Are AI Phone Agents Ready for Prime Time? | Alex Levin, CEO of Regal

Alex Levin is the cofounder and CEO of Regal, a platform for AI phone agents. They've raised $82M from amazing investors such as Emergence Capital.

Alex's favorite book: The PayPal Wars (Author: Eric M. Jackson)

(00:01) Introduction
(02:37) Evolution of customer contact tools and legacy players
(06:02) Launching Regal: Origin story and early challenges
(08:41) MVP strategy and problems worth solving
(11:46) Lessons from 0 to 10 customers: Growth mistakes and hiring
(16:13) Ideal early-stage team construction and hiring philosophy
(19:06) Sequencing hires as company scales
(20:58) What makes a good investor and how to leverage them
(25:42) Best and worst experiments while building Regal
(29:04) Internal use of AI at Regal across teams
(31:49) The future of AI phone agents and near-term blockers
(34:13) Rapid Fire Round

--------
Where to find Alex Levin: 

LinkedIn: https://www.linkedin.com/in/alexlevin1/

--------
Where to find Prateek Joshi: 

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 
X: https://x.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.747)
Alex, thank you so much for joining me today.

Alex Levin (00:04.546)
Thank you for having me.

Prateek Joshi (00:06.537)
Let's start with the basics of AI phone agents. Phone agents have existed for a long time, for as long as we can remember. If you want something, pick up the phone, you call, and you're there for hours sometimes. So now, before we get into what AI can do, can you talk about what are all the tasks that a phone agent is expected to do well?

Alex Levin (00:33.58)
Yeah, so today there's about 16 million people that work in contact centers. I'm not even going to talk about how many, or maybe I'll tell you the number, it's crazy. a typical American spends about 43 days of their life on hold, even waiting to talk with those people. And then there's another tens of billions of hours talking with those people every year. So it's actually a pretty large part of your life, either waiting for or talking with these agents. I'd say...

the majority of that somewhere between 50 and 70 % of the communication is via phone today. And that's with every brand playing hide the phone number and doing everything they can to move you away from phone because it's the most expensive channel. So despite it being something they're trying to eliminate, still it's over half of the interaction. I think that tells you sort of how much people wanna talk on the phone to your point sometimes for hours.

And what's happening, the large majority of that, 80, 90 % of that is customer support related, tends to be customers calling inbound into a brand. In retail, the most common things are, where my package, I want a refund. In more complicated industries, health care, insurance, lending, services, it may be things about a specific appointment or booking or something about your account with that company that you're talking with about with.

the 10 or 20 % is more what we'd call sales or sort of lead engagement. So it's a new customer, either they're calling in or you're calling out to them. Think of home services as a classic example. And you're saying, hey, let's get your kitchen remodeled and let's talk about what you need and how much that would cost. So that's the majority of the interactions. There are also some sort of operations use cases, which are not always counted in that 16 million seats, which might be things like.

payments or logistics or things like that, which are very common use cases as well.

Prateek Joshi (02:37.042)
And if you look at the current landscape of the tools available to do this well, can you talk about how this has evolved over time? Where did we start and what's the landscape today and where are the gaps?

Alex Levin (02:52.888)
Yeah, of course. So my background is that I used to manage a 3,000 person contact center. So I've sort of lived this, if you will. And I remember, for a long time, I was taught just do everything you can to eliminate every customer reaction. That was the ethos. And so that's why as a customer, you feel like these brands don't care about you because they're trying to eliminate those interactions.

Prateek Joshi (03:03.518)
Hahaha

Alex Levin (03:21.05)
And what, know, the main playbook was sort of build more self-serve features and move more customers to tech space channels, whether it's chat or email. But even then, you know, whatever, five, 10 years ago when I was doing this, you know, you'd have vendors come in saying, we can automate this with a, in those days, either you called it an IVA interactive virtual agent, or some of them actually called it like natural language processing and like automated agents, not quite AI, but in that vein.

The players in sort of those legacy players were Replicant, PolyAI, ASAP. Those are probably the three best known. And I'd say the reason I never ended up doing it is one, there were enormous costs upfront because they were building these customs models and you had to give them all this data and they trained their models and it's very expensive and even ongoing, quite expensive. And two, just it wasn't that good. Like even when you did it, it just didn't sound that great. And what changed is about, you know,

Two years ago now, everyone started to experience LLMs that were at a place where they truly could understand what you were saying to them. They could respond in a conversational way. They could reason in a way that humans might. maybe about 18 months ago, we started being able to make demos where really combining the LLM reasoning with the new modern voices that were available. So text to speech really is what it's called. We were able to make demos that were quite convincing. And we go, my goodness, like this is

good enough that I would use it. And the more interesting part is not only is it good enough, but you don't have to go and spend hundreds of thousands of dollars upfront training some custom models. Like we can just go start turning customers on at 10 to 20 cents a minute and that's it. And that's a huge unlock. So forget the legacy players, forget replicant, poly AI, Sierra, sorry, and not Sierra and ASAP. Like now you have a new generation of companies that are built on this modern infrastructure on the sort of

the developer side, you have sort of retail and VAPI that are very well known. And on the sort of more enterprise side, I'd say we see most often probably Sierra, sometimes one or two other players, like that sort of much more professional services, traditional enterprise sales. And now you've got this like fascinating market where, I'd say went from at the end of last year, people kind of being a little bit interested in this to now everybody's talking about how voice is the future again. And if you remember at the beginning, I was saying,

Alex Levin (05:47.042)
Every company was eliminating voice. Now people are going, wait a second, at 10 cents a minute, I want voice everywhere. Like this is my cheapest channel. So put the phone number everywhere, let customers talk with us anytime. And it's completely changing the trajectory of how brands engage with customers.

Prateek Joshi (06:02.541)
That's amazing. Let's go back to the launch of Regal. Now, obviously you are in the industry and you ran the contact center, so you know, but because it's a very famous problem, which means many people would have tried to attack this from different angles. So how did you validate the need before launching Regal?

Alex Levin (06:29.238)
Yeah. So when we started, honestly, AI agents were not good enough. Like that wasn't the main product we were selling. You know, we started in a simpler place or maybe more complicated place, let's just say a different place, which was, you know, in a context center, they don't have the tools that are necessary to make sure each, each conversation with the customers is going the right way. you know, so in marketing, you have things like braves and interval in, sales, you have all kinds of tools and product. have all kinds of optimizely and AB testing tools.

You don't have those sorts of tools in the contact center. Like literally the software they're using are older than their agents. So it's getting pretty long in the tooth. So, you know, that was where we started and we built Regal to connect with our customer data systems, allow drag and drop AB testing, omnichannel orchestration for inbound and outbound. And largely we had to allow human agents to do things. So the human agents either sat in Regal or they sat in their current contact center software, but it was still human agent doing the thing.

And we could give advice and scripts and whatever, but like that was always the weakest link. Not because there's anything wrong with humans. just, it was the hardest to get the human to do the thing in real time that we needed. And it was hard for a human going from call to call to call to take all that input and that feedback and have a good conversation. So, we got to a couple hundred customers that way and a good business. But when we saw the sort of advancements in voice AI agents, particularly we went, wait a second, like.

Why are we convincing everyone to try to like get humans to do the right thing? Let's just walk in and say the agent will do it. And that's been a big unlock because it means we don't need to rip and replace their current software around their human contact center at all. Don't worry, don't touch it, leave it there. know, it's now by, it's actually exactly the same platform. So we have a five years of building this very sophisticated platform, but it's just a human agent that's doing the end of it. So I think what's happened, it's very fascinating is

in the same 18 month period when there'd been a lot of companies that built voice agents. Sure, we also built a voice agent, but it was a top of five years of platform work, which allow us to be much more sophisticated in sending texts, emails, doing decisioning, know, engage, you know, doing post-call analysis, things that most of the players in this space don't have because they're only 18 months old.

Prateek Joshi (08:41.074)
And once you saw all this, now obviously you can't build everything before you launch the MVP. So can you talk about how you decided what should go to the MVP? What did your MVP look like and also the decision making behind

Alex Levin (08:59.758)
Yeah, I mean, we're extreme in that, this, I've, you know, I have friends who will spend a year building something before they show it to anybody. And I just don't believe in that way of building. You know, we basically like, we'll literally make a deck with some slides and some images and like, we'll make a demo like, so they can engage in there and say, cool, give us your feedback. Like, you know, where was it good? Where was it not good? What, you know, what would you need to turn this into production? Like how much would you pay for it? And like, that's our MVP is like something that we've strung together with like.

whatever existing systems there are now very quickly, like, you know, as a founder, you learn to tell what's the difference between somebody sort of being nice and saying, that was good versus somebody like reaching across the table and going that I want to pay for that thing tomorrow, like turn it off. And when you're getting the ladder, so if you're getting the form or forget it, it's not a product, but if you're getting the ladder, like that's where, you know, you do want to double down. And so from our side, we really focused on how do we turn this into something that is production ready for very large enterprises. So.

we pretty quickly went from like an MVP to like making a real bet on voice AI agents, particularly. We support other channels, but voice is really the most important. And now I think it's about something a little bit different, which is there are some things which are very easy to do. Like you can use existing LMS and existing voices quite easily using PypeCAD or LiveKit or things like that. But then there are things that are very hard. as an example,

Every time you build a new AI agent, you can't just test it once. It's not deterministic, right? It's stochastic. You to test it hundreds of times. No one has a good solution for that. None of the validation platforms, evaluation platforms, none of the other, nobody does. And so how the first person will really solve that problem in a material way is going to have a huge leg up. Or as an example in RAG, everyone sort of access the RAG now through pine cone and different databases. But

What happens when there's two different articles that say the opposite thing and there's a collision? That's quite hard for a human even, and let alone for AI to figure out, I blend the two? Which one wins? Why? How do I update it? How do I make a decision? There's formulaic versions of solving that, but more holistic versions of solving that rag problem are very material. Similar with, I don't know, there's 20 or 30 other real problems that no one has solved.

Alex Levin (11:23.712)
A lot of the like interesting work now in sort of MVPs isn't on the actual agent itself. It's on these very hard problems that are like difficult to solve and like figuring out like if there's a better way to do it. And like I said, all of the companies are working on this because it'll the winner will take a lot of business, but no one really has figured it out yet.

Prateek Joshi (11:46.463)
Now, you had the MVP. Going from zero to 10 is always a grind and it's different for different companies. So looking back at that phase of your journey, like what are the learnings going from zero to 10? Looking back, what did you get right? What did it get wrong? And if we had to do it again, how would you do it?

Alex Levin (12:06.818)
Yeah, honestly, for us, zero to 10 was quite easy. Like was actually after 10 that it became much more complicated because we had products pretty early that sold very well. I think actually like the problem was that we sort of didn't lock down the ICP enough. And so it wasn't until we got to 10 where we go, like actually we probably shouldn't work with these sites sorts of customers because in the long run,

It's not sticky revenue or we don't think we can do more with them or we don't think it's the product we're selling them is highly differentiated. So I think you're going to see that we learned our lesson, what, over five years ago as we were just starting with the sort of more data oriented product, the journey building product. I think you're going to see it a lot in the AI world today where people go zero to 10, they go, I'm so fast growing. And then they're going to find out actually like 90 % of that was either trial or wrong sorts of customers or completely commoditized things they were offering.

where somebody is just going to switch the next day and their churn is going to be terrible. I think the important part for companies that are, if you're growing at that rate, it's a different story. Let's say you are growing at that rate. It's actually like, how do you massively slow it down? If it's happening too fast, something's wrong, which is like actually a weird perspective to have. if it's too easy to come, it's too easy to go. So you know,

Figure out like which of the customers you really want to focus on. Figure out how you create a much stickier product. Figure out like what you're going to do, not necessarily today, but over time to make those contracts last longer. So as an example, we started at the beginning doing monthly contracts and then annual and then multi-year and then, you know, with larger and larger commitments so that, you know, we really were locking customers in for longer. And I think like too many people are not sort of taking advantage of.

those sorts of exercises to make sure that they do have sort of good revenue retention. And then I'd say the other thing that we learned always is when you're going through that growth phase, where we made mistakes was on the team side, hiring people that we knew had gaps. So we'd say, we want X-roll. And we look and look and look and couldn't find the person, find, this person's good enough, them.

Alex Levin (14:19.422)
never make that decision. Like it's always the worst decision. You always come to regret it later. And so it's actually nice. Like I remember like one of my pet peeves is when people used to ask me how many employees you have. And I go, what does it matter? That's not a sign of us being successful. If anything, that's a sign of us not being successful because we have too many people. It's nice to finally see that actually entering the common conversation where people are going, no, they're like the companies that are staying very small with great people and being really, really careful about who they're hiring are smarter.

And the revenue per employee is the number we should be watching, not just growth rate. And so I actually really liked that that shift has started to happen.

Prateek Joshi (14:57.593)
Actually, we need to, in the AI world, we need to find a way to be couple experimental ARR from actual real ARR. Because as you said, people, so many people are just experimenting and they want to know, Hey, I want to test a bunch of tools and something will work. Most of them won't. And then I'll, I'll

commit to a multi-year deal with that one company. So I think that's a very good point that you mentioned. And also on the team side, yeah, I think it happens so often that you just want to tell people that we are growing and headcount good or bad, bad usually. People use that as a proxy for growth and the higher they make these suboptimal decisions and yeah, it always bites back. right, so now moving to company building for a second.

Obviously, now you're bigger, but looking back, you have building a company, what kind of teammate profile would you look for, especially in the early days? And let's say we're going from zero to 10 teammates. What's the construction of the team? And if you were to guide a young founder who's building something similar, let's say enterprise AI product, how would you guide them?

Alex Levin (16:13.356)
Yeah. So I mean, a few, a few concepts which are not that new, but are a good starting point. So one is have a co-founder. This is hard enough to do. Like don't make it any harder by doing it by yourself. They don't have to be technical, non-technical. Forget those things that you've heard. That's nonsense. You have to have a person you trust. Like don't go on Reddit and like find somebody tomorrow. Like you want somebody you've spent years with where you understand them. And when things get tough, they're not going to like up and leave, you know, you're going to be able to lean on them. That is so, so important. Second, I think.

very early on, like obviously the founder should be doing everything, especially selling. But one of the things I recommend to people is get like a BDR plus pretty early. Once you're seeing a little bit of traction, it'll just give you more leverage. Like there were weeks at the beginning where like I had to go do something else. And like I didn't prospect or talk with customers for a week. And that was just, especially at that stage, that was so painful. Like literally you'd see like weeks later, the revenue dip because of that months later. So get yourself a BDR plus who can shadow everything you're doing, help you write emails, help you.

It can all come from your email address, not theirs, but just have that help. And then I think very quickly, you know, once you have like an engineering team and like the founder plus a BDR plus selling, you should start asking the question that I think it's the Shopify CEO that's, you know, put this in his, in his letter recently, but for every role, you know, could AI do this? If I had an autonomous AI, you know, could it do it? And before I go and hire anybody, like check that first, because these are new technologies and a lot of people who

aren't used to them, aren't even aware what's possible. hey, instead of a certain designer, instead of a certain engineer, instead of a certain whatever, maybe you need one engineer, but do you need three or can one do more with the tool or maybe one designer, but do you need three? Like that sort of is a big shift where instead of looking at AI just as a tool, look at them as a teammate and start thinking about it in the headcount discussion where it becomes part of headcount planning. So I think that is more more important.

Prateek Joshi (17:46.209)
Thank

Alex Levin (18:08.536)
You know, from there, like I said, just make sure you're clear on what you want and don't sacrifice. You know, it's not worth it. Like, you know, if you're wrong about the thing that you need, that's a different point. like Aaron Hoffman is one of our investors and he's always sort of good about this stuff where, you know, I think his point is like, make sure you're focused on the stuff that makes you differentiated. And, you know, he more recently started talking about like, you know, don't buy big executives, rent them. like focus on like, if you're really going to do

X feature really well and it's going to be different than anybody, focus there. And instead of like going to get the most expensive person who will never come to you anyway, get somebody who's going to be really good at building a day-to-day and find an advisor or three advisors who are the ones that are the very expensive people and give them little equity or cash. And that'll be a much better setup. So don't waste six months going to try and find the more senior person or don't waste time on hiring people who are going to work on problems that are already solved by 90 other companies. Focus on the thing that is most differentiated.

Prateek Joshi (19:06.721)
And as you're building the company, let's say you get a handful of customers, you are seeing something is starting to work. How would you sequence the hires? Meaning there is not the sales, marketing, VDR, engineering, design, product, so many things. And many times founders get zero customers want to have VPS sales. Like it's not, it's not going to work because what are they going to do? So how would you guide them in sequencing these hires?

Alex Levin (19:29.635)
Yeah.

Yeah, look, it depends again on what your differentiation is going to be. Like if you're going to be the company, like, you know, if I think of ASAP as an extreme example, they said we're going to go after the top 20 companies by size in the world. You don't do that by just like MVP-ing your way there, right? You go and get the right board, you go and get the right salespeople, and you do maybe hire that those people even at zero revenue because that's your strategy. But let's say that's not your strategy and it's more middle of the road, like...

probably should be like for every role, like the founder or at least somebody in that sort of founding crew doing the role first, understanding it, you know, then hiring somebody to do the day to day. And then hopefully you hire people who are high growth and they can move into the management roles. I think where it gets hard, honestly, is where the company's growing too fast, where the people that you hire that you think are talented have only been in that role three months. They're not ready to be the manager yet, but you need a manager because now you have six people in that role.

That's where it's tough because if you're damned if you do, damned if you don't. If you put that three month person in the role, they're going to make all kinds of mistakes, which maybe you're okay with. But if you go and hire the right person, it's not zero revenue more. You have real revenue. You hire the right VP. They're not going have any of the context of the IC role and they're going to have to like try to jump on as the buses go in and like make changes. So both are hard and you just have to know which problem you're signing yourself up for. I don't know that there's a right or wrong.

Prateek Joshi (20:58.697)
And earlier you mentioned Arn Hoffman, but in general, as you raise venture capital, what does the ideal investor look like? And again, you don't need to choose a person, but mostly like, in what ways have you found your board or your investors to be useful to you?

Alex Levin (21:14.893)
Yeah.

so it's two slightly different questions, but I'll answer them both. look, I think that the ideal investor is one who knows that this at this specific fund is the last job they'll ever have. That's a very interesting concept to think about. That means that they're at a fund where they have enough control that they're going to be the GP and stay forever, that they don't want to go somewhere else, that this is their legacy, that this is it. There's not that many funds like that.

I have lot of friends in multi-stage funds, but just to say multi-stage funds are not that. The big multi-stage bulge bracket funds are not the last job those investors are going to have, other than maybe one or two of the founders. So less like the founders literally joining your board, like they're not in this category. People in this category would be, let's say, early stage would be like a homebrew, it's Satya Patel and Hunter Wok, where they have no desire for this to follow. It's theirs. It's them. There's nobody else.

Again, in sort of later stage, emergence would be a good example where, again, if you want to be a partner at Emergence, you have to join as an associate, spend years training, and then you eventually become a partner, and that's your family for life. There's nothing else. So I think funds like that, if you get the opportunity to work with them, are unique for founders because you know those people are dedicated to you no matter what. They're not going anywhere. They're super high conviction. It's the best. As long as you want sort of a relationship with your investors, let's say it's the best.

Now, on the other hand, if you don't want that, if you just want money, there are plenty of funds who, let's be honest, are just money and are going to place 1,000 bets and are not going to care about what you do or not necessarily going to be there when the times are down, but maybe that's what some people are looking for. So just know what you're looking for and make sure you're signing up for the right thing. In terms of the things where think investors are very helpful, a couple areas.

Alex Levin (23:11.56)
I mean, the obvious one is, as you're going through and you need help fundraising, making sure you have people who can help you, like literally say, here are the five investors I'm gonna introduce you for the next round. And like of those five, all five get you through to like a real evaluation. Whether they invest or not, it's beside the point, but like that your earlier stage investor can literally put you in front of the right people is super, super important. I'd say the other one is having somebody who knows enough about those

the place you're in so they can help you benchmark and understand. So, you know, what are good growth rates? What are good hires? What are good margins? What are like, if you don't know those things yourself, like have an investor who's pretty well, you know, in that space. I think in terms of advice, like I find just again, knowing where investors are strong. So we have some investors that are very strong on product thinking, other investors that are very strong on hiring thinking, organizational.

others that are very commercially minded. And look, they all chime in on everything and that's wonderful. And it's a good conversation, but just know where people's strength lies so that you as a founder have somewhere to go. You don't need it to be one person. That's not the point. But when you have that question, you do need to know who, you know, who are the one or two people you can go to are really going to be able to help you in that area. That's the sort of more important part. So these days I do recommend to founders go have an advisory board or sort of, you know,

advisors or maybe just angel investors if you want. It's pretty easy. Even if somebody wants to invest a huge amount and you don't have room, let them invest 5K. They'll still give you the same amount of time, whether it's 5K or under K, and you'll get the advice. Not everybody, but most people, and you'll get the advice. I highly recommend going and finding the people that you think you're really going to need over going and trying to find people that are just like fancy names. The fancy names aren't necessarily going to help you.

but the right advisor could make a big difference in the right moment. then sometimes I think it's, look, I believe in having a board. I believe in having true sort of a person you show things to who then can kick your butt, right? That's what you need sometimes. They may not always agree with you and that is the point of that. It's not just because your board has a fiduciary responsibility, it's because they're truly trying to think about what's best for this company and...

Alex Levin (25:32.736)
It's good to have those diverse opinions. You don't have to agree with them. That's fine. You don't have to do what your board says either. That's also fine. But it is important that you listen.

Prateek Joshi (25:42.453)
Amazing. Moving to just the company building, the experiments you have to run to go from early days, something is working to today. So looking back, what's an experiment that, obviously we've tried a bunch of experiments, like what's an experiment that worked really well? And also maybe it's a different question. What has been the biggest challenge taking the company from the early days to where you are today?

Alex Levin (25:54.606)
Hmm.

Alex Levin (26:14.178)
What's an experiment that worked well? Just sort of like, in regal, I think there's different levels of experimentation, if you will. Day to day, there's a million different things happening. Like literally the product is built so there can be experimentation. So we'll go in and say, hey, let's help every customer set up something where they use SMS before calls and like that works very well. Or let's go set up everything.

where they use a voicemail drop, great, that worked very well. like literally from a product perspective day to day, like our teams are enabled to do all kinds of stuff. I think then there's like experimentation on like an organizational level, which is like a different more of a founder level experimentation. And then there's like experimentation on products and MVPs and stuff like that. I think on an experimentation level, what's worked for our, on an organizational level, what's worked for us very well is,

you know, much more investment in post sales than perhaps at other companies. You know, we believe in like for certain products, PLG, but we are not a PLG type product. Like we intentionally go serve larger companies who have real challenges. And even if they want the ability to self-serve, often they want us. So like we've heavily invested in post sales much more than other people in the space. on a like product standpoint, you know, we, we have our own opinions about like what

people should be doing. So just use SMS as an example. We think all of our customers should be using a lot of SMS. An interesting thing that we found is that in many of these companies, we work with the contact center or with the sales team or within the contact center or support within it. marketing has a separate program for SMS. So often if we offer SMS, we're allowed to do it in certain cases. But marketing says, no, no, we need to do our marketing SMS out of another product. And that's an interesting dynamic where

You know, we haven't solved that problem. think it truly would be helpful to context center teams that have much more control of their own destiny around SMS, but a lot of companies, the marketing team kind of is too tight with SMS and so they're missing opportunities. And that's a real shame. So it's an example of one that sort of didn't work so well, not because the product was bad, but just because we haven't figured out the right go to market and alignment internally to make sure people are really using SMS more as they should be.

Alex Levin (28:39.246)
in programmatic ways, let's say. And then in terms of like, know, bet the company type stuff, you know, again, I'd say, you know, do you go to Europe or not? You know, do you switch into a new product or not? Like those are more, you know, do you change your pricing model or not? Those are kind of more bet the company decisions, which are harder to sort of like A, B test and try. You know, you have to really go just take a swing.

Prateek Joshi (29:04.022)
Within Regal, obviously, just to take a step back, Shopify CEO's memo has been circulating. It's amazing. And the point he made is really good about AI, which you mentioned earlier. So within Regal, where do you use AI internally for your own work?

Alex Levin (29:22.926)
Yeah, so very early, I'd say it was used by the marketing team around content generation, like, and that continues to be a good sort of place, whether it's LinkedIn posts, blog posts, whatever, you know, copy for the website. We have like a regal version of, you know, regal GPT, as everybody calls the stuff, like just internal stuff. And then I'd say the other place that really we've started pushing it is on the engineering side.

I think we saw uneven adoption. Some people just like very quickly started using cursor or copilot or whatever they wanted. And other people didn't. And we've basically made the decision you have to be like, we'll pay for it. Of course we'll run the trainings, but you got to be using this because that's just the way the world works. Now, like you're not going to be able to do your, it'd be, like saying, I want to be an investment banker, but I'm going to use like a paper spreadsheet instead of a, you know, Excel. Like you don't have a choice. You have to use Excel and like we'll train you on it.

But like, if you don't use Excel, like you can't do your job. I think particularly on the engineering side, it's changing the way those organizations are working. Like I'm a terrible engineer. I was not a computer science undergrad. I don't know how, like all the important things, but you know, I can like go and copy and paste things and change the, you know, few lines and like make it work and like, okay. So like, how do you sort of take advantage of that? I think within, you know, engineering, let's say is important, but also within.

product is very important. So one place that we've not gotten to yet, which I'm hoping we get to soon is shortening the cycle from customer asks for a feature to it. There's a version of it they can play with right now. Customer asks for a feature. we'll add it to the roadmap. We'll eventually have a conversation on how much time it would take. Eventually somebody will spec it, design it, but it'll be months or using these new sort of vibe coding tools, whatever, you can go to Bolt and just go and say, make me a feature that does this.

Like how do you empower the team so that actually can somehow integrate into our system and that product manager can just instantly show the customer would be like, cause maybe it turns out they don't want the feature at all. Or maybe it turns out that actually they meant something different or maybe just that simple version of the feature was enough for them and you never had to get anybody involved. And I think there are things that real engineering teams need to do, but more and more, there are going to be features that you don't need real engineering for. I think like empowering your team to do it will be very powerful. We're not there.

Alex Levin (31:44.952)
today, but I'm hopeful over the next six months we will get.

Prateek Joshi (31:49.039)
I have one final question before we go to the rapid fire round. Now, how do you see the future of AI phone agents evolving over the next two years? And also, what AI advancements are most exciting to you as it relates to this?

Alex Levin (31:49.678)
you

Alex Levin (32:06.604)
Yeah, look, I think the 90s were when everyone started building websites to interface with customers. then it was, I guess, eventually like kind of apps and social. And now it's voice. I think the main way you're going to interface with these companies is through voice. And it's funny, people go, aren't the youth on SMS? And I go, no, the youth now are on voice.

Like the youngest generation, like that's just like my kids, they just like start talking to the TV. I'm like, no, the TV doesn't do that. But like, they don't understand like to them, that is the natural way of engaging. And so I think, you know, what I guess, you know, probably to be through your phone or through a computer, but for most use cases, you're just going to talk and it's going to immediately engage and respond and know everything about you and everything about the company and be able to do certain things for you. So I think like thinking through like,

Prateek Joshi (32:35.563)
Yeah

Alex Levin (32:58.304)
a world in which that's the prime engagement channel is pretty fascinating. So if you haven't done it with your executive team, or you guys are still on the VCA side, you had your portfolios do it? I'd highly encourage your portfolio companies to think through what happens if 80 % of the way people engage is voice, not mobile app, not social, not desktop website. That's the biggest one. And in terms of advancements,

Look, I think there will obviously continue to be advancements in the LLMs and the Voices. Cool, fine. But put those to the side. Like it is already good enough. I think the places where the major blockers exist today are different. It's in legal concerns. It's in integration concerns. It's in, you know, all the having the right knowledge written down. Because if it's in people's heads or in tribal knowledge, the AI agent can't possibly know it. So I think there's going to be a lot of actual

hopefully rapid development and how you can sort of fix the legal problem, the integration problem, the RAG sort of the knowledge-based problem. And the people who solve those will do very well. Yes, again, there'll be improvements in the LM's, but it's not necessary anymore. whereas the things I just mentioned are necessary.

Prateek Joshi (34:13.237)
Amazing. With that, we're at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. Ready? Alright, question number one. What's your favorite book?

Alex Levin (34:28.921)
man.

I love company histories. I think like, if you haven't read, I think it's PayPal wars was the one about PayPal, sort of a great one. Like just as an outsider, remember like PayPal being this huge thing and you you see this like beautiful company and then you hear about everything that's really happening under the hood. That's so like completely insane. And I think as an entrepreneur, it's heartwarming that like everyone has real challenges even when it's going well.

Prateek Joshi (34:59.487)
I agree. love company histories too. And also on the podcast side, acquired a podcast. do company histories as you may know. yeah, I love that. Next question. Which historical figure do you admire the most and why?

Alex Levin (35:16.41)
that's a good question.

Alex Levin (35:22.414)
Look, you know, in a sort of lame sense, I'd say, you know, I do really think that something special happened when the framers of the Constitution met and wrote a document that enabled us to do this stuff today. There wasn't anything particularly special about, you know, the land or the country or whatever, but somehow, like, we've been able to create a system which benefits, I think, a lot of people in sort of creating, you know, jobs and innovation and sort of open, free economies and open sort of free thought.

So I think if I could sit around the table with the original framers of the Constitution, that would be pretty fascinating. And then in terms of business, you do see these, a lot of them are still alive, these folks who were able to just will industries into existence. Maybe Carnegie's not alive anymore, a must literally will the private space travel into existence. No one thought that was possible.

Prateek Joshi (36:19.637)
what has been an important but overlooked AI trend in the last one.

Alex Levin (36:26.35)
I mean, look, I think the single most important tooling is going to be on the evaluation and sort of.

whatever testing side, if you want to call it that basically to what we said before, like, Hey, you create a new agent or you make one change. How do you automatically test it a hundred times? And I had like the CEO of a YC evaluation company telling me, no way. Like our model just runs it once. And I'm like, guys, like you just clearly don't understand the point of AI. Like the point of AI is that it is not deterministic. It's a stochastic model. It can do all these different things. So you can't run it once. so how are you going to create a system that builds in that?

randomness and diversity to understand like the crazy things that customers do on the call, right? What customers interrupt, they sort of go quiet, they change their voices, different people get on the phone, you know, they're asking questions you never would have thought of before. And if you don't understand how the change you made to that AI agent is going to be impacted by all that, you really like have to manually test a million times and that's a huge blocker. So once you can have...

the equivalent of like a real time development pipeline where you just hit something and it tests it and goes to production for AI agents, I think it'll be a massive unblock.

Prateek Joshi (37:39.167)
what's the one thing about AI phone agents that most people don't get?

Alex Levin (37:45.262)
If they haven't played with them, think most people are still thinking about IVAs and old style things. They, yeah, I've called lots of AI bots. And I'm like, I already know you don't know what you're talking about if you're talking about AI bots. Like there's a new technology. LLMs have changed the game. like, if you're not, know, even some of the people playing sort of the AI games today are not really using the most modern LLMs. Like if somebody's charging you a penny a minute, they're not on them new models. They're not on the new voices. not on the new, like they just can't be.

no way that they could do it for a penny a minute. So I think people have lot of misunderstandings there. And then I think, you know, a lot of people are, there's a, there's a interesting type of choice or sort of happening now where some companies are basically effectively using the voices, not the LLMs. And what I mean by that is they basically are having a form filled out over voice.

They're saying, do exactly this, then do exactly this, then do exactly this. And like, yes, there's an LM, but it's not being allowed to do anything. It's so formulaic. On the other hand, there are people that are trying to go the other route. They're basically saying, well, how do we make it so that all I tell the prompt is you're an insurance agent. Go on board and do customer. We don't tell it anything. And that's also bananas crazy if you, in my opinion. So, you know, I think the reality is like somewhere in between, like, again, that's why I think actually contact centers are very well suited.

or engineering teams are very well suited to using LLMs is because they've had to for years take people who are very smart, take humans who are very smart, but then put rules on them and say, this is how the coding language works. This is the way in which you work in sort of Ruby or, you know, on the contact center side, like this is how we want you engaging with customers and put certain rules that still allow creativity and great customer experience, but within the bounds. And that's the...

Yeah, anybody who's too far just letting the LM do anything it wants or too far just making it completely deterministic is not taking advantage of the technology in which it.

Prateek Joshi (39:46.832)
What separates great AI products from the merely good ones?

Alex Levin (39:53.322)
In our space, you know, I think what we're seeing from, or the feedback we get from customers is that there's a lot of people again, who can just have a voice that does a thing, but to take that into an actual production environment where now you're going to have to do A B testing and have multiple channels and integrate with systems and, know, do after call analysis and have compliance. Like that's currently the, the, the piece that is hard for most companies. There's a million YC companies that just like have a voice that does something.

That's not what you need. Like you actually need, let's call it a workflow tool, you know? So if you're making a movie, you need the tool that's going to help you use AI agents to make a movie. If you're running a contact center, know, customer experience, customer service type interaction, you want a different set of tools. And so the companies that are very clear about who their customer base is and how they're going to make the workflow tools for that specific customer base are going to do the best or are going to be the best AI companies.

Prateek Joshi (40:49.663)
What have you changed your mind on recently?

Alex Levin (40:54.574)
I'll give you one where I was wishy-washy and now I'm pretty extreme in one direction is a lot of customers when this AI stuff started were asking about chat. Hey, can we have a chat bot but AI? Even like Sierra, that was their first product was chat and a bunch that were very heavy on chat. I was like, okay, maybe we're going to need to build chat. We ended up building it. I'm at a point now where chat needs to die. Chat needs to never exist anymore.

Basically, I was talking with somebody and their point was, look, nobody likes chat to start with. It's a bad interaction. If you drop off the page, it disappears. It doesn't persist the history. It doesn't allow you to know. There's a bunch of reasons it's bad to start with. But if you're going to chat with somebody, just allow them to either do SMS or do voice immediately. There's no reason chat ever needs to exist. And so I think it's a case where I went from kind of...

unsure to having a conversation and going, yes, chat needs to go away from.

Prateek Joshi (41:56.361)
What's your wildest AI prediction for the next 12 months?

Alex Levin (42:02.326)
Look, I think the best companies in our market, yeah, will be 90 % AI agents in the next 12 months. that may not even be wild anymore. I think last year when I'd say that to people, they'd go, it's not possible, blah, blah, blah. Now people are going, yeah, that's probably you're right. It might be a year, might be three years, you're probably right. So if you want to go like one step removed, like the next step, again, it won't blow your mind because you live all this stuff, but...

I sort of explained to people, soon the AI will not be dealing with humans, the AI will be dealing with AI. I have more and more friends who are playing around with simple, they're not using a third, like a vendor for this, they're just kind of building simple AI that can do certain actions for them. Maybe it's call the pharmacy and tell them you're coming, or maybe it's make a grocery order. It's nothing complicated, but I think more and more, as a business, your end user is not going to be a human, it's going to be AI.

Prateek Joshi (42:54.721)
All right, final question. What's your number one advice to founders who are starting out today?

Alex Levin (43:01.974)
Number one, so find a co-founder that you're gonna go through thick and thin, for sure. If I'm gonna go a second one past that, think, I used the expression before of like have a product where somebody's like literally reaching across the table and pulling you like that. Like too many people are going to like their friends or their YC badge or whatever and people are going, that's nice, I'll pay you $20. Like that's not what you need.

What you need is something, whether it's a PLG motion or real sales motion or whatever, consumer business, or literally, even though it's half broken and it barely works, there is something in that product you're offering or service you're offering that is so important that people you don't even know are paying you money for. That's a sign that you've got something. And if you don't have that, go find that.

Prateek Joshi (43:51.849)
That's a very good point and many times when people are discussing like how do I measure or how do I know if I have PMF or not, you'll know at the point there like you'll know it won't be a subtle thing it'll be so obvious that people will be coming after you're pulling the product out and that's when you know that something's working.

Alex Levin (44:09.9)
Yeah, yeah, the hard part is you'll have eight things that you think you're selling and you'll be telling yourself you're selling, but there'll be one that everybody wants. And I think that's the hardest part as a founder. We're like, no, no, no, but I'm going to build these eight. People are like, no, that one thing. And you sort of have to like realize that you got to make a decision.

Prateek Joshi (44:18.07)
Right.

Prateek Joshi (44:26.473)
Right, amazing. Alex, this has been a brilliant discussion. Loved your insights, loved the way you think. And again, thank you so much for coming onto the show and sharing your insights.

Alex Levin (44:37.282)
Yeah, thank you for having me.

Prateek Joshi (44:39.542)
First, right.