what we do. So Alembic then is a system by which it can ingest lots of different types of data into one single place. It has a particular neural network approach, which is using a digital twin of the brain to mimic or simulate neurons.

44:25Then from there, you have a system to tie cause and effect causal AI. And you're applying it today at this particular marketing attribution world. How long, Tomas, until it's time for the next item at the top now of your website when you add a second vertical? And what will that be? So the funny thing is, is that one of the things I love, I jokingly say, you don't ask if you have product market fit as a startup. You hold on for dear life. I mean, like you're holding onto this rocket ship and you're like, if you ask, you don't have it. Right. You're just like, I cannot even get servers up fast enough.

45:01I'm dying under whatever. Is that Alembic today? Yeah, absolutely. Like we literally, we have a wait list that is quite long. Our next vertical will probably be FP &A, which is financial planning, like at the corporate level for like CFOs. Okay. You know what? I actually really think that makes good sense because that stuff is complicated. It's stuck in legacy software. No one likes to say ERP out loud. I don't think anyone actually knows what's going on inside of most companies, including their accountants. So, yeah. The more interesting thing that we've been working on is, like, say, a tariff gets announced.

45:30One that we do a lot of what's called graph work, right? So we have these, you know, it looks like connect the dots, but they call them this physical structure graph. And what we do is we look at it through time, right? So we have this thing moving through space. It almost looks like stars moving through space, picture of your head. So what ends up happening is when you have a big event outside, we call it an externality. like someone announces a tariff, you can literally physically watch it warp the graph beneath it. We can literally detect the effects of external events on the data sets below it.

45:58And so when we started seeing that, we were like, oh, that would be really, really useful. It's alembic.com. Tomas, thank you so much for coming on. We could talk for two hours. Is there any job though you're hiring for that you'd love to shout out before we go? Yes. I mean, I'm looking for anybody who is insanely curious and has a scientifically based mind that likes to code. But literally, we're looking for everything else, whether that be go-to-market or the rest. One of my favorite things about the company is a huge amount of original employees, even from years ago, still work here, right?

46:25Take a look at our site. Definitely hiring. Well, thank you so much. We'll talk to you soon. Tomas, a pleasure. We'll see you all next time. Good to see you. We're doing another This Week in Startups flashback today. We're going all the way back to March 1st, 2016. And I was actually, when I was going through the archive, putting together clips to make these show out of, I was really surprised to discover that this company existed in March 2016. The company is Zapier. On episode 2626 of This Week in Startups, Jason had Zapier CEO and co-founder Wade Foster stop by. This is his old studio in San Francisco.

47:02For those of you longtime fans, you'll recognize it. We use Zapier very regularly today at launch. It's one of the tools that we use all throughout Notion to help sort of tie all of the things that we're doing together. You know, we're spread out a bunch of different portfolios, a bunch of different sort of CRMs and management systems across different platforms. And we still use Zapier to like, if you tag a button on Notion, it'll trigger an email or it'll send you an update here. It'll post to Slack or all sorts of these integrations that we still use today. The fact remains, 10 years later, after Wade was on the show, itching this idea to Jason, Jason was like, oh, Zapier, interesting.

47:42Tell me more about it. Reminds me of If This Then That. It's not only still going and still a viable company, but it has grown and become so entrenched into so many of our workflows. So that's what I thought was really fascinating to go all the way back to 2016 and find this chat. And of course, with me to talk about it, Alex Will. Thank you, Lon. Zapier is a company that we all know, but I'd forgotten that it was founded back in 2011. So by the time Jason talked to Wade, it had already had a half decade under its belt. Now, it hasn't raised that much money, which is very interesting. It did raise a seed round,$1.2 million, October, 2012, Y Combinator, Bessemer, DFJ, and then nothing.

48:20Because as we learned during this interview, the company was cashflow positive. Now, Lon, I will say that more recently, Sequoia did buy some shares of the company at a valuation of more than$4 billion. dollars, but I think Zapier is a good indication and reminder that you can build enormous companies that are very important to the fabric of the digital economy without needing to burn 100, 200, 500 million dollars. I want to start long with this discussion about multi-part Zaps. Now, the idea here is that you can take a Zapier, if this, then that kind of statement, and then chain them together to do more complex, interesting things.

48:52Today, a little bit table stakes at the time, rather innovative. But what's interesting is how curious Jason was about them because he was curious about what could be automated inside the realm of white collar work. Take a listen. So Zapier now does these multi-part zaps, stacked zaps, or I call them zap bots. Obviously, this is like an amazing way to take something, a repetitive task that a white collar worker does and basically automate it. So now if you have a bot and, you know, this white collar worker in human resources creates the entire scheduling bot, can the scheduling bot then fire off another bot?

49:31Let's say there was a bot that somebody created in the developer team on like, let's say this was the sales team and the sales manager created a sales bot, a zap bot to do sales testing to test somebody's ability to sell, the HR bot, when it got to step seven, might be when the person is here in the building and they've logged in, fire off this test and send them through this test. Is that what the next step is to have multi-level zaps, querying other multi-level zaps? Yeah, exactly. Right. You can see all these cool ways that interacts with them. In fact, you mentioned when someone's in a building, there's a company called Envoy, which allows people to check into office spaces and collects their information, prints off badges, things like that.

50:19So that's exactly a tool on Zapier. When you're at the front desk, you sign it and you're like, sign this NDA that nobody ever reads that says if you see something here, you'll never be able to create it in your life again. There you go. So that's them. But yeah, they can hook into Zapier. So that could be, you know, okay, you're chaining off this app. Now when they show up in the building, we need to do this whole other second set of criteria that follow up from that. So what do you think the future of work is going to be? Because I'm seeing that we don't need as many clerical people to run a larger and larger operation here at launch.

50:51For example, we have, I think, eight or nine people, and the launch festival has gone every year from 1 ,500 people in year one to 15 ,000 in year eight. We've literally gone 10x in eight years. But the same number of people, because a lot of the things that used to take a lot of time are now becoming more automated. So what's the future of employment, do you think? Because a lot of the low-skilled, white-collar jobs are being automated. It's a hot-button issue. Yeah, it is. And we can put aside the fact that you and I probably believe that more jobs will be created because of this. Yep. But what's the way for somebody who's listening to become proficient at these things, do you think?

51:31And what will the future of work look like for those clerical people? Will they be more like developers? Is everybody going to be a developer? Because let's face it, this is like scripting. Yeah, it is, right? Like this is kind of stuff, people get hired to write this kind of stuff today, right? Like that's what Brian and I were getting hired to do five years ago. We were getting hired to write code that did this stuff. Yes. So now like a, you know, a savvy person could, could probably hire themselves out and pretend that they're writing code, but it's actually Zapier behind the scenes if they wanted to, right?

51:58Right. Are people doing that? Yeah. Some people do do that, right? They, you know, say that they, you know, are going to build this stuff and then they do it and then they wait a week and then tell them that they've delivered. Yeah. Right. So some clever, some young aspiring entrepreneurs right there. So I think in the future you're going to see people just using more and more of these tools that make, I hesitate to say it's coding, but it's like quasi-coding, right? Scripting? Yeah, it's basic scripting. Macros? Yeah, macros. That really puts that kind of power in their hand and lets the tools kind of just get out of their way, right?

52:36It's like the thing that's more interesting to you is how do I put on a good event, right? Who are the people that I need to have at my event that can talk and give interesting discussions and engage the audience and all that sort of stuff? How the ticketing system works is just a detail that you don't want to think about. So as much as that can be automated and pushed into one simple setup thing, that's what you want to do. And I think that's how these tools are really enabling more and more people to set these kinds of workflows up. So now you don't have to be as savvy as an internet user as you maybe had to be five or 10 years ago.

53:14So Lon, to me, this sounds exactly like the conversation we're having today about automating white collar work using AI agents, which are more flexible and more intelligent than a simple schema setup in Zapier. But my God, I feel like we're having the same chat still. And he even brings up If This Then That, which was a very popular tool sort of like 15 years ago back on the web. And it was always about the same. The end goal, I feel like, really hasn't changed, which is I want the computer to do more things for me. Like, I can automate some things, but frustratingly moving between platforms, moving between systems, moving between languages.

53:49There's all these different variables that then require me, the human, to sort of, like, intervene in the middle and say, oh, no, do it like this. Or, oh, no, move it like this. Or even just, like, something as simple as copy and pasting. We're still doing it all day, every day, constantly. And if you're moving, if you're on two laptops like me, you're doing it all. Believe me, you're doing it all the time. And so I think the promise, the promise of a computer is going to get smarter to the point that it's going to know you've done this 100 times. You want to do this 101 times. I should just do it for you.

54:21That promise has been so tantalizing and so just out of reach for so long. and that it does feel like the AI agent era were like on the precipice of maybe we're finally gonna figure it out. Maybe I could finally get my Amazon cart bought and unloaded to my house without me having to click the button. I'm gonna believe we're actually there when I'm doing something for the 14th time of the day and a little bubble goes, bing, Alex, you're doing this again. Would you like me to do that for you in the future? And then we'll be there. But it's interesting. We've been working on it for a long time. Not there yet.

54:52Zapier though, crushing it though. Now, Lon, while we're thinking about the historical context here, there was an era under which APIs were a bit more locked down than they are today. Yeah. Well, especially I feel like the Twitter era when social media was sort of exploding and there was so much happening on Twitter. So many conversations were going on. There was this huge wave of apps and other products that sort of built themselves on top of that API. Like we're going to, you know, like TweetDeck very famously, Periscope very famously. There were so many of these examples. And I think that we then learned this sort of valuable lesson, or anybody watching the industry learned this lesson, which is if you rely heavily on one or two other companies' API to build your company, it's sort of an unstable foundation.

55:37They can yank that API away at any time. And now suddenly you don't have a company. And we are still, of course, seeing this. Gummy Search, a tool that I love that was built atop Reddit that used AI to help you search Reddit. it's being shut down because Reddit has shut off their access to the API. So we are still sort of living in this world, but you kind of forget about it now because websites are, they're so eager to interconnect and yet everybody learned their lesson. And so founders really aren't building entire companies that are like, we scrape Twitter and then do, well, Twitter is X. I don't feel like there's a whole wave of companies that are built atop other apps that we all use because everybody sort of learned this lesson.

56:17And actually in an interesting way, Jason was just sort of recently talking about this in terms of open AI. Like if you build your product atop open AI's API, they're going to be peaking at what you're doing and it will be very easy for them one day to be like, we should start doing that and rip you off. So we are kind of still having this API conversation. It's really fascinating to go back 10 years and hear about an earlier version of the same discussion. So speaking of APIs and Twitter, how is Twitter and them pulling back on their API and then saying, then changing their mind now? When you look at that kind of change, did that change how you build your service?

56:56Because there's some limits, aren't there? There are. And you see this often with consumer platforms, right? Who are figuring out their platforms. Twitter has done this. LinkedIn's done this. Facebook is notorious for doing this. Pulling the rug out from underdeveloped. Yeah, or just changing things, right? It's not always pulling the rug out. It's just, you know, they're trying... Just in those three cases, it was. It was, yeah. But they're trying a new thing, trying a different thing here, right? And the thing there is that their customer is ad buyers, right? And so that's who they end up serving.

57:27And so the API platform has to serve the ad buyers as well. And so that kind of creates an interesting friction that you don't necessarily get in B2B, right? You know, when you're looking at Salesforce developer platform, it's all about empowering the end user of Salesforce.com. And so the API is built to empower people to use more of Salesforce and get even more out of Salesforce. They don't care if you log into Salesforce every day. They care if you're getting value of the data that's in Salesforce. So if the API helps you get data into Salesforce and pull data out of Salesforce and makes Salesforce more sticky, then of course Salesforce is going to invest more in their API.

58:05The more integrations you build, the less likely you are to get rid of that core component. Exactly, right? It's a building block. It is. And that's what you see with Slack blowing up and why they've done so much to invest in integrations is because it makes Slack sticky and you don't want to leave anymore because of it. It is interesting. One of the reasons we didn't leave HipChat for Slack, even though some people on the team were like, well, I'm using Slack for this other project and we move over to Slack. I was like, you know what? We have so much Zapier plus HipChat. I don't want to spend a weekend or two like changing all the integrations and then the people who are on Slack with my other company, inside's on Slack, not, and then launches on HipChat.

58:45They have the opposite problem. It's like, oh yeah, we built all these custom integrations that they built themselves. You can't get off Slack, right? Yeah, yeah. What happens when someone leaves the company? And yeah, I mean... The whole thing is super complicated. So with Facebook, LinkedIn, and Twitter, then you're saying, hey, you could remove the value, the core value could be extracted from those consumer platforms, and then a third party could kind of take away the business that should be their core business. Right. That's their concern. That's what Twitter was worried about, right? With people basically replicating the Twitter feed in Twitter clients, right?

59:24It's like now you don't have to use Twitter.com or our officially sanctioned Twitter clients. you're using some third-party client that maybe isn't pulling in ads in a way and showing ads to users. Was that a mistake? In your mind, would you do the same thing? It's tough to know if I was in their shoes. It sure angered a lot of developers, right? And I think it seemed to have hurt them, right? You don't see nearly as much interesting things happening within the Twitter developer ecosystem as you did four or five years ago. So if Twitter were to allow people again to make their own clients and to display the feed however they wanted and just have fun with it, that would probably be a good move.

1:00:09You know, I think it would. Maybe, I don't know if it's too little too late or, you know, if, you know, it's certainly, I don't know that it would hurt too much, right? If all I have to say is you can't monetize it. Monetize. Yep. I say it wrong. Monetize. But you can't monetize it, right? So if you want to do it, you just can't insert ads into the stream. If you want to put ads in, you have to put our ads in. And so if you created Wade's and I created Jason's Twitter client and we had different ideas of why ours were better, we could put in Twitter ads and then take a percentage of them, but we couldn't put a third-party ad in.

1:00:41That would be a fine concession. Yeah, it might be. It all comes down to Twitter's operating a billion-dollar scale, right? So now the thing is, is Wade's Twitter client or Jason's Twitter client enough of a big deal for them. And I guess they probably run the numbers and said, you know, it's just not that big of a deal for us. But I don't know. I'm not Twitter. It's interesting because if you think about it, Google allows you to do search from Firefox, Chrome, Opera, or any flavor of browser, Dolphin. And they give you 30 % if you have your own browser of whatever search results come up. So like we're getting 70%.

1:01:20Yeah. So if I was running Twitter, if I was Jack, you should just say anybody who makes their own client can take 30 % of whatever revenue is shown in that client. Because it would just drive people to create more interesting clients. Yeah. So if I want to create just basketball, just all the basketball stars, you open it up, it's basketball Twitter, and I take a third of the revenue or 30 % of the revenue, they're still getting 70%, and it might make people who are into basketball more interested in Twitter. I also like that Jason shouts out HipChat in this discussion. a service that in 2016 felt very relevant.

1:01:55And I honestly don't know if I've thought about much since then. It's funny how it all comes down to money, because in the conversation, Jason's like, what about ads? Are they getting served? How does the money work? And companies like Reddit were like, hey, we want to conserve monetization. Reddit went through a similar lockdown of its applications. You mentioned Gummy Search, but this actually happened as well a few years back when they were crimping down on third-party apps. And at the time, people were like, oh, they're being greedy. They just want us to use their application. No, it was all about AI training data.

1:02:20And so the idea of open or closed gardens, who can get over the wall for what reason? How does the money flow? It's almost like long, there are some things in technology that are just business and therefore will never actually be fully resolved. The other sort of very famous use case for this would have been like the Zynga games and Facebook, where for a long time it was like, you got to go Farmville. And then Facebook was like, why are we just hosting these billion dollar enterprises for Zynga? We should just make them make our own games or kick them out. And it's sort of always the push-pull.

1:02:51You want to put your product in a place where all the people are. You want to meet them where they're already hanging out. But then you have to play by somebody else's rules. And you're setting yourself up for a big fall if they decide to rip you off or cut you off. Speaking of playing by other people's rules, if you go work for a company, you are under their regime. You have to follow their protocols and rules. And Zapier at the time was a remote team. Jason was curious about this. How do you keep your team motivated? How do you keep them on track? and how do you ensure that they're actually doing the work?

1:03:19Take a listen. You've got a 100 % distributed team with no central office. You have 36 people working from home. Yep. So you're kind of like that Matt Mullenweg. Yeah, yeah, automatic. How do you know that people are working? How do you manage that issue? People leave a digital trail, right, when they work these days, whether it's Slack logs or GitHub commits or, you know, publications on a blog or what have you. There is a digital trail of did you do this thing or not, right? And so I think that's really what it comes down to. And in a lot of ways, I think it's more efficient because now you're looking at the byproduct of the work rather than necessarily looking out over the room and seeing people in the chair, which isn't work in and of itself.

1:04:04Absolutely. It's a lot more pure. Exactly. And do you actually do that or do you spot check it? How does one manage this distributed workforce like this at scale? Yeah. Because it's one thing to manage 10 people in a chat room, but to manage 36 and then eventually 100, slightly different, right? Yeah, it is a little different. We have a pretty basic way we do it, which is every week people write what we call a Friday update. And so we have an internal blog. You publish, hey, here's what I took care of this week. Here's what I'm going to tackle next week. And here's some fun stuff I'm doing this weekend or whatever, right?

1:04:35Oh, that's cute. Yeah, which is nice for remote. But then whoever the team leads are, they can go through check, make sure that, you know, people on the team are making progress. And not just making progress on the task, but are they making progress in the right direction, right? Do we have one person running this way and one person running that way? Well, if so, maybe we need to reconsider how we operate, right? And so those variety updates are just a really quick, easy way so everyone knows, like, hey, this is the pulse of Zapier. This is kind of what is moving. And you put it on the internal blog.

1:05:05That's brilliant. Yeah, it's all public and transparent on the internal blog, so everyone kind of knows here's what's happening. And you can go read anyone else's if you want, right? So like an engineer can go check out what's happening in marketing, can go check out what's happening. So just like a WordPress blog or something? Yeah, we used to use a WordPress blog called P2, which is the same thing that Automatic uses. We actually have now built our own variant of it because it was so important for us that we were like, we wanted to specialize in a few ways. I love that idea, though. It's just like, and so you get to work from home, but you have to self-report.

1:05:39Exactly, right. So the cost is, like, because some people get offended when you're like, write down a list of everything you did, like, for the week. And it's like, are you micromanaging me? And it's like, no, you get to work from home, dude, or do that. Yeah. It's pretty freaking cool. Exactly. And I don't think it's, you know, write every single micro task down, right? Like, it's more just, hey, give us the broad overview. Like, tell me where to go look. If I want to see the specifics of that commit, just link me to the commit so I can go see, all right, here's the hundreds of lines of code you wrote this week.

1:06:09Here's how it works. Right here, it's tested. All right, a manager can make sure to code review. Not just a manager, a teammate. Like, you know, a lot of stuff. To check in on it, yeah. Yeah, you know, you rely on your teammates that are doing work too. So like, you have to collaborate. You have to be able to check these updates, make sure that teammates have access to what other teammates are doing and can benefit from that shared work. Zapier's sort of ahead of the curve. I mean, we now think of having an entirely remote team as, you know, just another day doing business in the tech industry.

1:06:38But in 2016, this was well before COVID lockdowns. This was well before, you know, quiet quitting or whatever. Like, this was still an era when most companies were expecting most people to show up to work. Yeah, absolutely. But I do love that we're still having the same conversations because we're talking about end of week roundups. What are you working on? People taking a look into like the checklist of people's operations. It's almost like technology doesn't change that much. Like this is the funny thing to me. We're still talking about Slack in this conversation 10 years ago and today. We're still talking about remote work then and now.

1:07:10I'm still a big believer in remote work. I know Jason's kind of gone to the other side of things here, but it seems like it didn't hurt Zapier too much to have a mostly remote team on. No, it really doesn't. And I also thought this is a fascinating look back into history because sort of the idea of self-reporting, like, well, if you're going to work remote, you've got to post multiple times a day, let everybody know what you're doing. You've got to be so much extra communicative to make up for the fact that your boss can't just stop by the office and peer over your shoulder and see what you're doing.

1:07:41And this became such a huge discussion during the sort of Doge era when Elon Musk was asking government employees to do this. And a lot of tech leaders were saying, I definitely want to do a start of day, end of day, end of week type reporting. We do those at launch. It's a big thing that Jason's a big believer in. And they were already sort of discussing this same conversation back in 2016. I really liked Wade's point that if you really think about it, you can tell who's being productive because everybody leaves a digital trail. That's the phrase he used, which I love. It's like, even if they're more efficient, you can't tell how much time somebody is spending necessarily remotely.

1:08:21But you can tell they got these five big tasks done today. And if you're not looking at work that way, well, maybe you should be. I mean, there is a performative aspect to sitting in the office and looking very, very busy and very stressed out, arriving early, staying late, but taking a long lunch so you look more productive. And people play these games, right? And I'm not trying to say that every remote worker out there is an absolute saint who should be given a raise and more responsibility. But I do like that Zapier was ahead of the curve here, and I think it's worked out for them pretty well.

1:08:48I mean, that concept that Wade discusses in this segment of the internal blog. Well, we wouldn't call it a blog today, but the idea that there's a central place. everybody's going, they're posting about what they're doing. You can look at what everybody's doing. And so you could, if you were the boss, you could look at their blog post and say, well, that doesn't sound like a whole day's worth of work. What did you do for the rest of the day? So like, it's not that there's no way to check up, but it's not as rigidly like hour based as I think a lot of employers are where it's like, I need to know that you're very, very busy, sweat dripping from your brow hunched over the keyboard for 10 straight hours every day, or I'm not getting my money's worth.

1:09:29And, you know, it's not good for that. As a very bursty worker myself, I really approve of this model. Next up, Jason and Wade discussed how he forgets that he has the application that he's using it because it is essentially software that works in the background and does stuff for you so that you don't have to do it. Take a listen. One of the things about Zapier I find is I forget that I have it because I set it up to solve a specific task. It's an invisible tool too, right? It is an invisible tool, right? Like I don't have to log in to actually experience it. It's just occurring in the background.

1:09:56And so I've always like wondered with a product like yours, it's almost like I wish there was somebody who would come in and audit our company and teach everybody in our company, like all eight or nine or 10 people in the launch, here's how it works. Here's how to leverage it. Go. Yep. Totally. Yeah. This is something where we actually started experimenting with early this year. Like we have Jess, who's running our concierge program, who is doing this, right? Concierge. Yeah, fancy name, right? I like that idea, though. So yeah, Jess will talk to a person and say, hey, you're using these five or six apps.

1:10:29You know, well, based on what we've seen across our hundreds of thousands of other people are using Zapier, have you thought about setting this up? Would this be valuable? I can set this up for you, right? Wow. So we started experimenting with this. You know, I think some early signs are pretty positive. I think this is pretty common in a lot of industries, right? You have a services component to the company. So we'll see if that's something that ends up taking off or not. But I think a lot of people are in similar bows to you. They've got this tool, right? It's like having a spreadsheet. You have a spreadsheet.

1:11:00Yeah, what can I do? What can I do? How can this help me get my job done? Exactly. So you've got all these tools now laying around, and you just need a little inspiration to figure out what's the next. What do I do with it, right? Now, the reason why I wanted to bring up this segment is that, again, I'm hearing an echo of current trends in the past because they're talking about concierge service and helping and auditing, figuring out what people are using it for, where they're not using it, how to get more out of it. And we've talked on the show lately about forward deployed engineers from major AI companies that go into a business and help them set up, let's say, a new AI model or a set of AI agents, and then help them kind of figure out the process to actually use that technology.

1:11:40And here, 10 years ago, Zapier is doing a similar thing. Now, they were very early in this process. So I think we're doing more of an experiment, whereas today I feel like your Palantir's and your open AIs, etc. are more aggressive about it. But again, people need help. And that has not changed in the last decade. It echoes a lot of what you hear with sort of how AI is deployed and how a lot of AI engineers and researchers are feeling like the current era is very, like, demand-focused. Like, I go to ChatGPT and I type in what I want. I tell Claude what I need an answer for. I ask Gemini to make me an infographic.

1:12:17And one day, the idea is that AI is maybe just going to always be kind of hanging out, listening to us passively in the background. And then it will figure out, Lon needs some help. I'm going to just jump in and write that for him or give him the right answer. We actually, when we first brought producer Claude from Anthropic into This Week in Startups, this was my idea. I was like, can we just have Claude listen? And if he's got background, be like, actually, I could add, you know, like have him alert us that Claude has an insight. We're not quite there yet, but I'm really excited about the day when we can sort of forget that you've got these AI applications on.

1:12:56And then they will just remind you by being helpful. And then they will go away again. Obviously, that means we have to let them listen to us all the time. I know that's not ideal privacy wise, but. I really do think that we're going to have eventually our own personal instances that are so locked down that that won't be an issue as long as you can trust the overall privacy architecture. But you know what? That is not the topic of today, Lon. Let's stay on topic. I'm not going to let myself wander into that forest. Just to wrap up here, Lon, Jason wanted to know Zapier's holy grail. What is at the absolute end of the tunnel?

1:13:24What are they looking forward to building in the future? Take a listen. What are the services that you guys want to add and that you'll add in the future that you think will be like the holy grail and really sort of that are frustrating that you can't add them today, but that you could? Like there's a wish list of things like why doesn't Google allow us to put search in here? Well, I think Google search is a closed ecosystem, right? There's no API for it. I don't know. I think it. Yeah, I think it is closed. I mean, the one that we're waiting for bated breath with is Microsoft's Office suite, right?

1:13:57Like traditionally, Excel and Outlook and Word and all that stuff have been kind of a closed ecosystem. So I guess we'll see what happens. You know, I'm cautiously optimistic that this will happen. With Office 365. They're definitely moving in that direction, right? They've had some announcements around APIs coming soon sort of thing. So we're pretty excited about that because the world still, like as much as, you know, we like to talk about new tools and, you know, we've talked about all the cool tools we've used today, the world still runs on email and spreadsheets. Like most businesses are using that still.

1:14:31And so just, you know, getting access to the world's most popular email and spreadsheet tool is pretty exciting, even if it is kind of boring. It's really an amazing product. I highly recommend people go take a look at it. It's super cheap. five plans and they range from free to 125 bucks a month. And then have you gotten into any huge enterprises yet? Or do they just, you let like the groups just put them on their corporate cards? Yeah, pretty much. Um, you know, people will hit that one 25 limit every now and then and ask us for higher plans and we'll, we oblige willingly of course, but we haven't, you know, it's still all self-serve credit cards.

1:15:10Um, we don't have a sales team or anything like that today. I love when people get what their ultimate dream was, because he says in this clip, that the white whale, the ultimate integration for Zapier would be Microsoft's Office Suite. It's this closed ecosystem. So many people are on it. It's what runs the business world. Everybody's on Office. And it was like the far off distant fantasy. And if you go onto the internet today, Microsoft Office 365 completely integrated with Zapier. Wade made his dream come true and how proud of him we all are. And we make fun of people that want to manifest their success.

1:15:48Well, here's a guy going on a little podcast back in the day, and he brought Microsoft, now a multi-trillion dollar company to heal, to open up the darn doors. Shout out to Zapier, man. I mean, what an amazing company. I think it's going to be one of these IPOs that when it does happen, is going to really impress a lot of people. Now, there's not a lot of venture pressure to get them out the door because they've only raised a handful of millions of dollars, but I mean, they're going to want to list eventually, Lon, right? We're not going to be left in the dark for another 10 years wondering how much money they make.

1:16:14It's just become such an essential tool. I mean, that's why I was so surprised that it had such a long shelf life before we all sort of became aware of it. Because today, it seems unthinkable that we would run our entire system without Zapier. It's a fundamental building block of everything that launch does. And so that was really surprising. And back when Wade came on This Week in Startups, he was talking about, you know, his dream one day of Microsoft Office or whatever. They didn't even have a sales team. They weren't even focused on enterprise sales at all. They were selling Zapier to individuals with credit cards.

1:16:49And so to see the growth from then is unbelievable. Speaking of growth, this entire segment was shot in what appears to be a WeWork. Like there's just people walking by. And Jason's talking about launch having like eight people or something. Jason has a great love of co-working spaces. That's where we tend to find ourselves there. It's almost like he's an extrovert. Coming to you from the Capitol factory here in Austin, Texas. I'm coming to you from my home office of Introvert Castle. All right, everybody. This has been This Week in Startups. Lon, you're the best. Thank you for finding these amazing clips.

1:17:18More flashbacks coming your way.