Navigating Startups: Jake Reichert on AI, Iteration, and Timing the Tech Market | Techstrong Unplugged EP32

Hey everybody, and welcome back to Techron Unplugged. I'm your host De Ta Solomon, and this is episode 32 of our series. In this episode, we're taking you to AI Dev Summit 2024.

There our cohost, Cassandra Chen met up with Jake Reichert and they discussed AI iteration and timing the tech market. Jake is a seasoned software developer and engineering leader with more than 20 years of experience. And this episode they'll discuss the strategic decisions that drive innovation at major tech companies as well as the intricacies of launching new products.

Without further ado, let's it over to AI Dev Summit 2024. Welcome back to Text on Unplugged. My name is Cassandra Chin, and today we're here with Jake.

Can you introduce yourself? Uh, sure. Yeah.

My name is Jake Reichert and I am a software, uh, developer and, uh, engineering leader. And, um, I've been a developer and, uh, manager for 20 plus years at this point in anywhere from getting startups off the ground from a handful of people up to, uh, managing teams at Amazon. Do you wanna talk about, like, maybe how you got started into technology?

Yeah, so, um, my background originally was in math and physics. Um, and at the time that I was in school, um, there was still a lot of the internet that was being figured out and built. Um, and I realized pretty quickly that was what I wanted to be working on.

Um, I wanted to be a part of figuring out how to solve some of those problems and just kind of blazing this trail an area nobody really knew. It's, it was, um, not that dissimilar from the landscape now with, with ai, with that same kind of, uh, level of excitement and passion around it that people had. So, um, I went to school in, um, San Francisco, and that was clearly a place where there's a lot of energy around that.

Um, so just being around it all the time, I was, um, very interested to be a part of it. And along like your journey, like have you been to like startups and lots of different companies? Yeah, so I've, I've worked at, I would say predominantly I've worked at startups that are anywhere from, you know, 30 up to 200 people.

Um, but I've also worked at places where that are larger and smaller. Um, at the, uh, large extent I worked at, um, Amazon, uh, I was part of Amazon Music. I was managing a lot of the teams doing front end development, um, for, for, uh, pretty much everything that was not voice.

Um, so web browsers, phones, uh, TV sets, you name it. Um, and at the other end of it I've been involved with companies where I was, uh, where I was one of the co-founders. Um, starting, starting from that level and building up from there.

It's a lot of interesting things you've done. Yeah, I would say, um, you know, there's, one of the things that's interesting in to me is, you know, I've been predominantly in the startup world. Um, I, I went to Amazon mostly because I wanted to learn things that I felt like maybe I couldn't learn those at a small organization, but I wanted to take those lessons and apply them, you know, elsewhere.

Um, but uh, for me there's always kind of been this real excitement to, um, working in this startup environment because, you know, there, there is of course a high risk of failure, but there's also just this level of excitement that it's hard to find in, um, some other positions. Um, uh, there's, there's people are, you're usually not working at a startup because you're not excited about it. You know, like everybody there is usually a hundred percent behind whatever the company's mission is and really trying to, to, to make that thing a reality.

And sometimes you succeed, sometimes you fail, but the, the, the journey in it, I think is probably more important than the, than kind of the end result. It's a really different environment when everyone's really passionate for what they're trying to achieve. Yeah, for sure.

You know, there's, and, and also where, where people might have, uh, you know, particular expertise in particular areas that they're really looking to, um, apply. Um, early on, I remember one of the places that I worked at was a startup, um, that, um, uh, they never made it, you never, never would've heard of it, but it was, um, building the first, um, uh, chat software, like voice chat software over the internet. And this is when still there was, everybody was on dial up modems.

Um, and there was a guy in France who was our CCTO, who basically his background was in, um, signal processing, um, and, and data compression, which was pretty much what you're doing with, with speech recognition. So he had come up with this idea for how to, how to build this and, and, and, and even though most people at the time felt like, well, there's not enough bandwidth when you go outside of a local network, you know, over the, over the wider internet, um, to make that work. Um, he felt like, no, I, I, I feel like I have a plan to make that work.

Um, um, I was a junior developer that at that point, so I came in, basically took a lot of direction from him. I learned a lot in the process, but, um, uh, you know, seeing that actually work, I remember, I still remember the day when we started having our, our meetings, uh, over the internet and being able to talk through it and, you know, it's compared to today, it was like, it was scratchy. It was, there's a lot of, you know, it breaking up, but it was just like that factor of like, I can't believe this actually works.

It, it, it is, it is that same feeling that you have when you're like talking to, you know, Chad GBT, and it's like, how does it do this? It was that same level of kind of like feeling like magic. So it's not entirely polished, but it's really new and it feels magical.

Yeah, I mean, I think it's, it's, uh, I mean, one of the things that I think is, is, uh, true at any startup is, uh, you can't really be a perfectionist, right? You have to be willing to build something, put it out there, figure out what people like about it, really take that feedback into consideration and then quickly do something with it, right? Um, which basically means you can't try to, you don't, you're not trying to build a finished product right out the gate because, um, it might not be the right one.

Like, you have to actually put something out there and just like, that's enough for your customers to be able to say, is this, is this good enough? Is it not good enough? Um, my current company, um, that I, uh, co I'm a co-founder at s ai, we're building real-time conversational avatars, and we've gone through probably, you know, seven iterations in the last nine months of kind of what the, what the core of the product looked like.

Um, you know, who was the audience? Was it more like large businesses? Was it consumer end users?

Um, and a lot of that was just seeing the usage patterns when we put it out there. Like what are people using it for? Um, I'll give one example is that, um, you know, we, we'd come out with it originally thinking we're, this is gonna be really interesting to, uh, retailers as like a, a virtual customer assistant, um, agent.

Um, and we have gotten interested in that, but the place we've gotten, the moat, a a surprising amount of interest is from, um, consumer end users who are, who are paying to use it as a conversational partner for learning English. Um, which is not something we ever designed it for. Even, you know, we had our list of, of 20 things that we thought, oh, here are all the things that people might use this for.

That wasn't anywhere on that list. Like, we're as surprised as anybody else. But instead of just saying, huh, that's curious and ignoring it, we basically said, oh, okay, well if that's what people are actually willing to pay for, like, let's do more of that.

Like, let's figure out what, let's find out what is it that those people are interested in. You know, that could be just like sending out a, you know, a, a survey to, to them and saying like, what did you think of this and this, you know, what are the top five features you'd like to see in this? Um, and then using that to shape the next kind of iteration of development.

Um, and on the flip side, you have to be willing to throw out what you think are like your, your best ideas if people don't actually care about them, right? Um, you know, you can, you can work on these things all day long and you think are like the coolest CS or product feature kind of ever, but if nobody's actually interested in using that, um, from the perspective of building a business, it's not really worth putting a lot of time into. That's really difficult.

You have to reprioritize your ideas and like, uh, really listen to user feedback. You do. And it's, um, it's, I think a, um, a lot of people, their first instinct is to reject that feedback.

Um, even people who are kind of seasoned professionals, you know, if somebody gives you, uh, negative feedback about your product, it's very easy to get defensive about that. Um, and I, and I think it, it, it takes a, a while to get used to the idea of like, well, they're not putting you down. It's not about you personally.

You know, they're not saying that like you're somebody who comes up with bad ideas. They're just saying like, I don't, this isn't useful for me the way that it is now. You do have to weigh that off against, you know, like, don't, you can't like jump on one piece of feedback from one person and say like, oh, well that's a, we'd never thought of that, and then change your whole strategy, right?

But if you start to see that pattern, you know, I don't, I, I don't necessarily mean from like, you know, tens of thousands of users, you know, we, we usually are able to make pretty good decisions based on feedback from like, you know, a dozen, two dozen users. You know, you can see enough commonality in there. It's like, you know, it, it's not, it's probably not strictly speaking statistically significant, but it's enough that it's like, it's probably right.

You're directionally probably right if you're like, uh, five of those people said this, and that feels right. You know, so let's, let's kind of do things in that direction. And then there's other things that you have to listen to and say, yep, thanks for your feedback.

It's not the direction we're going in with this. It's either, you know, just like, not our expertise, not the company we are trying to build. Maybe somebody else will.

Um, you know, or just like, we're not excited about going in that direction. Um, before you talked about the use case for language, like, like, is that company still like going in that direction now With No, that company, um, was a, um, it was interesting, and I think we're gonna see a lot of this in the AI landscape as well. It's kind of a victim of, of timing.

Um, the, the, the infrastructure just wasn't there, or maybe the need wasn't quite there at the time that we were building it. Um, we thought it was pretty great, and we found a few companies who were, thought it was pretty great as well. Um, but at the end of the day, there wasn't, there was, there was still a, um, there wasn't a compelling use case for it of people saying, well, why do I need to talk to people over the internet?

You know? Um, now, now today, it's kind of obvious what some of those things might be, but at the time it was, um, I think that we were spending a lot of our time in customer meetings trying to convince people why they should use, why they would need something like this. Um, and in my experience, it's, um, it's, it's much easier to sell things into customers when it's like you're solving an existing need than trying to create a need and then telling people they should, they should, like, be interested in that, you know?

And that's not impossible. I mean, that's essentially what Facebook did, right? It's like, it's not like everybody in the world was on a social network, you know, they were looking for, you know, the next best social network.

When Facebook came out, they basically created this platform and people then like saw this thing that they wanted that they'd never been on, you know, never been on before. Um, same thing with, you know, just as simple as like early broadband providers, right? Like, why do I need this, you know, to pay all this money for a faster internet connection.

Like, I could check my email just fine, right? But like, you know, that, that is one approach to it. But for consumer products, I think it is, um, typically easier to, um, get people interested in something that like, where there's like, here's a need and we're gonna fill it.

So that one, you know, that was, that was an example of where a few years later, um, you know, Skype took off, you know, of course they had video in it as well. And, and the, the bandwidth was there to support that, which we didn't have available to us at the time. Um, you know, an example I think of a lot is, I had a friend who worked at a company who, that they were, um, they had this idea of building a portal where you could, you know, upload video of like, you know, your friends and family are just like short movies that you made yourself.

And like, then like, have it so that anybody in the world could like go and watch them. And, you know, sounds a lot like YouTube, you know, but it was like two years before YouTube and they didn't make it because same thing, like the, the broadband adoption wasn't there. And so the process of actually watching the videos was just like too slow.

And so, had it been two years later probably would've been a great idea. Um, but sometimes there's an element of timing that goes into it as well, that you just, you know, you, you, you have to take your best shot. You can't really factor that in, you know, and there's no timing the market there.

You just have to do it and see if it's, if, if it's, uh, people are gonna use it then or not. Yeah. That's very tricky when we're talking about timings, Right?

Right. And always hear the phrase like, don't time the market in stocks, uhhuh. Yeah.

I mean, it's a little bit like that, right? You, you, um, I mean, you do have to be thoughtful about what's possible with the technology that's out there, you know, so like can, as, as an example at, at Sage, and one of the things that we struggled with early on was the, um, the responsiveness. You know, you would say something to these 3D avatars and it would be this pause of, it started out fairly long, like 10 to 12 seconds.

It was down to four seconds, but even at four seconds, it felt like a little bit long. Um, and, and, uh, we got some sort of expert advice from a people, you know, in the, the AI space working on these topics in particular. And they did some research said, yeah, based on, you know, what we see out there, we don't think this is possible.

And we looked at it and said, well, we don't agree, we think it is. And we were able to figure out pull, you know, not using any tricks related to making the response from the LLM faster, which is kinda what they were looking at, but looking at making all the stuff around it faster, you know? So, uh, I mean, we did all kinds of things like, like, uh, introduced, uh, uh, message-based architecture.

So, you know, you could ask questions, have the responses come back, kind of, um, asynchronously, you know, if the response is gonna be five sentences, have your, uh, text to speech renderer, you know, process all five sentences at once, and then send them back. It gave the illusion, and then we would sequence them back in the right order when it would speak it back. That would then give the impression to the, to the end user that it was faster than it was, or one thing we did, which was kind of the audio equivalent of, you know, when you have see spinners or like a waiting button when you press, press a button to like submit a credit card number or something like that.

Um, which was, you would ask a question and then we'd have the stock phrase that would say like, Hmm, that's an interesting question, or, let me think about that. Um, and we'd throw those in every once in a while if the weight was starting to feel too long. Um, and then that gave the illusion that there was less of a weight than there actually was, so that we're trying to use as much as we can to our advantage of what's out there today.

Um, but a lot of that was us really thinking about what is, what, what do we think is possible with the given state of technology? Um, but then there's certain other things that we've said, you know what, that there x, Y, and Z could be cool things we could do, but not now, because we actually don't think the technology is there to do that today, even if we really push the envelope. It's just, there's, there's too many other problems that need to get solved first that we ourselves don't know how to solve.

Um, and so those are things we're just, we're not gonna make a gamble on those ones for right now. It definitely makes sense. There's a lot of strategy involved when handling startups and what to invest technologies.

Yeah. I mean, you're, you're, as opposed to a large company, you're always going to be struggling for resources, both, both human resources and financial resources. And so you really have to prioritize what's the most important.

Um, and that can be harder, that then you think, you know, it's not an obvious thing, especially as the priorities keep changing depending on the feedback that you're getting. Um, and a lot of it is just sort of going with your gut instinct, you know, or your experience, um, what you've seen work in the past, what you haven't. Um, but yeah, you can't, um, there's not gonna be enough time to work on everything that you think is interesting.

So you really do need to get really, really good at both prioritizing, um, like what order are you gonna work on. And also, um, and this is a hard one for a lot of people, is failing fast, right? You know, if you start, if you work on something and you think it's awesome, but you put it out there and people hate it, um, you know, don't double down on a bad mistake, right?

Like, you don't wanna say like, well, okay, but if we only did these five things, then maybe people really like it. And then you spend the time to do those five things and people still don't like it, and you're like, well, but they liked it a little bit more, so now I'm gonna do these I things, and the next thing you know, that's sucking all your resources for something that like you kind of knew upfront nobody was really that interested in, right? So that's something where you have to get good at recognizing when that's the case and say, I'm gonna shelve that for now.

Let me move on to these other things here that we think have maybe some, some more potential to try out A lot of strategic decisions. Uh, yeah. Yeah.

So that, that's an example. Um, uh, also just where to, you know, allocate your money. That's of course an important one too, right?

Um, you know, there's, there's certain things where, you know, if you're, if you're, you're experienced in space, you can figure out how do you get better deals on things, you know, how can you get free credits from, you know, Amazon or Google or, you know, whatever companies, right? But then there at some point, you, you, there's certain things you just have to pay for, and you'd say that you're not, you're never gonna have as much money as you wish you did to pay for all these things. You have to kind of decide where are the things that are the most important.

Like, if we don't buy this thing, we don't have a product, versus well, if we don't buy this thing, we have to build it ourselves and maintain it ourselves, and that's gonna take more time, but it's not gonna take half of our, you know, existing budget, right? So, um, you know, you have a lot of those things you have to think through as well. I think I've learned a lot today, so thank you, Jake.

Yeah, thank you. Happy to be here.