The Vertical AI Wedge Hiding in 25-Year-Old Software | Birju Kadakia, CEO @ Rec Technologies

77,000 people going on the system at once. All of a sudden, this is like when I was working at Uber and it's like New Year's Eve, demand's spiking in that way. So you have to build a technical system that can handle it. When I was growing up, I lived in Chicago. Chicago Cubs tickets would go on sale on a certain day, and I would go wait in line for like my chance to buy tickets for the whole season. And then you'd see people there paying people to wait in line. And so if you think about that analogy, AI should help there. Why can't you just set up your agent to wait in line for you? $400 billion are spent in the US on local activity, and that's kind of the economy that we're building into. This is my Rec ID. It's on my wallet. Any city that I work with, I can use my Rec ID to just scan in, and I can do anything at that location. The analogy I use is a recreation passport. It feels superficial. It's just I'm pulling out my ID, but under the hood, it knows that I am a tennis player, I'm an amateur golfer, I have my medical information on it, and so it knows who my emergency contacts are. It knows my allergies.

The example that I've seen or I've heard What's the single most important word as people start looking at token usage?

It feels like a false positive. I think we're going to look and laugh at ourselves one day. In the early days, it's all about iteration velocity and failing fast. The single word is speed because time is your most precious commodity. You can raise all the money in the world, you can hire all the talent in the world, but time is finite and it goes by fast. And so every time I look back at something that we did right or wrong, the first takeaway I typically have is I wish I learned that faster.

CEO of Rack. They are the company disrupting 25 years of software incumbents in recreation tech, leveraging AI. Today he's going to teach us how he did it. Welcome, Birju.

I'll help share how we're getting started doing it.

Okay. Okay.

Yeah.

I shared information from my research about you and your company. Is there a non-LinkedIn version of you that you would like to introduce yourself as?

I'm Birju. I started Rec Technologies almost 4 years ago now. I'm from Chicago, uh, grew up in the suburbs and was always a community recreation user without even knowing it, right? I used my local tennis courts, I used my local, uh, basketball in the rec center. My mom probably used me for childcare without me knowing it. And so that's kind of hidden away in my background. After that, I be— you know, I, I'm on the classic tech journey. So I studied engineering in Illinois, I moved to San Francisco, I was a really early employee at Uber, helped launch a number of the product teams there from Uber Eats to taking Uber Eats all over the world. And then I went and led product at The Athletic, the sports media company. The through line all the way through has always been, I love products that touch the real world. And I want to work in a space as a product leader where I'm talking to someone I don't work with, I'm talking to you or I'm talking to my dad and they understand what I do because it exists at the intersection of digital and physical. And that was why I joined Uber. I was like, oh my God, it's just so painful to get around in Chicago in the winter. And it's why I, you know, Joined internally the Uber Eats team because again, like food delivery, like my favorite restaurants weren't delivering food, and it was such it felt like such a simple problem. And fast forward all the way to Rec in 2022, myself and my co-founder Rachel were just excited to work on a new problem on the backs of COVID where we said, "Hey, like I'm trying my hardest to become more active and get back in the real world, and I want to play." I grew up a tennis player, so. play more tennis. I'm looking to build a new hobby. I'm looking to meet new people. And every time I tried, technology was not helping me. And technology was not helping me get a ride faster. It was helping me get, uh, you know, to the, the, the gym, the restaurants, uh, to get, uh, all, all of these like simple things. But for some reason it wouldn't help me get more active. And so I latched onto that problem and I got obsessed with it and I went really deep on understanding— we call it The activity economy here at Rec. And the activity economy is actually one of the biggest industries in the country. It's how you get active. It's how parents keep their kids active. It's the childcare. It's the swim lessons, the summer camps. It's all of the sports that we all play. It's how we stay healthy and fit. It's where we all want to spend money but are looking for ways to do it. And so look, that was a long way to say, like, I like working in things that make our lives a little bit better every day. And I think Here at Rec, we're embracing AI to figure out, hey, how do we just get people more active?

You're one of those founders who are building things for yourself first, because this is relevant to your day-to-day.

Totally. I can't get excited about a problem unless I'm going to be able to live it.

Some founders are solving really hard to understand problems. On the other side, I feel founders are solving problems like yours are very relevant and very understandable. It's a choice, right?

Yeah, I think, I think it's, it's two sides of the same coin to me. So you've got like a number of people that are scientists at heart, you know, and I really respect the folks that say, hey, this is a really hard problem and the hard problem energizes me and I'm going to solve it. And then it's going to have, I know that solving this hard problem will have impact to the world. And then you have folks, I probably fall on this side of the coin where I'm a product person first and I say, Hey, this thing is frustrating me so much. And what's really important to me is, let's keep it really simple. I can't figure out where to play tennis. The easy thing to do would be to say, let me try to build like a search engine that can find all the tennis courts in the area. But what I think you have to do, especially when you start a company like this, is be the type of person that sees that hard problem and then digs in and says, how does this world work? And how does this industry work? And how does How do municipal recreation departments work with the private clubs and who makes the scheduling decisions? And I think that's where all the founders are the same is like whether your starting point is I'm a scientist or I'm a product person, underlying that is you're willing to get in the weeds and the details to figure out what the problem actually is and solves that.

At some point you meet both worlds because you're solving a problem that consumers are facing, but deep down when you're actually trying to solve it, you meet with the scientific side of things. It's not just tennis, right? It's, it's more than that.

Way more than that. So what else? Yeah, the way I would think about it is the activity economy is how any one of us or anyone in our families do the things we love, our regular activity in our own community. And so every phase of your life, there's something different. When you're a child, your parents are putting you into summer camps and you're starting to start playing sports and you're building skills, you're building life skills like swimming. And so the— there's an entire economy, the youth sports economy, the aquatics economy that are around to get kids more active, right? Summer camp, the aftercare. Then we become young professionals and it's about just like filling our days with wellness and activity. And so you can, you know, the golf courses, the tennis centers, the gyms, the yoga studios. And then as you retire, a lot of your community is built around your hobbies. And so the aging population deeply relies on this economy, whether it's at the senior living centers again, or the gyms or the parks and recreation departments that are putting on all these activities. And so maybe to like contextualize it, $400 billion are spent in the US on local activity. And that's kind of the economy that we're building into.

What is really the problem you're solving? Can you describe it?

It's a 3-sided economy, right? You've got restaurants, you've got me and you who are saying, hey, I want a burrito today. And then you've got delivery people. And recreation, very similar. At the base, you have space owners. And so you've got your, the city of San Francisco, which owns spaces all over the city, or, you know, cities, municipalities, you've got the YMCAs, the JCCs, basically anyone, all, there's hundreds of private sports complexes that are built every day, right? And so you've got that group, and then you've got all of these operators. You've got people that run leagues, you've got parents that are coaching and volunteers that are, that are reffing, uh, lifeguards. And then you have all of us trying to be active, right? You've got that same, you've got that same intersection, but in between that, to make it all work, it's all pen and paper right now. And so if you think about—

I can't believe it's pen and paper.

Yeah. And, you know, and there's like a little bit of software. So it's like when, if I sign up for a swim lesson, I'm signing up online. You've probably done this before. It's—

I just set up my kids' summer camp, so I know what you're talking about.

Yeah. So it feels old and stressful, I think, is like what we saw. And so there was some like simple pain, but we said, hey, like, why, how would we change the system? And when you learn what's going on at the summer camp operator, they're doing so many things. They're saying, okay, we need to recruit a volunteer to run this camp or a part-time employee to run this. this camp. And so, they're out posting things online, then say we have to market this thing. So, they're taking huge catalogs and turning them into activity guides. And then they're saying, okay, well, we need to take attendance every day. They're doing it on pen and paper, and then they're data entering it into the system. And so, you've got all of these really just crazy manual processes that bog everyone down.

We receive a Mountain View catalog for the summer camps. It's a booklet that we receive in the mail every year, and then we have to manually look and search for which ones to sign my kids up on. And then we have to plan it because classes will run out quickly, and then kind of have to have a strategy. So it's a very painful thing to do.

Yeah, so there's 3 problems in what you described. Okay, the, the, the mountain— city of Mountain View, right? Someone on that team is spending weeks putting together this activity catalog that then gets printed. Problem 2, it's a paper catalog. And so you and definitely folks younger than us don't want to flip through a paper catalog, maybe don't even check the mail.

Right, exactly. Throw it away.

And then problem 3 is the stress. It's like, okay, now I got to sign my kids up at 8:00 AM on this day. And I don't know if the system's going to work. I don't know if I'm ready. I don't know if I'm planned. I don't know what the right sections are. If we just step back, we said, hey, this whole activity economy has so much room for innovation. And this time that we're at is a really exciting time because we actually think with the tools we have today, we can really innovate and bring something totally new and different to the ecosystem.

So I've been doing this for the past 2 years where flipping through the catalog, it never occurred to me that this is a solvable problem. You know, as a consumer, or I'm one of those Maybe 99% of the consumer who just suffers through it. I'm like, okay, I accept this as it. Was there a moment that you decided that this is the thing that you want to solve?

It's a really good question. Maybe I'll ask you a question. Why do you think you accepted this pain versus like you wouldn't accept this pain to pick a restaurant?

Very good question. I wonder if I'm, uh, I didn't suffer through it enough. It's once per summer. If I actively, you know, I'm scheduling classes. and the scheduling system is like this, then maybe I'll think about it. But I— but if we don't have something like that, yeah, everything's manual. Like, on the other extracurricular activities are manual signups. But—

You have to identify what side has the most pain. And I think for the consumer, what we believe is we can provide a magical experience for you. It doesn't need to be branded as AI, uh, because, you know, you don't care if it's powered by AI.

Yeah.

But it will just get really great and then you will just be happier and you'll be happier with your city and your community because you can be more active and then you'll probably spend more money. And so to me, the situation for you is an outcome of the work we're doing, the improvements for you. We decided that where we needed to start working is with people like the city of Mountain View. The reason for that is the space owners and the operators, they're the ones that are incredibly bogged down by work, right? To get the camp out and available to you and then to operate the camp and then to do the financial reporting on the camp and to get funding for the camp. There's enormous amounts of work there. And that's why our business has focused on recreation professionals, people that are putting on recreation. And through them, we're making the consumer experience a lot better because we realized that that's where the pain was.

When do you think I'm probably not going to receive any of those paper catalogs anymore? Like by which year?

Yeah. Yeah, it's a good question. I can't tell you when they're going to stop doing the paper thing, but maybe what would be better is we know what you signed up for your kids last summer. We know how old they are. We know what they're interested in. Why can't we just send you— well, why can't an agent just ping you and say, hey, we assessed the Mountain View summer camp plans. Here's what we think your son or daughter should sign up for. Would you like to accept this itinerary and would you like us to try to book it for you on booking day?

I love that. Can you do that?

Yeah, it's in the works. But, you know, what we want you to feel is less stress, more confidence, and like making this go away. But what cities want is they also want predictability in demand. They want to reach more people. And so the outcome for you is just You know, so much easier, but for them, it enables them to kind of figure out what classes to put on, to figure out where, what, what's going to be more popular and then get people in and simplify the operations.

I'm in a Mountain View neighborhood community and also parents community. When you have that ready, give me a shout out. I'll share a post so everybody knows this exists and we don't have to like do the paper.

Please do. I think this is what's exciting to us is like, When we go live in a community, yeah, people kind of like lost hope for something better. And so—

Right, right. No hope.

You know, we see tweets, people are like, we have a feature called Fast Track, which lets you kind of plan your bookings. And people are like, oh my God, I love Fast Track. And that's what we want to do.

There's another thing that I've seen. Every school year, we have this calendar from schools. It's literally an email with a list. I have to book myself, invite my husband as a calendar invite. One by one. So I did it by building a little script to do it more automatically.

Most people can't write a script. And so, you know, it is really challenging when you think about, you know, a lot of our cities are not in the Bay Area, they're all over the country. And we find that the access and equity to activity and childcare and all of these things is like a critical thing. And that's why people come work at Rec, right? Because They're passionate about this mission that we're on. But it's been really exciting to think like the types of solutions we can provide. So once we have the catalog in our system, so, you know, we have, we know the calendar, we know when all the events are happening, we know when everything's happening. Step 1 of building Rec was to get that data and put that data somewhere that AI can actually access. Because now that we have that data in a format and accessible, to agents that we will build or agents will let other people build on top of the tool or allow you to discover things on ChatGPT or Claude or wherever you're discovering things these days, the world becomes more accessible. And so, I think, yeah, step one is really about taking that calendar and putting it somewhere accessible.

Yeah, I like that tagline, making the world more accessible, making your own community more accessible to you, right? I have a use case I want to run by you.

Sure.

I tried to book the Ghibli Museum in Tokyo. The tickets drop at 10:00 AM sharp on the 10th of every month. So one day at one hour per month, I logged in early, 10:00 hits, and I hit refresh, and I was number 77K in line. Like, the site is like telling me you have to wait for an hour. I waited a full hour, and then when the page finally loads, it says it's sold out. I mean, when these Demand crushes a system like this, right? Do you have a fix?

So, so we work on this. We think about this problem every day. Like, look, as you're describing it, I'm like, oh, this is like, this is what we— every community we go live in faces this problem. And look, the, the first thing I'll say is like, part of this problem is the lack of supply, right? It's like what you were trying to book is incredibly popular. And so I feel I have built a lot of empathy for our customers. Our like parks and rec departments, our recreation operators, where they're trying their best to put on more activity, more summer camps, but there's just not enough. And so they're selling out. So like, I want to start there. It's like, there's, there's part of our solutions are creating more supply. How can we help them create more supply? Because now you have to give everyone a chance to do this with equity, right? Because like, I, I bet what you felt in that moment was this is really unfair.

Yeah, it's not fair. I wonder, like, how come I hit that button on the dot? How come I'm, like, already thousands of positions out?

The framework I would use is, like, you know, baseline is reliability. You have 77,000 people going on the system at once. All of a sudden, this is like when I was working at Uber and it's, like, New Year's Eve, you know, like, you have demand spiking in that way. So you have to build a technical system that can handle it. We can tell you, hey, this is a percentage chance you have of getting this right now. And then here are your options and we can give you alternatives and we can smooth demand out. And then we go into the booking experience and like, why are we making you wait in line for an hour?

Yeah, exactly.

It's funny, you know, like when I was growing up and like the, I always lived in Chicago, Chicago Cubs tickets would go on sale on a certain day and I would go wait in line for like my chance to buy tickets for the whole season. And then you'd see people there paying people to wait in line. Right. And then you have people and, and so if you think about that analogy, AI should help there where it's like, why can't you just set up your agent to wait in line for you in order to give you and everyone else a fair shot? Right? Like, what if everyone had an agent? Now there's like other, you know, that requires us to think about tech technology equity and things like that, but there is a path here. And so we're actually partnering with some of like the largest operators of scarce resources right now. Like the, I was just in, in a conversation with the city of Chicago talking about their challenges getting people into summer camps or swim lessons and just concepting, right? Just brainstorming with them on how can, how can we step by self-solve a problem like this?

So, in your mind, it's solvable?

It is definitely improvable, you know? And, and so, I would say the— there's a customer experience element, which you just described. We can help solve that. We make it way better. And then there's a supply creation problem, and we can help solve that too, right? Where it's just like, if there's one day of those tickets that were less in demand than the other day, and you knew in advance, like, what could you do to shift supply around?

That was just thinking, are you— and you're a global company, right? So one day you may be working with Japan, and then you can handle those tickets over there.

We're in the US right now. We will definitely be a global company. The way I see it is The whole world is trying to be more active. I was just talking to someone in India who said they're having thousands of pickleball centers all over the country being built. And so all of those things need software and they want software that can actually like meet the level of quality of the systems being built. And that's where it gets really exciting.

Other than this registration use case, do you have any other typical use cases that you have seen?

Yeah. Yes. That's a good call. So, so we think of our technology stack as everything that helps you get active. So. Signing up for activities is one big body. There's another body that's just like reserving spaces, right? So we work, like I said, we work at tennis, the San Francisco Parks and Rec Department on reserving tennis courts, but swim lanes and picnic tables and pavilions and all that stuff. Then there's a whole area around memberships. Like I'm a member to my gym, I'm a member to the local racquet club, and how do I show up? How do I scan in, et cetera? Then learning, how do I, how do I take lessons and coaching and things like that? So And that can be on music. It doesn't even have to be in sports. And so that's the way we think about the space that we operate in. It's everything that you'd want to do. And then we have software that supports the businesses. So keeping, keeping on the software analogy, like Toast is probably who you think of in the restaurant space. It's providing all the software that runs the restaurant. We're providing all the software that runs all the recreation ecosystem.

As a consumer, My understanding is I can also sign up Rec and then I'll have access to these, all the activities available out there. Is that what it is? It's like a Yelp?

Yeah, it's growing in that way, right? So we're, we often start, so in San Francisco, we work with the city of San Francisco. We also work with the city of Piedmont and we also work with the city of Emeryville. So over time in the Bay Area, we're becoming not just a a piece of software that helps all the cities operate and serve their residents better, but we're also becoming a resource for the community to find what they want to do. But I like to think of it as like we are building a platform for recreation professionals to really deliver their best service to their communities and constituents.

Because earlier you mentioned your initial thought was building something for consumers who can just search for activities. That search function, and then you turn this B2B and targeting at a recreation professional angle. Now you're, you're also potentially doing the two-sided market as well, eventually branching out to consumer as well.

Yeah, though, look, the, this ecosystem has, has had a severe lack of technology and companies like us that have, you know, great engineers that are building all these awesome tools. And so it's exciting to be in a space where there's so much white space. If I was, you know, doing the hard thing as a founder of prioritizing, I think the most important thing we can do is help businesses connect with their customers better. And so this can be everyone from the city of San Francisco to the local YMCA to a small pickleball and tennis club. They all have the same problems. And so we're helping all of them make their program catalogs more accessible, operate more efficiently. Efficiently. And so that's been kind of like problem and focus number one. The really awesome thing here is like, we're, we're gonna be going live with the city of Indianapolis in the fall here. 800,000 residents that are all going to have a Rec profile that enables them to take advantage of everything their city offers. We were talking to their team yesterday and they said everyone sees Indy Parks and Recreation as the city's best kept secret. And we said, We don't want to be a secret anymore, right? And that's the, that's like the role that Rec plays in that community. But what we will do is, as hundreds of thousands of city residents come online, we are going to help them get more active, right? And then so we will, we will spin up products that say, okay, we now, now we have 100,000 parents that have children on the platform. How can we help them? And Rec becomes a platform to enable more activity. Not necessarily just like quiet B2B SaaS software in the background.

Is that the first city that you're working with whose, uh, all the residents are getting, uh, this REC profile?

No, we have over 100 cities live.

Ah, okay.

Growing really fast. And so there, I, you know, uh, I like to use kind of like the iconic big cities as the examples.

What does it mean when you say go live?

So when the city of San Francisco went live with us, what happened was So they, they use us for tennis and pickleball court reservations specifically. And so if you walk around the city, go to Dolores Park and you'll see the tennis courts, you'll see a sign over there and it reminds you of a sign you might see at a restaurant wall that says like, book on OpenTable. And so it says Dolores Park Tennis powered by Rec. And now you can scan that QR code and you see all the resources available there. You can see when it's available for drop-in, you can see when it's bookable, you can see when you can take a lesson on that court with a certified coach. And so we're basically turning that kind of like passive tennis court into like a digitized resource for the community that enables people to use it more. And so that's the San Francisco example. But in Indianapolis and many, many, the majority of our cities, the gyms that they operate, all of their activities, all of their courts and their picnic tables, everything becomes powered by Rec. And so Rec becomes like the Shopify for the city for all their recreation resources.

If I walked in a gym and I want to sign up, they're going to probably ask me to download Rec.

Yeah. So like, here, I'll pull it up. So this is my Rec ID. It's on my wallet. And so I can, any city that I work with, I can use my Rec ID to just scan in and I can do anything at that location. I can like scan, if I have a membership, my gym membership will let me in. If they, if I want to sign up my child for something, I can scan and they, you know, if I'm in person, I can say, Hey, like I'm checking in my, my son today and it'll just buzz and check in the son because it knows what you're, what you're there to do. And so it's like a, it's a piece of digital infrastructure that makes everything way easier.

Wow. That's so interesting. It's like an ID in a sense, in a sense, like your activity ID everywhere.

Analogy I use is a recreation passport. And the reason I say that is like, it's, you know, it feels superficial. It's like, hey, like I, it's my, just, I'm pulling up my ID, but under the hood. You have all of your preferences. It knows that I am a tennis player. I'm an amateur golfer. I, you know, like to play in soccer leagues. It also knows that I might— I have my medical information on it. And so it knows who my emergency contacts are. It knows my allergies if I'm, you know, someone that's in a summer camp and things like that. And so it enables you to actually participate. And you've probably gone to a gym and filled out like a health waiver and things like that. We digitize all of that.

I remember seeing a use case online that's called refund. Do you also handle refunds?

Yeah.

Is that one of your most common AI agents?

Do you want me to show you a couple? So this is Pawnee, Indiana, right? And so it's a nod to Parks and Recreation if you've ever watched it. What you'll see here is this is the website that the city gets, right? And it looks and feels like especially on your phone, like all the other apps that you're used to, but you can see all these reservable sites, you can see programs you can sign up for, you can see events that you might be running. I can get coaching from certain coaches, I can book a bunch of things. And so the users get a simple, easy-to-use front end. We're not trying to make it too complicated. It's just get in, get out, get what you want. Now on the administrator side, I'm going to go into the The city of Palm Beach. Actually, I probably have it. Sorry, one second.

So does it mean that if I want to book different activities for, from different companies, different vendors, then I don't need to log into each one of them? I can just go to the centralized place. Yeah.

Any, and anyone that we work with. And over time, we'll, we'll kind of expand that coverage.

That's the idea, right?

But the more, like, I think like the more powerful part of this is think about, so this is the city of Pawnee. They're, they're probably managing a $20 million budget, right? Uh, and, uh, it's a huge business, right? These aren't small businesses. 50% of the country's recreational spaces are operated by municipalities. Uh, and so here they're managing Their whole business. They're managing what might be happening across all of their sites. They're managing any activities that they make available. They're managing any events they might be running. So if there's like a concert series, they're seeing people sign up. They've got memberships that they're using and people are checking in for things. So it's a very, people are showing up, to on location for things or checking in. And so we're basically like, maybe the best analogy is a hospital has an EHR that, that allows the hospital to operate. We are essentially the EHR for a recreation operator. And so there's a lot of work going on. And so I'll just pull up this automation I turned on. And so you mentioned the refund agent, and we can talk a bit about like how we think about these agents. We We're building agents that can essentially do a lot of the work that exists onsite. And so the refund agent is an example of, we found that when we talked to people and we're like, hey, why don't, why is your job so hard for you? And they're like, hey, look, if it rains in a city, all of a sudden all of the swimming pool classes are, are, are canceled. All the tennis court reservations are canceled. We have to communicate to users that, uh, you know, there's like potentially lightning. And so we're going to cancel other things. So there's all these tasks that kind of come up ad hoc. And these are the types of tasks that AI is great at. And the reason it's not easily automatable without AI is every class has its own policy. So you've got summer camps that have one policy and you have tennis court reservations that have another policy. And so it just becomes incredibly complex to build a system that could automate all the edge cases. And so what our system does is we read the policies one time and we read the policies one time. And then what AI is great at is writing a slew of deterministic code that basically codifies these policies. So what happened here is in the background, we had AI read our policies and then Set up what the policy should be in plain language so that our operators could understand what was going on. And then behind the scenes, it codifies that policy. And then maybe what's the most important is you have to make it feel very trustworthy for a, for a city to use AI, right? And so this is basically an audit log of the agent, and this is our test sites. There's not that much going on here, but. Taking a look at every potential section that was canceled and whether or not it should be refunded. And we make it really easy to kind of trace Rec as a human that's working on behalf of your city.

When an incident happened, like lightning or weather conditions changes and classes need to be canceled, is Rec initiating the call or making decisions? And then sending that information to the administrator and they're approving it. Is that how it works?

The way I'll describe this is there are workflows in Wrecked, right? And so, um, we go to cities, we learn their workflow, and then I'll, I'll like use some AI terminology here. But so we're finding that workflows begin and end on all sorts of different triggers. And so you've got a workflow that says like, hey, if it's rainy outside, Communicate to users that it might be canceled. And then you can imagine it's like, hey, if the weather API is saying that it's rainy, then cancel, cancel this set of things and automatically give them a refund. And so what Rec is building towards is an end-to-end workflow builder that can take any kind of experience and automate the work end-to-end. Now to get there. You kind of have to do this in a few steps. And so the step before that is we're building a bunch of skills and those skills allow you to codify, or whether or not it's codified, to operate any part of that experience. And so you can think about it as refunding a user is one skill in this broader workflow. Deciding if it's rainy enough is another skill in this workflow. And so we're the, where Rec comes into this is why can't you just do this on Claude, right? It's like where Rec is great is we understand how a recreation operator works. And then based on how it should work in recreation, we train the system to do things in the way a recreation operator would. And so right now what we're doing is we're getting all of these individual skills really sharp.

Yeah.

And then what we will do is we have a team that's working on building out the threaded workflow. So then you can kind of tie the workflow together in any way you want.

I wanted to ask you, what if I just pass these policies of each locations into Claude and would Claude be able to give me the decision that's correct most of the time?

Our belief is like, we should build things into the mediums that people will use them. And the thing that we have is a deep understanding of the way recreation should work and the data and the access to all the information that allows us to be very useful, right? We know the schedule data, we know the customer's data, we have access to their payment information so we can actually process the refund and all of that.

So, which means that, uh, if I'm an administrator, I can access my data, but accessing to RecMCP means it brings the recreation taste So it's not just about the data. It's also about the AI that's capable of understanding the nuances in this specific industry, right? You kind of have to have the taste to understand the true decision boundary in this industry. Exactly. Yeah.

I think you see this with a lot of vertical AI. If you look at like a Harvey or a Legora, it's like, what are they building, right? They're building workflows and an understanding of how law works and compliance and things like that. And you see the exact same thing with us, right? Even you could use Claude, but if you want to perform a refund all the way through, you're much better off using our agents that are built with sturdy workflows to manage your finances, right? You don't want to be doing that in Claude. And so that's the way we think about some of this stuff. It's taste. It's an understanding of how customers want to be responded to and things like that.

How do you build this skill or this taste?

It's a learning curve because Look, the models keep getting better and better. And so that's good. That's great for us. And so you have to really understand what you need to teach it. And, you know, what you saw a brief snippet of was a pretty messy LangFuse file. But what we're doing is it's a combination of classic, like, system prompt engineering, but building, going from there to building, like, really structured skills files. That allow us to say like, okay, what skills do you need to be— we'll just keep using activity registration because we've been talking about it— like a camp operator. It's like you need to have the ability to check in people. You need to be the ability to like assess for any allergies in your class. And these are all like unique things that we teach the model how to do. And then we own that, right? And so, and then we iterate because It's, these things require a lot of iteration to see how it interprets all this information and then gets it right.

How do you acquire this knowledge, you know, in the beginning?

Where I think vertical AI companies will truly succeed are when the founders and the teams are deeply embedded in that ecosystem, because it's going to be really hard for someone like Claude or OpenAI to train on how the work happens and build the workflows around it to have the work happen.

They're not going to go knock door to door to each city and ask, can I help you take care of your recreation process? Probably not, right?

Probably not. And, you know, OpenAI launched a legal team, right? And so they're trying to figure out how do we, how do we do law? And I think that's good for the industry. It's like their model is going to advance faster. than a model that Harvey is making by themselves. So I think you can expect that OpenAI will become a medical assistant and a legal assistant, potentially a recreation assistant one day. But what Rec is, is also the operating system for all of the data and the consumer interaction and the coach interactions that need to exist for AI to actually operate in a useful way.

Vertical SaaS is dead. For instance, You know, I'm going to name them ClickUp. It's on their landing page, so I'm not making this up. It literally says software to replace all softwares. What can you share your POV on this?

I think some software is dead. Like, I think, I think it will definitely happen. Like there's no reason to buy some, like a basic website creator, you know, it's like there's so many resources to make a website.

So which ones are going to survive?

You have to be a platform. We're a platform that connects different entities and we coordinate between them. So if it rains, we're reaching out to the coaches saying like, hey, we need to reschedule this class. When are you available? And we have all of their availability data. And then we're going out to the resident and saying, okay, like, when are you available? Can we process a refund or can we reschedule you? And then we have the calendaring information. So that platform and being able to coordinate across a platform is, and have all the data to do so, is really important, right? And so I think you have to be an owner of whatever the valuable data in your field is. So that's maybe one is for us, that valuable data is schedule data, it's customer preference data. And it's kind of like this compliance, healthcare, all the important data that's pretty private to people. 2, payments. Like I think transactions and money movements, you saw OpenAI kind of Back out of commerce. There's just really high, high risk, high important things, right? Like you're getting your kid into summer camp, the trust factor is really high. Um, you're spending $3,000 for the summer for them. And so you need a platform that's designed around that. And then I think the third thing is infrastructure, uh, and equity and access for us where what you, the way you just described it, like a community has, you know, they have 3,000 spots available to swim lessons and Because of that, there's layers of things that are happening. There's personalized pricing, there's the financial aid group that, you know, has early access to things at a lower cost. And to be able to build this personalized engine, like, that's the type of stuff that you have to be able to do to be irreplaceable. It's like we've built something that's very unique to recreation. And it's a, you know, it's a platform that I guess like a really enterprising city could vibe code, but it would be more expensive than And working with Rec.

How did you get the first dataset? Like in terms of user preference, registration data, you kind of need that to start, is it?

Yeah. I mean, we had to solve really micro problems. This is like, I think maybe my tip for any founder. It's like, pick a small problem to solve before you solve every problem. And so the first problem we solved was, it was 2022, pickleball was blowing up and everyone needed better software for their pickleball courts. We solved that, went live with 10 cities, and Now all of a sudden we had schedule data in our system and consumer data. And we said, hey, customers, look at how much more stuff we can do. And then we earned the right to grow from there.

Um, you know, since you're using AI agents and large language models, I'm sure you're battling hallucination. Do you have a way to address the problem? Do you have a way to measure when the agents go wrong?

Yeah. Uh, I mean, there's, there's, it's a layered problem. We, we kind of operate on like a, you can't trust inference in the moment of inference. And so some of this is just building trust as a layer around it, right? The human in the loop, classic concepts. It's before we allow you to turn, like before you turn on automated refunds, you can turn on refund suggestions. And so what a refund suggestion is, is it, it won't actually automatically refund them. It'll show you what it would have done. And it says, You know, here are like the last 10 cancellations. Here's what I would do and here's why. And then the administrator can say, okay, I trust this thing. I'm going to turn it on. So it's modeled after the way you would onboard an employee. Right. And so that I think makes it human because for us, it's really important to enable these things in a way that people will use them practically in the real world. So I think that's number one. And then Number, like maybe number 2 on this stuff is just, or on the other extreme is there are things where we shouldn't use inference or enable hallucination every time. And that's why with the refund agent, what we're doing is we're codifying policy, right? We're reading policy. We're using an LLM to write code, which is now we can serve any city's refund policy to any level of complexity, but we don't, we don't actually infer it every time. And that allows us to just avoid hallucination altogether. And then in the middle, you have everything else. You have like customer communications. Like we have agents that will help you find the right picnic table that you want to rent. And we are evaluating, we're doing evals on conversations, right? And there's not like an easy way to do that yet.

Right.

And so we, you saw we used LangFuse extensively, but even stepping back from that, we have people on our team that are like, okay, we're going to take a look at 50 conversations and we're going to score this thing. And that's the reality of where we're at, right? That everything is changing so fast. Some of it is just scrappy.

Do you have an estimated accuracy level that your agents are achieving?

Depends on the task. When we do the deterministic stuff, 100%. When we do the refund suggestions, I think we were, depending on like the tuning, we were between 60% and 80%, which was really great. And then we kind of stepped into the full deterministic and then, yeah, it's a spectrum. You know, we're trying bolder and bolder things. Like when we try to recommend you the schedules for your 2 kids that have very different preferences, right? I think, you know, we might be at 40 or 50%, but that's still a starting point.

It's not high stake, but if it's a refund, then it's probably a little bit more high stake for the business.

Exactly. So my, my like reference point on this is we have to be meaningfully more useful than what you would expect in a world without AI. And so if you work with a travel agent, you just want a really good starting point the first time the travel agent sends you back your thing, right? A proposed itinerary. You don't want them to send you to an island when you want to go to a mountain.

Yeah, exactly. You can't be like that.

Yeah, yeah. And so I think that's what we want to do is get to a point where we are Meeting, we're way exceeding the human expectation for this stuff.

When you were getting your first 10 customers, how did they happen? Did you literally cold call them? How did it happen?

This is just the, yeah, the grind of starting a company. You cold call 100, you talk to 5, and you get in. And so some of it is just that. There's no, there's no glamorous way to get your first customer as a startup. And then that's getting from like 0 to 5, 0 to 3. The next 5 to 10 sign up based on the references of your first customers. And so the thing that I really believe is like, you have to do really well and partner with your first customers and make them believers in you because they're going to be the thing that, that drives you forward.

How long did it take you to get your first 10 customers?

You know, it, it's interesting. It's like our first customer, we didn't even know that we were going to work with parks and rec departments. We were building software on some experimental spaces. They reached out to us. They said, hey, we saw that piece of software, we want to use it. That's a really exciting feeling because the market is pulling for your product. And so we got our first customer that way. And then the next 5, we were like, okay, like, are we doing this thing? Are we going to call a bunch of people? Are they interested? What we heard right away is everyone was interested. And it was really about us figuring out how to go to market, who to call, how to, how to Find the right people. And once we figured that out, we really started rolling. And so, I think it was maybe like 3 months to get to 10. And then it's really exciting when you can go, you know, we went the next year, we were pushing 100. And so, it's really fun.

Did you have to learn to handle rejection in this business? How do you handle rejection?

I think it's the same way you do in life. It's, uh, Especially as a founder, you need to believe in what your product market fit is. And what you should never do is not listen to your customer. And if they're telling you that this thing sucks, or like, it's not what we need, don't force it down their throat. But there will be lots of other reasons for rejection. It's the wrong customer at the wrong time. And so I've learned to take every rejection and learn from it. Sometimes it's, hey, we're a little bit off on this product, and that's feedback we need to meaningfully put to the loop. Sometimes it's we need to change our pitch or we need to change the way we're— who we're reaching out to or something like that. And you just have to like really accept it.

I want to ask you about your team for a minute.

Yeah.

How big is your team?

We are about 35 people right now.

Has AI changed how you think about how many people you need and who you hire?

Yes and yes. Everyone is way more productive, right? And so I think this is the answer you're gonna hear across the board. So, Yep. We can do more with less, and that's absolutely true at this point. Who we hire is a little bit different, and I think this has changed for every skill set. And, you know, is there one in particular I can help answer or talk about?

Are you hiring now?

We're hiring across the board. And so, you know, you got the way we think about our companies, we've got go-to-market sales and marketing, we've got deployment. So it's like once someone signs on, it's like, how do we make them really successful? And then you got product design and engineering that are building the product. On the product design and engineering side, it's really fun. And you're probably hearing this from a lot of people. All the roles are colliding and we're not exactly sure where they're going to end up, but I think the most important thing someone looking for a job right now can do is embrace a little bit of all of it. So if you're a PM, you want to really flex your ability to have taste, right? Like do some of the design work. And then also be comfortable in the technicals because you can probably, like, you know, one of our PMs here has shipped code every day this week or shipped a feature every day this week. And that's becoming the expectation really quickly. It's like, why are we going to our engineers for some of the faster, easier to do engineering work? And then for our engineers, our best engineers right now are managing many agents. And so they've become mini PMs running a team of engineers, right? And so I think everything is kind of converging to the middle. I still think that each one of those, those functions has something that they're great at. You know, the engineers are great at the architecture and thinking about how these systems are going to collide and dealing with the scaling issues and the queuing issues and all that stuff. Designers, you got to be pixel perfect at some point. You know, it's like we're building a consumer experience. You have to be pixel perfect. And then you've got PMs that have to now kind of like be more technical, but they still have judgment and they're looking at the business and we have hundreds of customers that are asking us for things and prioritizing. So everyone has something they're great at, but you have to be kind of more of a jack of all trades.

Do you name them differently on the job description? It's kind of strange in a sense that when you're saying that you're hiring an engineer, but you kind of need the design experience and PM experience, do you actually Explicitly say that in the job description?

Our first step has been to change the way we hire. Our interview loops look a lot different than they did even 3 years ago, right? We, in our, our prompt now is like, use AI to X, right? The expectation is you're using AI. Um, so that's simple. We screen for people as AI, like we use, we have introduced like a full product interview into our engineering loop because they have to be a The product skills, especially at a company like ours, which you're working in the physical world. Same thing with designers, right? Designers now need— designers are with great communication skills and great structured thinking that kind of like looks more like a PM are way more effective as well. For PMs, we want to see technical skills, right? So we've kind of merged all the interviews so that you're looking for folks that are kind of like really deep at one thing, but reasonably deep at the other things.

For my audience who are potentially data scientists, engineers, and do you have any suggestion on what should they learn in order to, you know, meet the demand of this AI age?

I think like one simple thing to do is think about the engineering manager at some point in career, in your career, that was the person you looked up to, right? And what they were probably doing. was they were giving you great technical advice, but they were also unblocking you and keeping you working really efficiently. And then they were also communicating really effectively with the rest of the org. That, I think, is the job of an engineer today, because you are doing that inherently with a series of agents. And so before I, you know, like someone takes my advice and says, I'll go be a designer, go be a PM.

Yeah.

I think become a great engineering manager. And I think that's kind of the new baseline is that we want to see people that can progress into that.

It's a weird feeling to think that for junior engineers, they haven't been even learning how to do their jobs on their, on their regular roles, in their regular roles, and now they have to learn how to become an engineering manager. Isn't that a gap there? Like, I feel like there's a gap in education or training.

It's interesting. There's a little bit of a gap, but think about it this way. Like, I, you know, my first management experience was something like 10 years ago now. When I, when I started managing, the hardest thing for me to learn how to do was to deal with all of the people things.

Yeah.

It was the career ambition that people had and making sure I was unlocking that, or the compensation problems and things like, you know, things that are totally different than me as a PM giving coaching on how to write a better PRD. And so that's the stuff that you don't have to manage with agents.

Right?

It's like, you're a great engineer. The skills that you now need to work on, or you're a great data scientist, is how do you communicate your thoughts in a structured way? How do you prioritize information correctly? And I think these are all things that everyone can be great at. It's just easy in your day-to-day to kind of like want to just solve the problem in front of you.

Do you have different metrics or signals that you watch for, for like high-performing engineers? The example that I've seen or I've heard is people start looking at like token usage.

Token usage. I don't like token usage.

You don't like it? What about, what do you use?

It feels like a false positive. I think we're going to look and laugh at ourselves one day. I think it still just comes down to shipping, right? And so this has been hard to codify for us, but shipping velocity is very important. So some of this is like retroactively looking at stuff and Kevin leads our engineering team and what we'll sometimes do and say, hey, last, you know, in traditional engineering world, here's what we should have shipped. Are we shipping more than that? Because it is very easy to generate 100,000 lines of code and then have it stuck in code review and bog the rest of the team down. And then the code is sloppy and then nothing gets out. No, everyone was actually less efficient. And I think this is what teams are struggling with today. And so we're doing this in a very coarse way right now where, you know, myself, Kevin, many folks on our team have been leaders of these product and eng teams for a long time. And so we're trying to do it by feel and we're saying, hey, like, are we moving way faster than we used to be? And I think the answer is absolutely. And it's our job to set up the ecosystem in a way where everyone's moving faster, understands how to use the tools. Things like that.

Are you using AI personally every day? And what are you using? Can you share?

Yeah, I mean, every day for sure. Many, many, many times a day. You know, some of it's simple. Some of it's like, I've got routines set up in Claude to get me the information I need to know. I think what I'm not doing is I'm not saying, hey, I should be— like, I'm saying, here's what I need to do. Here's what's most impactful. That hasn't changed, right? I need to Keep the rest of my team unblocked. I need to simplify the logistics in my life. I need to communicate to these people. I need to meet with my investors. And then I'm using AI to make each of those things easier. So summarizing what I need to act on, I am having Claude brief all of my meetings so that everything is ready and easy to use. And so I'm using it probably more like an EA. And then I am sitting with each of my managers and saying, what is the strategy to make your team go 10x with AI? And then it's their job. To come up with a process for their teams.

If you were to teach me one thing to help me become a better founder, what would that be?

In the early days, it's all about iteration velocity and failing fast. A single word is speed, because time is your most precious commodity. It's like you can, you can raise all the money in the world, you can hire all the talent in the world, but time is finite and it goes by fast. And so, every time I look back at something that we did right or wrong, the first takeaway I typically have is I wish I learned that faster. And so, I think it's all about tuning your company to always be able to move fast, learn fast, pivot fast, and so on and so forth.

How do you know that you're moving fast in the right direction though? You could be, you know, wrong in the wrong direction and so on.

Like simple examples, right? It's like, You have a tendency to like, do you have a meeting? And then you say, okay, we'll check back in next week on this thing. And we say, no, no, no, just like if you're learning something, share it right away. Right. And so it's like a cultural thing. If you want to change direction, let's talk about it tonight. You know, and I think this is kind of the thing I've learned from people I've talked to is like tune your company to speed. And that's just, it's culture more than anything else. And being a willingness to like to almost overcommunicate at the right times.

I have a closing tradition to ask all my founder guests to recommend maybe a book.

We had the whole company read Amp It Up by Frank Sloot recently, and it was a book that, uh, you know, he was the, the CEO of Snowflake and ServiceNow before that. And it's a book that kind of gets at what we were talking about, which is how do you build a culture where you're setting goals correctly, you're failing fast, you're moving quickly. And how do you actually set up an org to be really successful through culture? I thought that was awesome. The other thing I would encourage people to do is not only read for self-help and, and, and actually, and read for enjoyment. And so I think there's just something about freeing up your mind if you are spending your time reading as a healthy way to free your mind to be able to come back and refocus on your work.

And the way to free your mind and rejuvenate is to maybe be more active and cycling.

Be more active and yeah, and participate in recreation.

Thank you, dear Drew. It's so nice talking with you today. The vision that you have for making the communities more accessible and not some consumer experience that we all have to suffer through is such an amazing pursuit, and I really appreciate you making this happen for everybody.

I think that we're in a really exciting time in technology, maybe the most exciting time in technology in all of our lives. There's so much energy going into how do we, how do we make the time on our screens more exciting, right? And more powerful, and we can do more on our computers.

Yeah.

And I think for me, it was very important for everyone that joins our company to say, how can we embrace this technology to get back into the real world? And, and just unlock healthier, more active communities, because that's what's going to make us all happy at the end of the day. And I think that's the mission we're on, and I think there's a lot of nuanced problems to solve.

You're working on a technology solution for people to be more efficient on screen so that they have time to spend outside.

Get off your screen.

Awesome, thank you so much.