Tech Brew Ride Home - (Portfolio Profile) Freeplay

Episode Date: November 19, 2023

Find out more at Freeplay.ai. And find out more about the Ride Home Fund at RideHomeFund.com. Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to another weekend bonus episode of the TechMean Ride Home podcast. This is another portfolio profile episode. We're going to talk to a couple founders today of an AI startup.

It's not, however, an Ride Home AI Fund portfolio company. We're going to get into that because there's an interesting connecting story there vis-a-vis that. By the way, Chris Messina is all. also joining us as well. But let's start off by talking to the founders. We have Ian Cairns. Hey there. How's it going? And Eric Ryan. Hello. Hello. Greetings from Boulder. Yes. Today we are going to talk about freeplay, which is found at freeplay.a.I. I want to always call it freeplay AI. But free play is the name. Give me whoever wants to lead us off here. Give me like the two minute elevator pitch what FreePlay does.

Yeah, awesome. Well, this is Ian. At a high level, we help product teams figure out how to build better software that makes use of large language models. So if you're an established software company, been building software for years, there's now these superpowers available to you through LLMs. But it can be really challenging to figure out how to get them to do the right thing in production. and we help solve that problem. So we build tools for prompt management and version control, observability, for testing and evaluation so that you can both automate testing and measure and quantify

how well you're doing in production. And ultimately, you know, the goal is to help people build a feedback loop so that they can make their features progressively better over time. So give me a specific use case. I mean, what you just described seems like a broad tool that anyone could use. But someone out there listening that is either an AI startup or a startup or a company that is dipping their toe into AI, wants to deploy models, wants to do whatever. Explain to me specifically what I would use you for and why I would use you.

Yeah, yeah, for sure. Well, I think one of the fun things for us just working in the industry for so long and talking to people in this space has been most folks when they get started trying to prototype a new idea and see like, you know, what can they build with an LLM? They don't need us. It's, you know, the common story is, hey, you know, we had this idea. Maybe it's been on the roadmap and we thought it would take two years to build. Now it seems like open AI or another model might make this really easy. And it's really common. We hear the story. People are like, holy crap, you know, we thought that was. take us a month. We got a really awesome prototype done in a couple days or a week.

And so I think those folks, that's the story that where we end up coming in is downstream of that. They are able to build a really cool prototype, but then it gets really complicated to figure out how to put it in production. So the path from this works, and it's really promising 60 or 70% of the time to it works reliably so that we can actually trust it's doing the right thing for our customers on a regular basis. We start to help there. So imagine like, you know, example, and one of the customers I talk about that always just makes sense to people, help scout customer service software company. They've been building a bunch of awesome stuff in their product. Their whole goal is to make customer support agents, superheroes, you know, build better

relationships with their customers. So, you know, one of the things they're doing is helping people automatically draft an answer to a customer support email without having to press any button or like think about interacting with AI. You just log in as an agent, it's there. But you know, getting that feature to work on a consistent basis is a trick. You know, so how does FreePlay help? Gives them a way to experiment with different versions of a prompt. So, you know, say you get live and it's working in these cases but not other cases and you want to fix it, really common stories that becomes like a whack-a-mole game. You know, you make some changes to a prompt. Maybe you try a different model, you change some settings, and it's still not doing the right thing. You know, you want to figure out

how do I make this better, but not just chase that, you know, forever. You know, you need to figure out a way to test across a representative set of examples. So, you know, FreePlay helps you save those kinds of examples, basically like your test suite with a large language model is a bunch of examples that you might see in production and you run them through one version and you run them through the next and you compare the difference and see which one's performing better. So we help with that and then once things actually get live, you know, we provide a framework to see what's happening in the system to continually test and evaluate. But maybe that's kind of like a basic. I'll pause there. Yeah, so there's an observability angle afterwards. But also you're you're also allowing

people to like flip a switch and go between different LLMs and test different things like that? Yeah, exactly. So one of the things that we've really focused on when we got started on this, I think there's a lot of people in this space that have been approaching this set of problems as like a, quote, developer tool problem. And we certainly have a developer tool. There's an integration with FreePlay. But we also notice there's a lot of people getting involved in building software.

or improving software here who aren't just software developers. We have cases where product managers or designers are getting involved in prompt engineering and experimentation. Maybe they're closer to the problem. Maybe it's just English and they want the freedom to do that. You have examples where we've talked to accountants, medical doctors, agronomists, people that are getting involved in QA because you know, you're trying to use AI to generate a result, in a domain where even people on the product development team don't have expertise to know if it's doing the right thing. So FreePlay is helping people both do that QA and experiment. So it's model agnostic.

We give people the tools to use OpenAI, quickly try Anthropic, quickly try another model. And our SDK makes it really easy for developers to do that integration once, basically turn the code into a playground. and then a product manager or somebody else can be a developer, but doesn't have to be somebody who's working in code, can actually swap out models or make changes to prompts and run experiments without having to bug somebody to go do a deploy. Well, here's the time where we usually pivot to, we'd like to hear about your background and your entrepreneurial journey.

But to do that, I want to bring on someone that the listeners know very well, but he hasn't been on for a while. Hey, Chris Messina, how you doing? You. What's up? It's been a minute. Yeah, we wanted you on because, as we'll explain, after they give their backgrounds. Well, first of all, you have history with these founders, at least one of them.

But then it's worth talking about how FreePlay sort of led to the Right Home AI fund. But all right, let's do the thing. So, Eric, since Ian's been talking for a while, how about you go first? Why don't you give us your background, your entrepreneurial history, whatever you want to share with us? That sounds great. Hey, Chris, great to see you, I guess, virtually. We did cross paths once upon a time and then get up years. Yes.

Activity streams era. I mean, it's kind of wild just to think about where we are now with like Macedon Federation activity pub threads. Like all that stuff that we've worked on before now is finally coming to. ahead. So it's good to come full circle. Totally. Agreed. Well, everybody, Eric Ryan, I grew up in Delaware. I've been playing with computers since I was a kid. I've been building software and engineering teams for the past 15 or so years. Early engineer at a company called Ginnip, which is a strange name, but it's ping spelled backwards. We built out the pipes for the social analytics,

social listening ecosystem. And Ginnip was actually acquired by Twitter in 2014. And there I eventually went on to lead engineering for Twitter's developer platform and say engineer at heart in many ways. But my main focus has always been hiring folks that are much smarter than myself and building a strong culture around them as a foundation and then try and couple that with a strong execution machine. But yeah, Ian and I were chatting about what's next last fall. We wanted to work on LLMs. We saw how transformative that was going to be. We also knew developer products. So we ran a process in Q1 last year or this year spent time exploring a couple different verticals, things in education

and the customer service space. And we became convinced that that probably wasn't for us. And we wanted to get back to our infrastructure routes. So that's about how we got started with free play. And Ian, you go and feel free to bring in your Chris Messina anecdotes as applicable here. Cool. Well, a lot of my background overlaps with Eric. We've worked together most of the last decade. But, you know, on my career side, I've always been on the product side of things and always working on, you know, I've described it to people as like developer products that are way high up in the stack that are actually helping other people build products. So way in the early days, back in 2005,

I first got involved in the Drupal Open Source CMS project. That's right. I forgot that we go back that far. So I knew this designer factory Joe in the Drupal ecosystem. And I was working for a company called Development Seed that we built a few products. So we were involved in Drupal, but we also launched a news aggregator. We launched a collaboration tool. The third project that we launched was called Mapbox.

Mapbox became a much bigger thing. So I got to be around for the beginning days of that. Never got to actually work for Mapbox, the company. But if you don't know Mapbox, Mapbox is used to early on help developers put maps in their products and do cool map visualizations. Now you can also use their SDKs to run a self-driving car program. But check it out. So yeah, I think from there I went on, I get to spend some time, bouncing around DC, get to work in the Obama administration as a consultant and do some product-related work,

help the White House ship their first API, and had a great experience in public service, you know, that side of things for a little bit. But Ms. Startup Days came back to work with Eric at Ginnip and then went on at Twitter, you know, spent my time, I was there for almost seven years, eventually ended up leading product for the developer platform and the Twitter API. But yeah, coming full circle, like, you know, Eric and I both got really interested. in LLMs together, missed working on technical products together. We'd both gone to other companies that were more tech-enabled businesses and couldn't stay away from what we saw happening. And this was even before chat GPT launched.

So we were ready to go just in time. Right. Well, give me a Messina anecdote if you've got one. And then I will tell you why you guys are so foundational to what Chris and I are doing these days. Well, we already heard one, I guess, which was, you know, the activity streams protocol that turned into so many things happening with Federated Social today. That was a, Gnip was one of the earliest adopters of that. Chris helped work on it. And I feel like a lot of that's actually relevant today with what we're facing with large language models. One of the most common problems we face right now

and activity streams has been mentioned is, you know, how do we help normalize the interface between a bunch of different models so that developers can easily experiment and test different things. But I don't know, maybe one of my favorite anecdotes was sitting with Chris the day that he gave up his factory Joe name and decided to be real Chris Messina on Twitter. I don't know if he wants to comment any more about that, but there was a deep conversation about having real identities on the internet. I feel like the last decade there's been a bunch of issues that have persisted from there. Yeah, actually, I remember pretty sure David Recorden was there and several other folks, whatever. We were at South by Southwest. And as I was want to do, I surely had several cocktails probably.

And I was feeling just very frustrated because I think I'd probably given a talk. And people had probably come up to me and refer to me as Joe, which obviously was not my name. And so that was very confusing, except that was my online handle. And so it was starting to become kind of a drag on my real world experience where my internet identity was somewhat better known than my real world identity. And the, I guess, juxtaposition of those two things made it seem like it was time to perhaps retire the handle and actually become myself on the internet. Up until that point, I think I was terrified of that whole prospect of outing myself on the internet and therefore being accessible to all the trolls and all the other people, which of course I probably was one. But anyways, I think what ended up happening was Twitter had the ability to change your username, as many platforms do. And so, you know, in this sort of, you know, drunken moment of, of O'Carrage, I suppose, decided to go ahead and change my username from factory Jada Christmasina.

And man, it was such a weird sort of feeling of of nudity in that moment, of internet nudity, I guess, or death. Either one. It was, it was, I felt vulnerable. I'll put it that way. That it took me a moment to just, like, pause and kind of like collect myself. And I was like, okay, it's done. And within moments, I went back to sort of like claim my old username and someone had grabbed it.

And I was like, you've got to be kidding me. Like, this is terrible. Because back then, like, hermalinks were like a really big deal. And like link rot was like something we were worried about. And so all of a sudden, all my links to all my old. tweets had broken because of course now I had a new username and now someone else had already grabbed that name. I was like, you got to be like, who would want that name? It's the dumbest name ever. And it turned out that it was either David or Corden or Max Engel, some of our friends from the

identity space. And it was just such a like transitional and pivotal moment, both for me personally on the internet. And then also for the degree to which I guess we started to really recognize how much these online personas and personalities and identities really make a big difference. And fortunately, I think I was able to punch record in the shoulder several times until he finally gave up the account. But there was just a moment of real loss. And it's funny, I guess, reflecting on this year, and I think getting into this conversation about free play, once Elon started making his changes on Twitter, now X, and I decided to

leave in April, you know, I went through my second digital death. And so that first digital death of Factory Joe was sort of like, I don't know if it prepared me for this year, but effectively allowed me to let go, you know, my Twitter account, my 100,000 followers, and to move on. And I feel like that's a similar sort of journey, perhaps, that you've had to go through, but maybe in a little bit of a different way with regards to, you know, your former employer. Yes. So if I could bring that up, and this will say into how free play, led to the birth of the Bright Home AI Fund. So essentially, you sort of yada yotted over this part.

But last fall, you were looking for something new to do, and that was because the great Twitter calling happened. I wouldn't call it great. Well, unfortunate, but sure, yes. Yes, the substantial. Yes. So we don't need to go into any of that drama or anything like that. But one of the, I'll give you the connection to the AI fund.

You know, as I've said many times, you know, Chris has helped me out with my other fund for many years. And I had always said, if there's somebody that is doing something new that by definition, you'd almost invest in them site unseen, let me know. And Chris says, oh, my God, like this whole team from Twitter is, is jumping into this AI space. And so that was almost the, that was the birth of the thesis of the fund, is that if folks like you were coming off the bench to do something new, then we were like, well, this is, because we were already activated,

and Chris can speak to this as well. Like, we felt like this was the biggest new moment in both of our professional lives in tech since Web 2.0. And we were like, if there's other folks that are like us that are like, oh my God, this is so new, this is so amazing. Those are the people that we want to fund. And so literally, the whole conversation about getting connected to you all then led to Chris and I talking about, well, then let's just do a fund to do that.

And so now the Right Home AI Fund exists. And so FreePlay, that's your fault. Love it. Amazing. I don't think we actually knew that story until recently, but yeah. It was a fun time. Like I said earlier, Eric and I had both gone and done other companies in between. You know, Eric was at a company called Main Street. I was a company called Fur Space.

But that was definitely part of our story. The two of us were both excited to build with LLMs. And we knew a bunch of our crew were looking for new jobs. On the market or soon to be. It was an exciting time to get a lot of the old band back together and get to build with people who we really love working with and trust. And for all the other things that might have been heavy about that time, there was a lot of just really great people and incredible talent at the company.

So let me ask you about that specifically to bring it back to FreePlay Story. What was that like in the sense that, like, as you described it, you see the LLM and AI moment happening. You have previous experience with developer products and APIs and product itself. and it's not like it's when when people are people who are listening are thinking about well I would like to do a startup I'd like to do a thing can you walk through that process of saying we know we want to do a thing some people just know they want to do a thing some people know they want to do a thing in a space some people know they want to do a thing to solve a specific

problem can you to what degree you're able to remember this sort of thought process like walk us through how that worked for you. Like, we know we want to do something here, but what's the best thing that we can do with this? Yeah, that's a good question. I mean, I don't know, trying to remember. I'd say this is the first time that I saw a help wanted sign. Both we talked about this, Ian and I saw a help wanted sign for something that is truly

transformative that our skill set is applicable to, right? So I think the altruist in me was always like, oh, I'd love to get a. into green energy or carbon credits or something like that. But never really saw a direct connection with being able to help. And I don't know, I think that was the initial spark. Like I can actually jump in and help here. Yeah. Yeah, I think we also, you know, we, we had a gut check.

You know, we talked to each other a bunch early on like, wait a second. Anybody that's starting a company or anyone giving advice about starting a company, Like the classic line is, you know, don't start with the solution. Go start with, you know, a problem, fall in love with the problem, be ready to find different solutions to the problem. And, you know, we were trying to gut check early on. Like, wait, we just want to work on this amazing technology. Is that the right answer? You know, so we actually did try to be faithful to go do some real, like, bottom up research and understand, like, what was going on in this space where were the opportunities.

I think we did come back around pretty quickly to the things. that we've known and are good at, but, you know, that we're a unique new problem. Like there's a paradigm shift happening in how people are building software. We're still in such an early, you know, stage of that, too. And I think we saw that at Open AI Dev Day last week. You know, they really rolled out an even more bold vision, I think, of what, you know, the next stage of interacting with computers is going to look like. But we saw that it was different enough that for people that we'd worked with,

and built products for a decade or 15 years, you know, there was going to be a real learning curve. And we also knew early on there was going to be this explosion. You know, there's an order of magnitude, if not two orders of magnitude, more people in software development building with ML technology today than there were a year ago. And most of them have never run ML systems. They've not had to deal with non-deterministic systems. They've not had to figure out like, okay, you know, how do I deal with something that's an approximation, much less generative models where they're not going to do the same thing twice. And there's just enough differences that there'd be a real need there that again, like Eric said, aligned to our experience and our background. So, but yeah, that took us three months of doing some real exploration and research before.

But by research, you mean like literally talking to people. And you had the advantage of, you know, you had folks that you had worked with before that you can reach out to and essentially say, if you're exploring the space, what are the problems you're having? What are the pain points you're seeing or whatever? Because I even think we invested before you settled finally on what the product was going to be. I think we invested while you were in this process of literally asking people, if you're trying this, what do you wish someone could help you do? Yeah, totally. I think at this point, we've probably talked to well over 100, you know, early on that first sprint, you know, several dozen CTOs, heads of product, design leaders, senior engineers, that companies ranging from seed stage to public companies, and, you know, and that was one of the crazy things is everybody was saying the same thing. No matter what stage you were at, there was this, like, the last time I saw it was with GDPR, but not such a positive, you know,

reason for the whole industry just stops and they're like top down we're going to do something different you know next week we're going to tell our teams go build something with open AI and and and so they're all saying we're going to we're going to build something but what was the thing that kept coming back in terms of the answer of what the pain point is that led you to what free play is today yeah totally i mean that that sense of people saying hey we got through our v0 we created the prototype that's promising our customers actually do like it, but we feel insecure, uncertain, maybe terrified about the lack of understanding about what the things actually doing in production. We have no way to, you know,

improve on it with any kind of quantitative discipline or confidence. You know, talking to CTOs and heads of engineering here, we're saying, you know, I've been running a professional software organization for a long time. We have, you know, reliable testing practices and discipline around QA. And we don't even know how to QA this new thing that we built. I was going to say, so one of the things that that's, I guess, coming out for me and how you're describing this. And I think I'm trying to sort of, I don't know, like, figure out and relate to both how like software design and software production and engineering is changing, but then also how product development and that work is changing. And I think what FreePlay is doing and what

is increasingly happening to product development is there's a different type of scientific method that's being brought into the fold where it's a lot about experimentation and it's a lot about sort of maybe seeing around corners and cutting out like edge cases and kind of generally like steering an AI kind of like in a direction, but in a way that it's almost like a herd of cattle and it's kind of like you can get them to move in a certain way, but there's still some random aspects to it. So I guess my question is around positioning relative to existing solutions for product development and for engineering relative to where this goes. So things are moving so quickly. Obviously Open AI Developer Day also brings a whole new set of concepts, you know,

clearly they're grabbing hold of this GPT's concept. What do you see is maybe where FreePlay is in like like one to two years in terms of the way in which you want to be used by your customers. Is it something that is for post-production and post-launch where you're like there's a constantly, I guess evolving the product surface area and what it can do? And so it's basically like an infinite game or does software, does AI software ever kind of end where you really cool? That's like R1.0 and it's like done. Because it feels like, yeah, like that doesn't exist anymore.

So how do you see this field kind of evolving? Yeah. You think you're done and then a new model comes out that's compelling because it's state of the art or it's more cost effective or it's better latency or whatever else. We're seeing that in the past week. All of a sudden, people that have said we're done suddenly want to test the new open AI models. There's a ton of interest in open source models right now for privacy reasons and otherwise. So yeah, we don't think it's going to be done. I mean, I think about it like, you know, what's the enduring, need of any product development organization, like everybody's ultimately trying to build a better

product for their customers and make sure that they're growing in a positive way. Right. And I think about that way, like maybe free play in one to two years is something like maybe amplitude if you're familiar with that has been for Web 2 software. What's the way that you understand how the product's performing, get insights that help you figure out how to make it better, run a experiments to actually quantify that you are, in fact, making it better and really bring that feedback loop to life.

So that brings this really interesting question, I guess, around the relationship of generative AI applications and perhaps conventional applications that are well served by an amplitude. Right? So is there sort of a collision course where these things come together? An amplitude is offering similar types of functionality that you are? Or is this sort of, you know, product teams will be using three or four tools that are kind of adjacent to what amplitude does, what you're doing, and it's all part of a suite that is required to build both conventional software that has somewhat conventional interfaces,

you know, like pixel-based interfaces, and there's generative AI aspects, which are kind of anticipatory and probabilistic in that they work with the customer, the client, the user, to achieve some sort of outcome that's a little bit more abstract. Yeah, for sure. I mean, it is so early. Like, it'll be fascinating to see. I think a lot of people are rushing into this space. If you've been building tooling and observability or tooling and continuous integration testing,

like you're seeing a lot of established companies say, hey, we have an answer for LLMs. And the interesting part is so far it's needed to be a different answer, right? Like so much of the way software has worked in the past is around structured data, you know, things that can be clearly and easily quantified and counted. And now, like a lot of the folks we're working with, you know, you're generating these large blocks of text that are very hard to make sense of and very hard to quantify, especially if you've not been doing that. You know, when we were at Dev Day last week, some of the Open AI team did a session. I think the talks are all online now, but, you know, they were talking about just the challenges of going to production and one of their big slides said lack of evaluations has been a key challenge for deploying to production. You know, what are they getting at?

It's like, you know, at some point you have to quantify, what is the characteristics of how this thing is behaving across a bunch of example use cases. And that's really what we're doing. I think we're taking an AI first approach. I think we're, you know, like just very focused on the problems that are unique in this space. And I think we're also really focused on this change that is happening with who and how software is getting built, you know, bringing more folks into the process than just the developers. But yeah, I mean, I have to imagine a lot of existing tools will add LLM features.

I think we're bullish on the opportunity to just reimagine from the ground up what these kinds of solutions should look like to really help people build better products when you're starting with generative AI as a way to create them. I think it's, you know, it just occurs to me and I'll like a Brian going a sec. You know, there's a pretty well-known McLuhan kind of, I don't know if it's an adage or a dictum. I don't know what the difference is, but something he said. So, you know, McLuhan, famous media theorist often sort of pointed out that new mediums are often kind of at first filled with the content from the previous era, from the previous media epoch, because people are just trying to figure out how to do it.

So, like, they'll take, like, radio shows and they'll try to adapt them to television. And it just doesn't make sense. It doesn't work. And in this case, I don't know that we've, it doesn't feel like we've had this conversation in the tech world as much, but it's occurring to me that code is a type of media format and that we're moving into a world where, like coding and software development as a medium for collaboration or for solving problems, is suffering a similar type of circumstance where previous developer tools are tempted,

attempting to be brought into the LLM world, the world of like sort of software coding and generative AI. And it's not a good fit. It's a different medium. It's a different format. It requires a different way of working with the tools that allows for a lot more nondeterminism, you know, as you pointed out. And so it requires, and so I guess like the question I have is to what degree, and I suppose this is what you're doing. And so maybe the answer is in what you're building, But the degree to which you start from carte blanche to solve these problems for generative AI application developers versus building bridges to existing IDs or software that already exists. Like the co-pilot model kind of builds into an existing context where there's like code completion, as opposed to what I'm increasingly seeing like I'm product hunt are people that are like almost like launching like application templates where you can kind of interactively choose the parts that you want and the problem that you solve.

And the application kind of designs and builds itself. and you iteratively improve it. So you're not really getting into sketch. You're not really getting into Figma. You can use those tools if you want to, but they're almost clumsy relative to the way in which the software allows you to just express an idea and move towards like an outcome. I don't know that makes sense as a question,

but I guess I'm asking, you know, again, like does it make sense to sort of build free play into more existing developer tools that already exist or to focus more on a green space? Yeah, I mean, it'll be an interesting question to answer in the coming years. I think what's been happening so far, you know, I was just on the phone with the team last week, like a household name media brand, and they're building in a space, you know, and they described tools that they're working with, like traditional ML ops tools, you know, even tools that have been created in the last, you know, five, half dozen years somewhere.

And some traditional observability tools, and they're like basically trying to map them into this space and then we walk through, you know, what we're doing at FreePlay. And, you know, it was really encouraging. You know, kind of see the lights come on for folks and like, wait, like you've built the workflow, not just the instrumentation, but you built the workflow that is the workflow we need for this new reality. And I think that's what's different, right? It's like it's not just an observability problem.

It's not just a testing problem. It's not just a version control problem. And there's a different enough way of creating with this kind of technology. that we've found a need for a fresh workflow. So we'll see how it all unfolds, but that's where we are today. Nice. Second to last topic, since we've been mentioning the OpenAI keynote, essentially, or everything that they did last week,

I'm just curious for your take and almost the macro-level take in terms of what they announced for what, what they announced and what it means for the overall AI ecosystem right now. I've heard some people say they announced more than I expected. They didn't announce as much as I expected. Where's GPT5 or why didn't you open source GPT3? But then also some people are like the things they did announce are weirdly underpowered. I'm just, and whatever I don't, I'm trying not to lead the witness here,

but whatever your take was on what they announced and what it means in a macro sense for the ecosystem. Well, I mean, my quick takeaway was I feel like, you know, we saw a company that's executing at a level that very few others in the industry are. What they rolled out, I thought was incredibly polished and, you know, well done. If you haven't watched the keynote and you only have a couple minutes, jump straight to minute 33, watch Roman, who leads developer relations over there, show the new assistance API and just the story that he tells and how it all comes together with voice and, you know,

images doing OCR on PDFs and creating an application pretty quickly on the fly. It's impressive. I was actually going to ask you since you mentioned, Ramon, like, I presume you guys overlap to work together. Is that right at Twitter? Briefly, yeah, we were there at the same time. And so I'm just like, you know, obviously he spent a stint at Stripe and so has that background. But not to be too insidery, but I'm just curious, like, what your impression is of having him kind of at the head over at Open AI on their developer relations front. And if there's any observations you have, having worked with him.

Well, I mean, I think if you watch the keynote, like it's similar work that I saw him do at Stripe. He's just incredible at putting the pieces together of an otherwise complicated set of APIs and developer surface area and just showing in a simple way how they can be useful to people. So, you know, Open AI has got this opportunity to get the entire world using their platform. Creativity might be the limit. Like, I think it's great. But yeah, I mean, I think what we, what we saw there was there's a ton of stuff. If you're working in enterprise software, let's put it this way. Most of our focus is with people that are building B2B software.

You know, they did show a new horizon line. You know, if you're maybe working on AI research, you were left wondering about, you know, why didn't they open source GPT3 or some of the other aspects of it? But I think that the practical application of how LLMs can come together and multimodal can be used, we just saw it in a really polished way. So I think that's what they did well. And I think that's a lot of where our customers are focused is, oh, great, now there's a new horizon line, not just how to adopt these new models.

Like we spent last week, you know, basically at lunch break at Dev Day, getting the new models added to free play. But then, you know, what's multimodal testing going to look like? like what's the assistance API going to do. I think those are all things that most companies are going to start to step into throughout 2024 and maybe not until 2025. There's just a lag for adoption. So I think in that way, it's like there's a lot to grow into for the ecosystem there.

Eric, I wanted to offer you the last word on this, either, again, a macro-level sense of what you took away from that, or if you have a sense of where Open AI itself is going, Did you see a vision of what that company, what direction they want to go in? I don't know. I think on the first part, one of the things that really stood out to me was the focus on cost optimization, especially in a world where there's been a lot of chatter about open source models and them being competitive. The fact that they are deeply optimizing on cost there, I think just changes the calculus

for a lot of people that are building with this. So I don't know, the land grab is interesting. Land grab in the sense that are they grabbing part of the ecosystem for themselves that is the ecosystem that maybe some of us are hoping for? I don't know. Remains to be seen. I mean, I think if you're, you know, someone that's building with this and you're worried about cost, you know, there is real cost to running an open source model. And you just need to pay attention to that. Okay, what I do want to wrap with is our usual bits, which are, again, this is FreePlay. You can find out more at Freeplay.a.I.

But also, anyone listening who is interested in maybe becoming a client or working with you or a partner, some are you guys hiring? Basically, this is at the end of the show. if people are intrigued by what y'all are working on, how can they get in touch? What are you looking for? What are the asks from the audience if you're looking for something right now? Yeah, well, I mean, I'll talk about one, let Eric talk about the other. On the using FreePlay side, we just launched a public beta two weeks ago.

So we've been heads down building with a small group of customers for most of the year, now open for business and ready for anybody. to come check it out. So if you're interested, freeplay.a is dot AI is the place to go sign up for that. And then we are hiring. Yeah, on the hiring front, I mean, look, we're always down to talk to fantastic engineers. Right now, we're focused on full-stack folks. We actually are now hiring our first full-time product design lead. And we're also looking for someone, maybe a little bit of a Swiss Army knife,

call it Devrel, call it sales engineer, someone that can help us really drive and focus on customer success and onboarding and can really get deep into the technical details with our customers and make them happy. Obviously, I'm sure there's a place on the website that you can look at those opportunities, but also I will offer up my email, Brian at right home fund.com if folks are interested in those two roles. And I will forward those along to Ian and Eric. Gentlemen, thanks for telling us about FreePlay.

Again, not part of the AI, the Right Home AI Fund, but literally the inspiration for it. How the AI fund existed, it would have invested. 100%. So, you know, thrilled to be an investor myself and Chris has been working with you all as well anyway. So just thrilled with what you're doing. doing and thanks for coming on the show to tell us about it. Yeah, thank you guys, both for the support and for having us on. We appreciate it.

Thanks so much, gentlemen. Good to see you guys. Hey, Chris, we got to find a way, at least in the new year for you to come back and do an episode, just you and me jamming on something. Yeah. We'll do it. All right. Thanks, everyone.

Love y'all.