All these companies are working on intelligence and our belief is that empathy and emotion is not an emergent property of intelligence. So we need to build the systems and interface for humans to be able to properly engage with this intelligence and use it. When you combine this right brain and left brain, that's really when we have what we consider to be a true digital human. So like imagine you're, you know, a first grader and you're not quite clicking with content. Imagine you could just like go spend an hour one on one with your teacher.

You can ask them any questions, they can read you a storybook at at your level. You know, really just like explore the bounds of creativity. And that teacher wasn't actually there, it was a digital AI teacher that actually understands the context of what was going on. It interacts just like a human does in a face to face scenario. It's able to engage and immerse at the end of the day this this student.

How do we make that so simple for them to use that they don't even have to realize the crazy complex running under the hood where they can just oh, it works. We're all changing the world together. Right? Like it's it's not enough to just build the product. Like people have to be using it and creating value with it.

Yeah. Especially because I have the exact same camera at home and it records way more than 20.

Yeah. Yeah.

Yeah. Well,

if you have you probably have the version four or Symfote.

I well, so I have the the seven version four and then I had the 6,000 for a while. I I'm big into photography.

Okay. Like, I love it.

Yeah. Videography, not not so much. But photography, I'm a huge fan of.

Yeah. That's funny. Kind of ironic a little.

It really is. It's ironic because

You know?

I'm more into photography than videography. In my personal life, I hate technology. Like the the whole nine yards, it's it's hilarious. Appreciate technology but but hate using it. I wanna be as far from it as possible when I'm away from work.

Yeah. That makes sense. I will OpenAI and and meta Can you dig into Not an Emergent Property?

Yeah. So as intelligence becomes better and better and better, right, and and eventually reaches this like, you know, gray area concept of AGI, right? Whenever that means to everyone, I I think there's different definitions. That intelligence our our belief is that intelligence will not become more accessible or easier to extract, right? It's almost like, have you ever like talked to chat GPT and you're like, you know, do this and then it does something completely different?

You're like, no, no, no.

Yeah. Yeah.

Right? That's a prime example where it's like we have this concept in our head where maybe we're actually physically talking about it or our face conveys, you know, the details of it and and, you know, we're communicating, but it's not being received properly on

the other

The reality is like when you're able to see micro expressions, when you're able to understand, you know, peripheral vision and and the focus, when you're able to understand patterns of speech and you're able to communicate that way, you're able to really get the full picture of knowledge. And so when you combine this right brain and left brain, that's really when we have what we consider to be a true digital human.

Scaling DevTools is sponsored by WorkOS. If things start going well, some of your customers are gonna start asking for enterprise features. Things like audit trails, SSO, SCIM provisioning, role based access control. These things are hard to build, and you could get stuck spending all your time doing that instead of actually making a great dev tool. That's why WorkOS exists.

They help you with all of those enterprise features, and they're trusted by OpenAI, Vercel, and Perplexity. And if you use them for user management, you get your first million, yes, million, monthly active users for free. I honestly don't know any dev tools that have a million monthly active users apart from GitHub maybe. So that'll get you pretty far. Here's what Kyle from Depot has to say about Work OS.

We use Work OS to effectively add all of the SSO and SCIM to Depot. It's single handedly like one of the best developer experiences I've ever seen for what is, like, a super painful problem if you were to go and try to roll that yourself. So for us, we can effectively offer SSO and SCIM, and it's, like, two clicks of a button, and we don't ever have to think about it. It's like one of the best features that we can add to Depot. It's super affordable, which effectively allows us to like break the SSO tax joke.

Essentially say like you can have SSO and Schim as like an add on onto your monthly plan. Like, it's no problem. So it really allows smaller startups to essentially offer like that enterprise feature without a huge engineering investment behind it. Like, it's literally we can just use a tool behind the scenes, and our life is exponentially easier.

It's pretty it's pretty mind blowing when you get into that. But it's, yeah.

I mean, it's it's an awesome future. It's one where Yeah. I think this intelligence will be democratized in a lot of ways. And like, I mean, let let's think of like the AI therapist use case. I've said that once or twice.

Yeah. It's a little bit cliche. Think everyone uses AI therapy as

I'm gonna say I have to use that.

Exactly. You know, like with an AI therapist, right now I have to pay $200 an hour to see one. Yeah. I have to schedule ahead of time. They're not always accessible when I want them They to don't really understand me as a human.

Like, they they don't have this understanding of like, who I am, like, my background, like, what I'm saying, like, things get lost or are not related versus, like, the AI human perception of that is fascinating. Like, I can get home after work on a Friday at midnight and, jump on a call with with the AI therapist and, they remember exactly who I am. They have this built up context. They're able to see me and understand I'm actually communicating with my microexpression Yeah. Subtleties.

And the the, like, you know, accessibility of that is just so high, not to mention the quality of that as well, right? There's no trade offs to this, like, scaling intelligence with with humanity involved as well.

Yeah. I mean, yeah, I agree.

And it's I think it's often even just like the timing is like such a big deal. It's like, usually like, you know, when you go to a therapist, it's like a normal hour of the day at like a respectable time like off the run.

It's like usually like you feel feel worse than like like 1AM.

You know, maybe had a few drinks, you're just like

That's why I I I tried having I ended up building my own little like, you know, Tavis Therapist bot which was awesome. I use that now. Yeah. But, you know, I had a I had a a real human therapist for a while. And the issue was it'd be like at 2PM in the work day or something Yeah.

Of the availability they would have. I would go from like meetings meetings meetings to like this therapist calling like, I wasn't mentally ready for it. It was like context switch, I wasn't in that frame of mind, and it was just not not as good of a conversation as it should have been on either side. Yeah.

It's a I gotta guess like, very very controversial one to say but like it's almost like sometimes, I mean like when you talk to the on the speech with Chaji Bhatti, already really good. Almost feels like it's like okay saying this, not that it's actually God but it's like it kind of almost feels like you're talking to like a god or like because it's just just like has so much

It's sci fi. Right? At the end of the day, it's we're we're living in this like futuristic world where these things that we saw in movies growing up and you know, the nineties and the February like they're here. They're they're real in And many I think, I think that's one of the maybe one the coolest parts about building Tavan is like, you know, obviously we we do a lot of research and we we build a lot of product here. But what always makes me the happiest is when we go in person and actually show someone what this is.

Because we're talking about it and it's hard to visualize, right? Like, it's actually a pretty hard product to articulate, like, what it does. And I I would venture to say, like, a lot of people listening to this, like, might not have the exact concept of what it is until they actually try it. But when you see someone try it, you actually see their jaw drop. Yeah.

It is a like physically like jarring moment where it's it's magical. It's like, woah, holy shit. Like Yeah. This is real. This is here.

And like Like Waymo. And quite literally, Waymo's a great example. Yeah. Chat GPT for the first time was Yeah. Was another example of of one that I've used.

And like, you sort of have this like series of magic moments that help you realize just like where we are and like Moore's Law of like, oh my gosh, like it's speeding up.

Yeah. Like going forward in the future, so who do you who do you think would be like the biggest customers of Tavis in like, you know, like ten years or something?

That's a great question. So we Tavis is is built Tavis is a series of models to build digital humans, right, and and to provide these digital humans to the world. And what that means is, you know, I see a world where everyone and anyone can go in and and use Tavis and say, hey, I want to, you know, create a, you know, AI blank. Here's where it connects to my knowledge base, here's where I want to deploy it, and it's a, you know, natural language interface, which is really easy for anyone everyone to use. And it becomes, you know, a tool that is that is seen in day to day life just like a chat GPT or or just like something like that.

Now, when we talk about like where we're seeing the biggest impact today, there's a few spaces that we get really excited about. One of them, interestingly enough, is healthcare. So we're seeing this left and right in healthcare. Pre op and post op care, patient intake, companionship, elderly care, fascinating stuff. We're seeing a lot of use cases in recruiting and, just like talent in general.

Mhmm. Not just for interviews, but actually on on both sides of the fence. And then the final one that I I personally get really excited about, and this is the one that that brings me a lot of joy, is education. And obviously there's the, you know, like corporate l and d, role play, coaching Yeah. Training.

But what what really excites me is seeing this actually pop up in schools. Yeah. That was a lot of my inspiration for what brought us to Tapas. And I think like seeing that like K through 12 use case where everyone's having that personalization and that like actual like touch that makes education engaging has been really cool.

So I think you spoke about on another show was like the the teacher could like send home, like kind of do like a tailored lesson every day to the Yep. Student as homeworker.

Yeah. So like imagine you're, you know, a first grader and, you know, you're at a different reading level maybe than everyone else or, you know, you're not quite clicking with content. You have some questions that you weren't able to answer in class. Imagine you could just like go spend an hour one on one with your teacher, later in the day. And you can ask them any questions.

They can read you a storybook at at your level. You can, you know, really just like explore the bounds of creativity. And that teacher wasn't actually there. It was a digital AI teacher that actually understands the context of what was going on. It interacts just like a human does in a face to face scenario.

It's able to engage and immerse at the end of the day this this student. Like, that's awesome. That's really cool. I mean, I would use that today as a non k through 12 present. Right?

Yeah. That that would be like I remember like when YouTube first came out because my school, some of the teaching was like fairly questionable. They had like a few a handful of good teachers and it was like, most people were just there.

I think that's a pretty universal

experience. Yeah. And it was like He's from Khan Academy. Came along, was like just wow. Like, and and the likes of Khan Academy, it was just like, you could just go learn from these incredible people.

Yeah. Explain it in a way that you actually understand it. It's just

Even GPT for the first time. Yeah. Right? Like I remember driving home a few months ago and I instead of playing music, I would just put on, you know, voice to voice and talk to him. I'd be like, tell me about blank.

And then there'd be this back and forth lesson where it was really immersive and I I would actually like learn something real.

Yeah. That's incredible. Tavis will not be building like this company that that that able to teach the kids. You're gonna be like a kind of plumbing underneath. So

Yeah. We we build the models that power these experiences. So, our research team who a group of phenomenally, brilliant engineers and and researchers, we build a series of core models, in rendering, in perception, in turn taking, and others. And then we bring it together in our conversational video interface, which effectively serves as the operating system for this, you know, human AI computing. CVI, conversational video interface, is then accessible to developers and product teams to build on all via dev first APIs and it's super easy to use in day to day.

And that's really where we specialize. So we're this, not quite infrastructure layer, but we provide this operating system to, you know, other development teams and engineers to really build on top of and then deliver those experiences like the teacher.

Yeah. So could could you would you would it be like fair at all to say like, most like the Twilio of like, kind of like conversational AI?

To a degree. The the the infinitely more complex, like, like, sort of varied, version of Twilio. But you could totally use that as an example. Right? And the way I'd say it is if you're looking to build a a experience or a product or a platform with conversational video, that's where you would use Tavis and and the APIs.

Yeah. Yeah. And do you plug in the kind of like so if someone wants to be able to like understand what is the student is asking and then like generate the thing, You work with Chantry, OpenAI's models.

Exactly. So we build all of our models in house except LLM and text to speech. Yeah. So text to speech, we use some amazing partners, Cartesia, Eleven Labs, those folks.

And then

on the LLM side, you're able to integrate the Tavis conversational video interface with your own, you know, sort of knowledge base or or information, whether that's, you know, a llama model, ChatGPT, really anything else. The way we've structured the pipeline is there it is a, you know, fully end to end, out of the box, ready to go pipeline, which you can use, you know, with the, you know, Tavis like fine tuned optimized LLM right away. Or you can treat it as building blocks where you can sub in your own LLM, you can change out the step, things of that nature.

Yeah. How does it work like in terms of like actually like developing? Because it's I guess like you're used to this like, okay, a user does this and then my program should do this or this depending on some variable. And then but for for like what applications like an AI tutor are, like tutor, it's like you kind of it's like different to the program. Right?

Yeah yeah yeah. It's it's not quite a if then tree or anything like that. Yeah. So we have like a layer in the ecosystem which we call personas. And a persona informs the LLM like who it is, what the conversation is, what the goals are.

And it has, you know, things that you can talk about on or off like memory, guardrails, and things like that. And what you're able to do and and actually, this is for prompting. It's really important. I think a lot of people still, like, underestimate how important prompting is. Like a one word change, like, fundamentally can change how smooth a conversation is.

But you can then prompt in and say, hey, you're you're in an interview. You know, you are the interviewer. Here's some background about the candidate. We want to understand if there's a good fit. Here's the 10 questions we want to ask, make sure if the response isn't detailed to, you know, ask an elaboration, etcetera.

There's different styles to that. Some people it's like very firm prompting, right, where they have, you know, a one by one, they want the answer, then they move on to the next. Other people, they leave it more open ended. You can also use a second LLM as like more of an orchestrator. So it's pretty across the board in terms of how you can actually execute on that.

Mhmm. But the persona is what informs the the LLM, like, what the goal is, who it is, and and the context behind the conversation.

Yeah. And do you do any like sort of development stuff that, you know, is like maybe non obvious to someone that's doing like a traditional dev tool? Like how does someone is that dev mode in like Tavis like? Yeah.

As as actual, you know, like Tavis developers or as people building with Tavis?

People building with Tavis.

So there's a lot that happens behind the scenes that people don't don't necessarily see. For example, you know, we have to teach the LLM how to just have a human conversation in the first place. Like Yeah. Before it's even told, you know, hey, you know, I'm I'm Jack. I run, you know, the the DevTools podcast.

It has to understand, okay, how do we just communicate in a human manner? Like, what is conversational versus like non conversational? So teaching and things like that. But we have quite a few different like playgrounds and parameters in DevTools. Pretty much everything that you can use within the Tattoos platform is exposed as a parameter for you to play around with.

So you can adjust things whether it's like interrupt sensitivity and pause sensitivity all the way to, you know, like, what is the accent the replica is speaking with and, like, how does it respond, even to how is the player embedded and where can someone access it. So really everything is fully customizable, and that's the beauty of it. Yeah. So most people, whether they know it or not, have probably interacted with something like this before. And it might have just been such an immersive experience where they didn't quite realize that.

Very cool. Very very cool. How is it how does the English accent do in the desserts?

That that's the hard one. You Europeans, if you're GDPR, you know, you always want the, you know, everything to be perfect, but then you won't give us your data to train on.

I'm sorry. No, you can

you can blame the text to speech engine. I go that way. I get it. But, no. Surprisingly, the European accents have gotten a lot better over the past few years.

I I don't think they're quite perfect just yet. To my untrained, you know, American ear, they sound great. But, you know, to to you folks it might be, I I sound, you know, Northern English. Exactly.

Choose American. Shout out to Level Labs. That's fun. They do a great job. Yeah.

Okay. Very very cool. One thing I wanted to ask you is most people that I've spoken to, in hindsight, it feels simple when you not not not like they they don't have this complexity to deal with which is that you are a product company and a research company. I guess 11 labs have that but like I would imagine that how you run a research company is or like research division company. Research company.

Yeah. It's very different to like product.

Well, yes and no. It's very different in some ways. It's very very similar in others. Like in in spot on, we are a research company at the end of the day. Like that's what that's what keeps us alive and that's that's really what powers the vision of of what we're building here because, you know, the models and the tools to deliver the product that we're trying to do just don't exist in the world.

You you literally can't get them anywhere outside of, know, building them yourself. Yeah. So research is is our primary focus and where we spend a lot of our time. Now, when it comes to product philosophy, you know, on one hand, research is product. And with that, it's run very similar to a product org, right?

Where we have to understand very deeply what are the user's needs? What are they trying to accomplish? And how do we make that so simple to like for them to use that they don't even have to realize the crazy complex running under the hood where they can just say, oh, it it works. Right? And it's a it's a easy to use like magic moment for them.

Like, that's one of the goals. But on the flip side where it gets really interesting being more research oriented than than maybe some of some other companies out there is we have to almost tell users what's possible. Yeah. Right? We have to be thinking out, you know, several years on how do we push the boundary here.

And sure, like what do people want in like a general sense, not out of a product, but what do they want in a general sense and how do we deliver that? And Yeah. So what we landed on is people want you know, talking to a computer to feel as natural as a face to face conversation. Like, that's the interface that we should deliver. So then everything we do gears around that.

Right? And we we call it almost the Tavis Turing test internally where, you know, you have you have to be able to have a conversation with one of these digital humans and not realize it's AI. And once we do that, then we realize we have the product for folks. But I think being a research orb, we have to be really future focused in the sense of, hey, how do we not worry about necessarily putting Band Aids on something today versus building the product that will ultimately, like, deliver the best experience in the future? So it's it's planning, you know, months, years out.

It's making sure that we are, really pushing the boundaries of what's possible and thinking through that and, you know, ultimately like delivering the end experience that gives people what they want or at least what they think they want.

Yeah. And like in a small way I think I've seen this and just like in teams I've been in where like if like there's like the go to market team and they're like they want the customer said they really need this thing and it's like can we just get the research team to just solve this like thing? Yeah. Like, but then they're like, no, we need, you know, we're building AGI here. Come on, we need to

So our exam and and I think I think this is a good example. Our example of this would be turn taking. Like we got a lot of feedback early on when we first launched the conversational video interface that, hey, the the bot is responding too fast or too slow. Like either it's cutting someone off or it's not giving them enough time to think or it's giving them too much time and now it's an awkward back and forth. Because really matters in a human conversation is that, you know, like I just realized that yeah that you said is actually an affirmation, not interrupt.

It's that, you know, when I finished talking, you're almost cutting off the end of my sentence and it's a a snappy back and forth versus like a deep introspective. But I didn't quite realize that. So, like, the customer request at the time was we want interrupt sensitivity toggles, we want parameters, we wanna be able to control it. Whereas, to us, that wasn't actually the right answer. The right answer was, how do we solve the problem of helping, you know, AI communicate intelligently and and, you know, have intelligent So turn we actually built a model for it.

We call it Spero. It's a turn taking model and it's trained on a ton of data that helps us understand when a user is actually done with their sentence. Is it an affirmation or an interrupt? And that model, as as you can guess, is far more effective than these like sort of like rigid parameters that really help adapt to a user's flow of speech and and understand that. So that's like a prime example where, yes, a user wants something and they're giving you a feature request, but the reality is like, by deeply understanding that want and understanding where technology is going, like, the right decision in that case was to build a model for it.

Right. Or or to, you know, apply a research solution as opposed to putting a band aid on them.

So true. Like, because they don't even really care about turn taking

from No. No.

Turn taking. It wasn't an issue, no one would care.

If we if turn taking was perfect and we did not offer a single parameter for it, that would be awesome. Because no one wants to think about ASR and VAD and these things. They're doing it out of necessity. Yeah. Right?

And that's where, again, it keeps coming to like, we need to keep pushing the boundaries where things get better and better and better. We're like, ultimately, someone can just drop in a line of code and say, here's my digital human. I just got it ready. Like, people should be having to think about these in-depth things or what TTS or text to speech model is being used. Like, all they want at the end of the day is an experience which gives you the most natural conversation.

Right? True. And if you can deliver that, that's all that matters.

Although, I'll throw it back at you. You know, sometimes you speak to people, it's often someone that's quite intimidating. Yeah. You say something and they don't say anything and they laugh. And they just hold it and you're like you kind of like walk yourself around in circles Yep.

Like good negotiation tactics, stuff like that. Do you, like, I guess, like, do you have like do you do you hard code kind of like how how deep does personality get? Is it like does it come into it, that sort of thing?

Yeah. So our our belief is that two two things should happen in this case. The first thing is, like, our users should be able to inform the AI who it is, what its goal is in the context. Which means maybe you can say, be a really firm interviewer, throw them off their game, intimidate them. Maybe for some reason, maybe someone wants that, right?

Or if you're interviewing for Tavis, you know, our prompt might be be really friendly, help them be open, you know, make sure they they can be themselves in the interview. But so so first you can inform the personality where where it's at. But secondly, that's where the turn taking model comes in because the turn taking model's job is to understand in every specific like conversation and and instance, how does the user speak, right? Maybe are they a little on the older side and, you know, they're a little more thought out, they they take time to respond, they need time to think and time to process and we need to adapt to that. First like, hey, maybe it's again, like we said earlier, this quick snappy back and forth where we're building off of each other and we're having a heated debate or an argument.

And, you know, what makes humans humans is you can be put into any of those situations and know how to respond or know how to handle it. And at the same time, like, communicate in a way where your style is known and your expressions are telling a story within themselves. Whereas like, you know, if you're talking to a voice only application right now, like, you're not really getting that full story.

Right?

Even on the phone, like, when you call someone, you're probably cutting them off or cutting over them because, like, you don't really realize when the end of their sentence has finished.

Yes.

So I don't know if that, like, fully answers the question, but but that's a quick thought.

Yeah. But it I guess it's like the broader question is just that, you know, I think there was like this almost like meme where I was talking to I think there was like a week where was like constantly talking to Chatuchu of Tea. And then someone posted meme about, I think they replied to like Sal Almond and said like, that's so brave of you to say or something. And I was like, oh my god. It said that to me like a million times.

So this is just like, it's I don't think they realized they were like, okay, we're we're actually gonna like dial it back a little bit because it's a little bit too like affirming. And like,

this is just like my my personal perspective on it. This is this is outside the scope of of, you know, work. But I sort of want the AI to push back on me. Like No. Want it to tell me no.

Every time I like, you know, have a query or something like that, always replies down with, oh hell yeah. That's awesome. But it's like, no. Like I want you to like challenge my thought. Wanna build on on top of it.

Wanna argue and I wrote a little like your GPT four that where it responds in a different manner and it's a little more, you know, constructive as opposed to just affirming. I I it's scary to me because real humans aren't affirming. They they question, they ask, they dig deeper. And I think affirmation's a scary concept where you almost get used to it, or it becomes very one-sided. Mhmm.

So I think they'll they'll get much better at it that at it, excuse me, as time goes on. But I've totally seen the same thing and I I see that day to day. Even this morning I was, you know, using it to plan a trip or something and it was like, oh, that's a great itinerary. And I was like, no it's not. Yeah.

I could tell it's that. And is that like kind of like as a developer working with Telus, like, they're like, I don't know, let's say they're building like a customer sales rep startup or something. Like they're building they're able to like code up essentially with prompts different sales reps and like maybe dynamically assign like, oh, like I think that based on the interaction, they would do well with this sales rep or like this AI sales rep or something or like

Spot on. So let let's use sales and and to make it really literal, let's say you have, you know, three buyers. You have the person who's ready to buy, they just need to sign the You order have someone who needs a little convincing, and you have someone who's never heard of you before. Yeah. You could just create one persona that accommodates all of those and sort of adapts out the conversation.

Conversation. But But let's let's say say you you wanna wanna get really specific and they're completely separate. What you can do is you can create, you know, not only three different, we call them replicas, but the replica is like the clone of you or the digital human. So you can say, you know, hey, I want a replica of Jack. I want a replica of Quinn.

I want a replica of blank blank blank. And so you can create those different like replicas that actually visually look different. Yeah. But then you can create the different personas with that as well that give it the personality. So for, you know, stereotype one, it's, hey, we want you to just like be really to the point, make sure you don't have any questions, and get them pricing in next steps.

First, you know, number three, like, hey, spend a lot of time in discovery, learn their needs, you know

Yeah.

Ask these types of questions. And then when a developer's actually going to implement this, they can programmatically say, yep, we're gonna just send off a quick API call, here's the replica ID, here's the persona ID, boom, here's the conversation

Very cool.

Way to go.

Very cool. I'm trying to still understand like the complex of building this because it feels like very like different way to like Yeah. Build in like kind of like AI, almost like AI humans. If someone wanted to kind of have this like like, you know, on their docs page have like, you could talk to a dev. Right?

Like, that'll be kinda cool. Like, you could talk to a dev. You could ask some questions like, how do I integrate this? Like, what is it? How long will this take?

I don't know. Like, is this something that they should just like go to Tavis and like start building it themselves? Or do they like go build with like someone that's like kind of the intermediary between like Tavis and Yeah.

Yeah. It's a a terrific question. And we have, you know, a ton of folks who build on Tavis and build amazing products that are really easy to use. They put great UIs around them. They have really, you know, like strong data or tuning or prompting and give a really high alpha on on top of, you know, the models that that we've produced.

So, you know, whether that's, you know, like interviewing again or some of these L and D tools all the way to, you know, even someone was using it the other day for like a drive through like food and beverage thing. Like, there's some really great applications out there that are already powered by Tavis that are just easy to use out of the box and ready to what's really happening though and and what's been evolving over the past few months is that Tavis itself is becoming easier to use out of the box, right? Where you don't actually have to put in a ton of development time on top of it to get a, you know, ready to go digital human to serve your use case. So, you know, this morning for example, I was working on I I'm a huge foodie, like love restaurants, love cooking, it's one my favorite things. I was working on creating a digital human guide for my website.

Okay. You know, to walk someone through and give them food recs depending on what city they were in. That took me maybe thirty minutes to spin up. So it was, you know, quick, easy, ready to go. So I'd always recommend someone, you know, come on in, try to build it yourself first.

But if it's a mainstream use case where you want it almost like a SaaS tool Yeah. That's when it's it's best to go through, like, one of our our partners or someone who's already built on Tavis to to use it.

Yeah. Okay. That's very cool. I will and if anyone does work on it, definitely let us know. I'm very interested.

For sure. And I think by the end of the year, like what I'm most excited about too is by the end of this year, maybe early next year, I think that this will be much more of a DevTool. Where someone will be able to come in and actually spin up this digital human and, you know, have a conversation where they don't actually need to write code to do it. Right? Where it will be a lot more natural and and be a lot more accessible to to everyone out there.

Super super cool. I wanted to ask you if you

have any advice for DevTools founders, especially people that are building like that kind of like research slash product side of things? Because I think that's I mean, more and more, like, I know there's like AssemblyAI eleven Labs. Yeah. Like, more and more DevTools are like Deepgram. Deepgram.

Yeah.

Yeah. Great question. I mean, there's the traditional advice, right? You know, know your users really deeply, make your APIs easy to use and and, you know, docs are products, support us product, all of this stuff. And I'll I'll, skew away from that a little bit because I think those things are are talked about elsewhere and and have become more obvious over time.

The things that I have learned at Tavis that have been like new learnings for me or or things I wish I knew earlier, first, make it a magical experience. Right? Like, someone should have surprise and delight when they use your product for the first time. And I think, like, Linear is a great example of this. You log into Linear and you almost, like, wanna use it more.

Like, I made tickets for myself cause I'm like, it's so funny. Right? Oh yeah. In Tavis we try to do the same thing where I think you experienced it actually where you went on to the Tavis website, you started your first conversation and Charlie, the dude, he was like, hey, I like your black hat or your white shirt looks good. And you're woah, that that's crazy.

And you know, every part of that is magical and immersive. Right? You have, you know, experience where at the very end it's like, hey, I've gotta wrap up, I've gotta go. Right? It's not like an abrupt stop.

Right? You have, actually like intro music playing like a bum bum bum, like Yeah. With like sort of futuristic like suspense like right when it starts. And all these things come together to make this beautiful magical experience that really shapes someone's first perception of of what you do and and how you do it. So building something that really delivers a magical experience and having a great demo of that magical experience is is like so so important.

Especially in today's world where like, it's hard to understand like what people like what companies really do and the product they produce.

Yeah.

The second thing I I give advice around, and this is something that we need to do a little better at still, but support is product. Right? Like, really understanding the questions users are asking, getting them quick answers where, you know, it's very clear we're building with them, we're not supporting them. Right? Like, making sure that they know that we're building alongside of them, we're here to help.

And ultimately, we're part of their team. We're almost their internal AI team as opposed to, you know, they have to go through like layers of support team and answers to more. And that's somewhere where we need to keep doing better and better, but that's that's the second thing. And the third thing I would say is really just have fun with it. Like I think especially in today's day and age, people have this personality in products, right?

There's Yeah. It's sort of the end of, you know, every website is all blue with, you know, the same Webflow sections. Like, people want to understand the vision behind it. They want to understand the story and the why and the the, like, emotion that drives it. And again, going back to linear as an example, that's a big part of it.

Like, I remember using linear for the first time I read the manifesto. Right? And I was like, oh, I totally get it. This is the way that the world should work and I I subscribed to this. I agree with this.

Again, Tabas as an example, we have a mission statement online Yeah. Where, you know, our biggest power users like see that vision and they're like, yes. This is the world I wanna be living in. And of course, you need the quirky social media with it Yeah. And the brand.

But having fun and having a personality is so important.

Yeah. Yeah. I think that that makes sense to me. And I think like I I heard like I was listening to what you do described on a show and it was like really kind of they were talking about it in a really kind of like specific way of like how it solves like sales stuff. And I was like, this wasn't the vision that like I was excited about.

This is like and I think you have done a really good job of that like on the website where it says like, it's not just about like what the use cases, it's like the vision and it's like, I don't know like and yeah, you say the magic moments is like I've told so many people about it. And I just held like my I told like the guys in the studio and they were all like, wow. And then like I tell like random guys in the co working space like I'm so excited about it. And like it's like I think when you hit that, it's like, it probably takes care of a lot. You probably like do a lot of other stuff wrong if you just have got like the magic moment and like and a big vision.

I feel like it's like people wanna get on board and like get excited about it.

Exactly. They we're all changing the world together. Right? Like it's it's not enough to just build the product. Like people have to be using it and creating value with it.

I I think, we call it internally like teaching the art of the possible. Yeah. Right? Especially with with Tavis, that's very specific because we're almost creating a a ripple effect where someone builds on top of us and then, you know, distributes it in all these ways or provides value in all these ways where, you know, teaching the art of the possible is what's cool. So Yeah.

Whenever a new employee comes in at Tavis, they have a use case in mind that Yeah. They want to see, like, our technology, like, change the world with. Yeah. I mentioned earlier, mine is like the k through 12 education. Yeah.

Someone else on the team, it's, you know, when you get off the ride at Disney World, it's Tony Stark saying, you know, hey, thanks so much for riding the ride. Another person at Tavis, it's about, you know, ecommerce. Another person's, you know, in the Waymo, you can have this

Oh, yeah.

Personal assistant. Oh, dang. All these different things. And that's what makes it like so unique is everyone has their perception of how Yeah. It can be used.

And I think that speaks a lot about their backgrounds. The only other piece of advice I I just thought of this as well when we were talking about the team. Like, the other piece of advice I'd have, not just for dev founders, but for any founders is build a phenomenal team. Like, true truly. And and it's cliche to say, of course, right?

Obviously, you want a phenomenal team, but there's such a difference between good and great talent. Finding people who really care about what you do, like, will move mountains to accomplish it, take pride in their work is like something so special. My my cofounder coined this, and I've I've started using it a ton. But, like, what I love about the Tavis team is in a different world, everyone here would be, you know, craftsman or woman. They would be artists.

They would be poets. They would be woodworkers. They would be, you know, building machines. They would be these like, you know, craftsmen who take such pride in their work and and, you know, see it as art and see it as craft at the end of the day. They apply that same mentality to the product here, right, and to the research we do where, you know, it's not just a science, but it's really an art with that that craft involved.

And I think, like, that's what's so special about working with great people is, like, you see those differences in the day to day actions. You see those differences in how the product comes together. And I think, like, again, it it's becomes very clear, like, good versus great as you build those teams.

Yeah. Do you have any advice for finding those kind of people?

In some ways, those people find you. Right? I I think if if you're really mission driven and and visionary and, you know, you you sort of have your philosophies and your principles out in the open, like, it attracts a certain degree and talent, and and it attracts people who are more so passionate, right, about what you're doing. And that's irreplaceable. Yeah.

So, you know, some of the best folks we've have on board, like, they found us, they've applied, they've reached out. Right. So think I just like being very public again with how you're building is is awesome. But the other part is getting really like specific with who you're trying to find. Mhmm.

Right? And finding people from diverse backgrounds with creativity. Like, we don't really look for like a traditional background here where, you know, they've been at Google for two years and they've done a startup for two years. Like, we find people who are either fresh out of college or very senior and somewhere in the middle, it doesn't matter. But finding something where like a piece of their work is really interesting to us.

Yeah. It's

where we read their paper and we wanted to reach out to the author and it turned into a hire where, you know, like they've worked on something similar at a different company where it's like, oh, that's a really interesting piece of effective computing here. Like, clearly you're passionate. Maybe that can apply here. Then obviously like giving them the means in the interview process to like express that and to share that and to to understand our vision of what we're building. Yeah.

Right? It's it's a mutual, you know, relationship at the end of the day.

Yeah. That's that that makes sense. Yeah. Kind of just like attracting them being open minded. Yeah.

I well, just wondering here like riffing. I wonder if we could do like as a kind of before we close out like maybe I don't if there's stuff that we could do like where we say like, okay, this is this is AI Quinn speaking. That people can like, here and then we'll do like, I don't know, like, is this Yeah. That a stupid idea? No.

I know. I know. Know. Yeah.

Sweet. Yeah. That's great.

Yeah. Yeah. Okay. Amazing. We'll to we'll have to

pull some training footage and actually train it. Yeah.

But And then, could we do lip synchs? Yeah. We should be able Quinn as well. Yep. So we could say, okay, do wanna just say like

We could even do it at a random segment in the video and see if people noticed. Okay. Yeah.

Oh, yeah. That'd awesome. Okay. We'll talk back to it.

Yeah. We'll see if the training data is good enough because I've been like looking at you the whole as opposed to looking here. But I'll talk for a minute looking at the camera and we could actually see if we can get this training data. But I think it would be really cool to just include this at a random point in the video and see if people

noticed it. Like, that would be awesome. Okay.

It proves our thesis in the beginning.

Yeah. Yeah.

Or hopefully proves our thesis.

Okay.

Amazing. Okay. So hopefully we've done that. We'll see. We'll check the check the description.

And and remember remember the the point of it is not tricking people. The point of it is showing that like the only thing that matters is immersion. Yeah. Like the whole the whole ecosystem, the whole environment is is so immersive that like, the rest of it doesn't matter.

Yeah. Yeah. Completely agree. And that was like my thing to it. It wasn't like hiding that it was AI.

It was like Yep. This is better because it's AI. It would be it would suck if it wasn't AI. Spot on. Exactly.

Yeah. Super cool. Where can people learn more about Tavis?

So I'd always recommend just go on

Super super cool. Where can people learn more about Tavis?

So I'd always recommend just go on to the Tavis website, tavis.io, tavus.io. Talk to Charlie, have a conversation, have an experience. It's really eye opening. It's a really great way to just like understand what we do better. If you're interested in building with Tavis, there's a free plan where you can get a ton of credits, you can start building right away.

The docs are public and live. So just start fiddling around, hacking together, and see where it goes. And if I can ever be helpful, just reach out to me or our support team as well, and we're happy happy to spend some time building together.

Yeah. Amazing. Well, thank you so much Quinn and thanks everyone for listening.

Awesome. Well, Jack, thanks thanks for having me. Such a such a fun time and glad you could swing by and and check out the office as well.

It's a very very cool office. I feel like I'm seeing the last the last days of this one.

True. True. True. Fair.

Fair enough. We'll have No. A new office as Amazing. Thank you very much, Quinn.

Of course.