Boosted AI’s Josh Pantony on Selling Maluuba to Microsoft, Building AI Workers for Hedge Funds, and Transforming Capital Markets

Josh Pantony has been building AI companies since before it blew up. After co-founding Maluuba in 2008 and selling it to Microsoft, then helping Bloomberg build their first machine learning systems, Josh is now co-founder and CEO of Boosted AI, where he’s creating AI workers that automate the daily workflows of hedge funds and investment banks.

In this wide-ranging conversation, Josh shares the story behind Maluuba’s exit, why he believes traditional SaaS is dead, and how his company is building technology that could make Wall Street more efficient.

He also explains the technical breakthroughs that made modern AI possible, the challenges of scaling an AI-native company, and his vision for becoming the “cyborg element” that guides how trillions of dollars get allocated across global markets.

Read on for the full discussion.

Mark MacLeod:

In this episode, I sit down with Josh Pantony, co-founder and CEO of Boosted AI. When I think of founders who are just born for the role that they’re in, Josh comes to mind. Josh has been in the AI space, specifically in financial services, for a really long time. He’s worked at Bloomberg, which many consider to be kind of the default tech player in the space.

We’ll unpack his journey on his prior company. He’s already had a great exit behind him, his decision to leave that company to go to Bloomberg. And we spent a lot of time talking about AI generally. He actually gave me a big education on AI and why it’s kind of such a big deal now.

And his thoughts on SaaS Punchline. If you’re a SaaS CEO, you better be figuring out AI and just talking about the impact that his company is having on the financial services market, which is, of course, one of the richest markets in the world.

Josh, welcome to The Startup CEO Show. It’s a true pleasure to have you here. How are you doing today?

Josh Pantony:

Excellent.

Mark MacLeod:

All right, cool. I’ve got lots of questions for you, Josh, but maybe just to set the stage for what we’re going to talk about, just tell people what it is that Boosted does.

Josh Pantony:

Yeah. So, Boosted AI is a system that makes it easy to build these sorts of little AI workers, specifically in the investment management space. So, hedge funds, portfolio managers, analysts, large Banks, and ETF providers basically use our system to create these little AI workers where they’ll teach them to do some kind of work, and then it’ll do that work on a continuous basis.

I think one of the biggest things that’s maybe different between this and other agentic plays is there are a lot of workflows in finance that really repeat. You know, every time there’s an earnings call that comes out, there’s a specific set of work you’re going to do.

Every time there’s an update to an 8K, there’s a specific set of work you’re going to do. And so we basically help users build these little workers that just automate parts of their daily processes, right?

Mark MacLeod:

Oh, that’s fascinating. I have lots of questions about Boosted, but let’s take a step back. I mean, I want to get into all of your history, but when I used to be a VC, and this was a lesson I learned kind of after a little bit. I didn’t enter the industry with this insight, but founders who was uniquely qualified to build this company. Like, I always found that the best investment opportunities were based on a really simple crisp insight.

And so often, the ability to create said simple crisp insight came from actual deep domain knowledge. And so, when I think about you and I think about kind of founder problem fit, it seems like it’s bang on. So maybe talk about that, and maybe even just the origin of this. I don’t know, I feel like you’re uniquely qualified to do what you’re doing, but yeah, maybe let’s talk about the origin.

Josh Pantony:

Yeah, so I’ve been in AI for actually a relatively long time, given the current crop of AI founders. I really got started back in 2008, 2009 as one of the co-founders of Maluuba. For those who haven’t heard of Maluuba, we were basically like a Siri competitor. We powered LG, Samsung, BlackBerry, and Windows Mobile devices back when that was a thing, and eventually got bought by Microsoft and turned into what was called the Turing team.

And that team eventually went on to actually build the first version of Copilot and a bunch of other really cool things. I was actually the CTO at that company, not the CEO. And that was really, really fun. I built a lot of cool technology there. I was there for a number of years. I eventually got recruited by Bloomberg to help Bloomberg start one of their earliest machine learning groups.

At the time, they had basically no AI. I’ve been talking to a few companies, I’ve been talking to Google, Amazon, and a few other folks, but what really intrigued me about Bloomberg was this massive data terminal with huge amounts of very difficult data, very hard to use. You can take an entire MBA course on how to engage with the terminal.

Mark MacLeod:

Oh my God.

Josh Pantony:

And the goal of the team that I ultimately helped build was to make it simple and easy. It was like, come in, ask a question, get an answer. So it was kind of what I’d done at Maluuba, but really, really supercharged for the finance space. We ended up building this really advanced search engine, and a bunch of other question-answering stuff. It was a lot of fun.

So I did that for a number of years, and after the Microsoft acquisition started thinking a lot about what I wanted to do next, and I guess to kind of rewind on that a little bit. I’d actually left Maluuba before they got bought. So I helped build it up until we got our first, let’s say, 20 million customers, something like that. 20 million users, I suppose you should say.

Mark MacLeod:

Sounds like a nontrivial amount.

Josh Pantony:

Well, especially back then, actually. These days, I feel like you get those numbers fast. But back in 2012, and 2013, it took a lot longer, usually. And so when I ended up at Bloomberg, it was actually kind of, let’s say, a challenging tax situation. Anyone in the world who ever has a Canadian CCPC and then goes to the U.S., I’ll just say, do your homework. It’s a little more complicated than you may realize.

And so kind of, long story short, I ended up back in Toronto with a whole bunch of liquidity. I had to actually leave what had been a really awesome growing career at Bloomberg. And I had to think really quickly about what I wanted to do next.

Mark MacLeod:

Sorry, you were forced to leave because of the Maluuba exit? That’s a champagne problem.

Josh Pantony:

Yes. It was kind of funny, though, because it was like when I was there, my bonus was like 40% of my comp or something like that. So I actually walked away from that to get Canadian residency and a whole bunch of other stuff in time, and sort of walked away from what had been this development career, suddenly had a whole bunch of cash, and I was kind of thinking about what I wanted to do next.

And we thought about a lot of different things. Just some folks I’d known at Bloomberg, but it was actually my old boss at Bloomberg, a guy called Jonathan Durando. The same guy who had convinced me to join Bloomberg, who kind of forced my hand. We’d been talking offline about a bunch of different stuff, and he called me up one day and he basically said, hey Josh, great news. I’m like, what? And he’s like, I quit my job. I’m like, what? He’s like, yeah, I’m ready to start building that company we’re talking about.

Mark MacLeod:

“I was totally joking. What are you doing?”

Josh Pantony:

I still bug him about that to this day. And so it really started off with the two of us, and we thought about a lot of different things. I think kind of right after Maluuba sold, I think it was the year 2017, the paper Attention Is All You Need came out.

And I don’t think we quite realized just how important it was going to be yet. But there was this general sense that a lot of the kinds of approaches we’d used at both Bloomberg and that I’d used at Maluuba were about to fundamentally change. And there were about to be types of work that could be unlocked that you really couldn’t unlock in the past.

In particular, after spending all this time at Bloomberg, I saw just how sort of behind-the-curve finance was and investing overall. And it’s this really interesting space where I think today, maybe as much as 50% of what analysts do could be automated. But it’s also one of the absolute biggest total addressable markets in the world. It’s one of the most high-value and high-impact types of work that you can apply automation to.

It’s also a space where people are constantly bombarded by information, and just giving them even a marginal ability to absorb more information than their peers can have a really high impact. And so it seems like a really interesting space to get into.

So John basically forced me at gunpoint to start this company with him. We started looking at the space and in the beginning, I think we actually went about it all wrong. We started thinking about it as more of like, could we do a better robo advisor?

Mark MacLeod:

Oh, nice.

Josh Pantony:

We had some success there, but the robo-advisors had already kind of played out a little bit. At that point, there were already a few people who had spent a bunch of cash on it. And I think VCs were starting to figure out that just because you have a billion dollars in AUM doesn’t mean you should have a billion-dollar market cap.

So then we started kind of playing around in a few other spaces. We started interacting actually with credit unions and groups like that, and then finally I met who eventually became the third founder of Boosted AI, a guy called Nick Abe. And Nick had actually been a former portfolio manager himself. He was working in a sort of middle management at RBC when I came across him.

And he suggested the idea that if you’re going to make the most powerful weapon in the world, you should sell it to the people who need the most powerful weapon, who use those things. And so we started going out and actually talking to a few portfolio managers and very quickly found some folks who were willing to pay reasonably large amounts of money for a competitive edge.

You know, we had this really wild thing where we’d spent years in the kind of like R&D space exploring and trying it out. And then late 2019 went from like zero to a million dollars in revenue basically overnight in a six-month period, raised our first venture check from there, and was off to the races.

Mark MacLeod:

You wanted to avoid a poor market. Like, you wanted markets that could afford to pay for your product. So definitely sounds like you’re obviously the choice of kind of your vertical, very rich vertical, clear ROI. Anything that gives you alpha, you’re going to pay for. So yeah, definitely checking all those boxes.

Also thinking about your timing, like, so AI is everyone’s obsessed about it now, but of course, it’s existed for a long time. It was just called machine learning 2008, 2009. That strikes me as like, I don’t know, Peak Web 2.0. So you were ahead of this new curve.

So, I’m going to ask a completely technically dumb question, is it just Moore’s Law that has made AI so possible now, given that like, to your point, you were working on it kind of a long time ago, and of course, as a concept, it’s existed for decades.

Josh Pantony:

So it’s a combination of two things. Certainly, Moore’s Law is an element of it, but I actually think we had enough compute as early as 2013 or 2014 to start making the kind of models we have today. And so it’s not just Moore’s Law. You know, I think the other thing, of course, was the amount of data you could train on. But arguably, we also had that amount of data in 2013 and 2014.

I think it’s actually two fundamental unlocks that have massively changed it. The first is that we got really good at what’s called unsupervised training. So to give a little bit of context, when I built my first company, Maluuba, we were one of the first companies to start figuring out how to get large amounts of effectively annotated data.

So we trained our systems on maybe 300,000 examples of basically think of queries, and we’d sort of teach the machine, okay, this is a query that’s talking about restaurants. This is a conversation that’s talking about movies. This is a conversation that’s talking about purchasing a retail product or something like that. And before we had done that, you’d had a bunch of basically only the really big corporations to be able to do it. Like the really big corporations like Google and AT&T, this company called Nuance, and they were paying millions of dollars.

Mark MacLeod:

Nuances like natural language processing.

Josh Pantony:

Yeah, yeah. I mean, Siri started as a DARPA project, right? So Siri basically had a ton of people who were basically hand-coding different systems. We made two fundamental unlocks at Maluuba. Number one, we were the first to start using, or one of the first to start using mass-scale crowdsourcing, where we would go out, pay people a penny to do an annotation, and then attract hundreds of thousands of annotations that way.

Before we even did that, we did another interesting thing where we were trying to find a way to get hundreds of hours of data to train on. We couldn’t afford anything. We had zero dollars. We hadn’t raised any money. And so what we noticed was TED Talks. TED Talks had audio clips. And then they were one of the first to try to transcribe it. So they’d actually have humans transcribe it.

And so they had this massive set of data. It was like hundreds of hours of someone giving a TED Talk, and then the text that went along with it. And then we started thinking about, well, can we get subtitles from movies and things like that? And so that was one of the first really cheap annotated data sets, as we’d go out to stuff like that and we’d start training our models on that. And we were one of the first to do it. And we just did it because we fundamentally could not afford to pay the huge prices that big companies were doing.

We built a really quick data set that way. And we built something, but you still needed some form of annotation. You still basically need something where it’s like, here’s a piece of data, and I’m telling the machine what this means. And there was this period of time where people started thinking, well, could I just give it the entire web? Could I just give it literally every piece of text?

In 2012, 2013, and 2014, the kinds of techniques we used to get value out of that data, the sort of what we call unsupervised training, only worked for abstractions. It was like trying to understand how similar two words were, or trying to understand the general context of what something meant, or trying to understand the sentiment.

It was like really, really basic kinds of embedding models that we were doing. So, one of the inventions that was necessary for modern AI was the concept of next-token prediction. Given a sequence of words, can I predict what the next word should be? Interestingly enough, we call these things large language models, but they actually come out of these old hidden Markov model systems that were trying to basically do the same thing, but in a much, much, much more basic way.

So step one was learning to train a model from all data through this concept of next token prediction, which was a very novel idea, at least in the way that Attention Networks articulated it. Then the second thing that Attention Networks did that was really revolutionary is really prior to that, we had these sequential models where you’d use this thing called backproc through time, which basically meant that if I was trying to predict a sentence, I’d have to kind of predict the whole sentence in sequence. So I predict this word, this word, this word, this word, this word.

And you could not paralyze that, really. You could not run it across multiple GPUs. You just had to have one really powerful GPU. And so what happened with Moore’s Law really was we got to the point where you could get much more powerful compute if you could figure out how to spread the compute across, like a million computers. And I don’t exactly remember what year, but at least in 2015, 2016, everything was still these sequential models that you could only really train on one or two computers.

You could take a pedantic argument that there were these averaging techniques that would kind of do it across different computers, but they weren’t truly independent, and there was a limit to how much you could scale. So the other thing that Attention Networks did that was really, really smart is they created a way to train every word independently, every token independently, if I were to be a little more pedantic.

That meant you could train a model a thousand times faster, a million times. I mean, just take your context window. Now, if I have a context window of 200,000 tokens, that’s 200,000 times more computers I could use, let’s put it that way, than an old sequence-to-sequence model.

Mark MacLeod:

That’s a complete game-changer.

Josh Pantony:

Massive, massive game changer. So number one, now I can train on the entire web, and number two, now I can use all the compute in the world. I want to train this. Those are the two things that fundamentally led us to the modern era of generative AI. And in 2017, intention is all you need. You could kind of detect that those things were maybe possible. I think what shocked most of us was just how much intelligence you could unlock.

If you jump all the way back to like 2010, there was a general belief that feed-forward neural networks were the smartest thing that you could do. Like the old-school nets that we had back then, there was this fundamental belief that everything else was just people playing around with stuff. Attention networks really completely fundamentally changed the game and made things possible that frankly, I didn’t think would be possible in my lifetime. So sorry, that was a very long answer.

Mark MacLeod:

I feel like we could do an entire podcast on this, and for me, as a buzzword-compliant neophyte, I’m drinking from the fire hose. But maybe we’ll move on. You know, you glossed over what feels like a really important decision when you were talking about just your journey to Boosted, which was your decision to leave Maluuba. Maybe talk us through that.

Josh Pantony:

Fundamentally, I think the biggest shift that happened there was actually the sort of founder’s journey. So when we started that company, we were really young, I think, like I might have been 22, 21, early in life. And you know, we spent a few years before we were an official corporation kind of working out the prototypes and thinking about it. And we went through a few different iterations there. We were at one point going to build a mall kiosk kind of product. At another point, we were thinking we were going to replace cashiers.

Mark MacLeod:

Very wide set of ideas.

Josh Pantony:

Eventually, someone said, put it in a phone, dummy. And then we were like, oh, that’s actually a great idea. That’s how the first version of a remote consumer app was built. Fundamentally, I think the biggest difference between me and the CEO at the time was just a belief in how big it could become.

The interesting thing is that Sam was the CEO, and I was the CTO. Sam believed that he could become what OpenAI eventually became. That was really his dream. He fundamentally wanted to build a multi-billion-dollar company. And the ironic thing, as an aside, is if he’d waited a year, he might have been able to do it.

After we got Samsung, BlackBerry, and Microsoft as customers, I don’t know if we’d quite landed Microsoft yet, but we were in conversations. But after we got BlackBerry and Samsung as customers, we got to really high revenue very quickly, but I didn’t see much more room to grow. I was like, well, we could try to go into call centers, we could try to go into cars, we could try to go into TVs.

Mark MacLeod:

This is strictly like an OEM play white label. You stick it in your device and off you go.

Josh Pantony:

Exactly, yeah, yeah, yeah. So we used to get money basically based on every BlackBerry that was sold. It was a great deal. Pleasant side, if you can get an OEM business, very long sales cycles. And they tend to be like seven-year-long contracts, huge revenue, great arrangement.

So we know we’ve done pretty well at that point. We had a few folks offer to buy us, and you know, all of us came from like let’s say, middle-class, lower middle-class backgrounds. So when you’re like 22, 23 years old and someone’s offering you like $100 million for the company, it would have been life-changing for all of us. Yeah, it was extremely compelling.

And I have to give huge credit to Sam. He had no interest whatsoever in doing that. He wanted to build a billion-dollar company. He was very convinced he could do it. Fundamentally, at the time, I did not agree. I thought there was a limit to how far this technology would go, and there was a limit to how big the market could go. And we know nothing about selling to call centers. The only reason we were successful selling to these big carriers is because we are engineers, we know how to talk to engineers. Selling to a call center is going to be a different process. We didn’t even know how to approach them. We didn’t know the first idea on how to sell into that.

And you know, we had started the process of selling it to cars. For anyone who wants to go into automotive, it’s like a seven-year sales cycle. Like by the time we were landing our first auto deals, we’d already sold to Microsoft. So you know, we hit like in the range of 10 million in revenue, and then I just saw no path to go past it. And by the way, I was right. Whoever it was, four years later, we were still at 10 million in revenue. But he was right that the premium on AI would go up. We sold for well north of 100, I’ll say that.

I think the other thing that ended up being somewhat prophetic is, again, if he had actually waited a year, attention networks, all you need would have come out. Maluuba would have been the best NLP company in the world. And we actually maybe could have been OpenAI. It might have been a realizable vision. I just didn’t see that technology coming.

Mark MacLeod:

Did he call you up to say “I told you so”?

Josh Pantony:

No, he’s actually eventually started his own company. He’s raised capital. I don’t know if he’s out of stealth yet or not, but he’s doing very well. There’s another founder, Jiwoo started a company called Mind. I think he’s also doing well. So all of us went off and you know, I talk about trying to be the Maluuba mafia, and I think the great thing about having success in life is it lets you sort of play in a bigger sandbox.

All of us are thinking big this time, all of us are aiming for a multi-billion-dollar IPO someday, and I think it’ll be easier to do than if we tried to do it with Maluuba. I think the other thing we didn’t anticipate way back in 2012 was back then when we trained models, we would spend $5 in compute training a model, something like that. You know, it cost a lot of money to get the data, but once you had the data, you were like training it on a graphics card for 10 minutes.

I think the thing that none of us anticipated was to become what OpenAI eventually became. You needed hundreds of millions of compute, and that need and that discovery of the scaling law was something that probably would have taken another two years, I think, even after the acquisition.

So, I like to joke that we could have done it if we waited a year, but I think in actuality, we sold at a good time. So yeah, I mean, long story short, I left fundamentally because I didn’t think the company could go where the CEO thought it could go. And I had an opportunity to make a life-changing amount of money at a different job, and I took that opportunity, and it was a positive ending. We still keep in touch and converse every month or two. Obviously huge overlaps in the network, but he was just a little more aggressive in the first run than I was.

Mark MacLeod:

You know, there are two things you need to succeed in business and in life: alignment and shared values. This was a line that Mike, co-founder and former CEO of FreshBooks, beat into my head all the time. And you two just weren’t aligned in terms of what the terminal kind of vision of this thing was.

Josh Pantony:

Yeah, I always tell newer entrepreneurs to treat having a co-founder as very similar to having a marriage. You know, you may in fact actually spend more time with your co-founders than your wife, and it’s oftentimes very high-impact, very stressful scenarios. And so, yeah, I totally agree with you.

Mark MacLeod:

Talk to me about how you transformed yourself from a CTO into a CEO.

Josh Pantony:

Okay, so this was really interesting. Just between you and me, and I guess everyone listening to this podcast, I really enjoy the CTO job. I’m happiest when I’m tinkering and when I’m playing around with new tech, when I’m thinking about where the technology’s going. And I think it’s a good thing that John was my old boss because otherwise, I’d probably drive him crazy with the amount that I get involved in the technology stack to this day.

But I wanted to try something new. I built up the first company as CTO. I also felt like, with the kind of background I had, I couldn’t think of a better CEO. That’s not hubris, it’s just an interesting combination where I knew the technology really well, and I knew the base we’d be selling into really well.

And I also thought as CEO, I’d be better able to use the network that I built between Bloomberg and Maluuba to recruit some of the best talent. So there were a lot of things that were different. I would say the biggest challenge was hiring salespeople. When coming as an engineer, there are so many different things about managing sales that are just fundamentally different. First of all, how do you identify a good salesperson?

Mark MacLeod:

Especially because if they’re good at selling, they’re going to perform well in an interview, and then that’s a false positive.

Josh Pantony:

Yeah. And I also think you can. Like in the early days, you’re kind of fighting a little bit for product-market fit. And so it is that your PMF is bad, or is it that you’ve done a terrible job hiring? Turns out that for our first two years, it was both. But it took a long time before I got a person who was good at sales, and I did not hire through a standard practice at all.

What actually happened is that eventually, when we got the first version of the product we were able to sell, we started getting some portfolio managers who wanted to invest in the company. And there was this gentleman, actually Michael Priest, who’s still with us, who was an early angel investor.

He was a co-founder of a hedge fund out of New York called Mira. They were like a three-and-a-half billion-dollar hedge fund. He wrote a check, and then he got really excited about what we were doing, and he said, do you mind if I join the team? I said that sounds fantastic. He was my first real sales guy. And he wasn’t really a sales guy; he just knew the space got us some of our clients.

Mark MacLeod:

You know, a three-and-a-half billion-dollar hedge fund. I can run the math on the management fees on that, let alone the carry potential. So, he walked away from that?

Josh Pantony:

He walked away from an absolutely insane comps package. I don’t know if I’ll talk about it online because I don’t have his permission, but he was both really convicted in our mission and I think honestly he had more fun here.

Mark MacLeod:

That’s an endorsement.

Josh Pantony:

He was the closest thing to a proper sales guy that we actually had. And then really quickly, there was someone else who had actually been a portfolio manager who started off as a client and then got excited and then joined us in sales. So those were my first two salespeople were basically an angel and a client. Everyone I tried hiring before that just went terribly. They didn’t know how to sell, or you know, they misrepresented themselves.

And also, like for me personally, there are a lot of times in my past where I’ve hired someone and actually gained mentorship from that person that I was hiring. At Maluuba, we were really selling to engineers, and we were selling engineers on an engineering problem. I think one of the other biggest adjustments in selling to this space is that most engineers, until recently, didn’t really have much budget power. They weren’t really the decision-makers on where to allocate.

Whereas when you’re selling to a big tech company, everyone up the entire stack understands technology, and you can sell based on the technology itself. So that was one adjustment like just learning how to sell value in a different way. The other adjustment was like learning the kinds of personalities that can sell.

And then I really just did that by kind of getting lucky with an investor, learning from them, and then after I’d seen them sell to a few folks then I learned enough to understand, okay, here’s the kind of people I need to find. Then I started getting a few others and at that point, we started to build a proper process.

You know, today I’ve got a CRO who was the ex-CRO of Blockdaemon and has built a bunch of very large, very high-impact organizations. I’m learning a ton, and he’s awesome. But in the very beginning, building a sales org was just very, very challenging. Something that’s fundamentally different.

I also think the other big difference between a CTO and a CEO is that you have to learn so many other parts of the business. You need to be good at making financial forecasts. You need to understand the concept of client acquisition cost to lifetime value. You need to understand the idea of actually doing marketing. What does marketing look like? How do you think about product positioning?

Some of those things I’d kind of learned in my first business. But many of these things were fundamentally new skills. And I’d say if I were to condense all that into one coherent piece of advice, get as many mentors with as many different disciplines as fast as possible.

Because no matter how smart, or intelligent you are, there are going to be things you’re going to need to learn, and there are a lot of things that you just, I think, fundamentally can’t learn in a textbook. You have to encounter the problem in a real-world situation and then get advice on how to navigate it.

Mark MacLeod:

Yeah, by the way, this could also be an entire podcast. So we’ll skim the surface on this one. Maybe let’s talk about disruption, and there are probably a few different ways we could approach that. You know, back in my operating days, it’s actually going to be 10 years this week since I left my last CFO gig.

Like, the world’s changed a lot. You know, back when I was helping run SaaS companies, we had these product roadmaps that went out six quarters with great certainty. And we just slowly twist a knob here and turn a lever here and just gradually move up and to the right. And it feels like this long-term planning is kind of a luxury or maybe a thing of the past now with AI.

And so I’d be curious about your thoughts… Maybe let’s come back to your thoughts on SaaS afterwards. But just give us a view into running a kind of AI-native company today. Like, I just get this impression that if I’m the CEO of an AI native company, I’m like trying to disrupt myself before I get disrupted. Is that your world?

Josh Pantony:

That’s such a great description. You’re like constantly trying to cannibalize the product you had six months ago. This is the biggest difference, I think, between when Maluuba really hit its stride in like 2012 versus where we are today.

When I first got started, I think my biggest superpower was learning what was possible in academia that hadn’t been applied yet to a commercial setting, and then learning user needs that could be unlocked in a fundamentally new way. And I’d say, like, I did that at Bloomberg, I did that at Maluuba, I did that for many years.

And in an environment like that, it’s possible to have a technology model that can be years ahead of your competition. Ultimately, we hit a point with Maluuba where if you wanted the best MLP technology in the world, there just was nowhere else you could look. Anyone else trying to build it would take them so long to figure out how to formulate the models, to figure out the method we had for getting our annotation for just all these different things that sort of lead to creating this really valuable NLP product.

The new 10 years are like six months now because things are moving at such an insane exponential gradient. I don’t believe it’s possible for any company to have a moat that big. Actually, I think the incumbents are going to be the most smashed by this because I think there are many incumbent companies where their moat is like billions of lines of code that have been written by humans, and I think the cost of producing that code is going to reduce by 80 or 90%.

And so, small startups that iterate fast are going to be able to smash the incumbents in a big way. But that also means if you’re a high-growth startup, it’s fundamentally easier than it’s ever been for someone else to come in and smash you. And so the kind of like durable moats that we used to have are just very, very different today.

I think that the biggest challenge about being the CEO of a startup is that your pace of innovation has to be insanely high just to keep a few quarters ahead of your peers. And there are some fun elements to that. I mean, the sort of flip side to that is like every quarter we built technology that did not feel possible to me when I was younger, every quarter we’re seeing something new get built that is just absolutely insane.

We do things in a week that I think fundamentally would have taken us a year back in the day. The flip side is that you always have to maintain that pace, and it creates a very interesting dynamic. I like to remember having a nice, predictable forecast. It feels very step-functioning these days when you make these ginormous breakthroughs both on the revenue side and on the product side.

But are you also seeing that happening with your peers, and you’re kind of, I like to call it, like riding the bull? You’re constantly riding the bull, and that makes it a very different dynamic than the old days.

Mark MacLeod:

That’s fascinating. You know, back when I was a CFO, it was easy. I’m thinking about annual budgeting, right? It was easy to kind of sit down with your go-to-market leaders and figure out costs because there was like a known determined result, and ROI and payback period and all that stuff.

And I’d always have a bit of a bun fight with R&D as to how much we would spend. And I typically worked for product-oriented CEOs who were just like, we’re going to spend x percent of revenue on product and engineering, and that’s just what it’s going to be. I feel like in this world today, this is the temptation to just continually amp up the spending and therefore continually postpone profitability. Does that feel right to you?

Josh Pantony:

There’s some truth to that. I’d say the flip side is it’s also possible to build highly scalable organizations on a smaller amount with higher talent density, but also a smaller number of bodies than it used to be.

There are lots of examples of companies. I don’t remember exactly how big Kershaw was, but a relatively small team that really quickly got up to 300 million in ARR.

Mark MacLeod:

That’s bonkers.

Josh Pantony:

Yeah. Even when you think about the number of employees to revenue at Anthropic or in OpenAI. I mean, OpenAI was making billions in revenue with like a few hundred employees or something like that. So, definitely, any situation where you have a huge amount of venture capital pouring in is a situation where there’s huge pressure to effectively expand at all costs and not care about capital efficiency.

And we certainly feel that very directly. I’d say the flip side though is that you can build so fast with a smaller set of people that if you’re strategic enough in thinking through the direction you want to go in what you want to build the market, you want to capture that it’s possible to capture absolutely massive upside with way less capital than you used to use. So you’ve got both things happening at once.

Mark MacLeod:

Maybe you’ve seen the clip where the CEO of Microsoft talks about how SaaS is dead. And you told me something once that scared the crap out of me, which is that you felt like UIs would auto-generate in a few years. I don’t know. What’s your prognosis on SaaS?

Josh Pantony:

Oh, did I say a few years? That was way too conservative. It’s happening now by the end of this year. Yeah, it’s interesting. I mean, you’re starting to see this a little bit with Repplit and Lovable and some of the other products like that. But fundamentally, where we kind of see the world going, it’s going to get like 80 to 90% cheaper to produce code.

You’re going to increasingly see these agentic systems that can kind of generate code on the fly. It’s going to mean the cost of software is going to go down to zero. And I think fundamentally, the kinds of SaaS products that we’re used to are either going to change or just go away.

I’m obviously highly focused on my industry, so if we just for a second think about finance, what is the biggest player in finance right now? It’s the Bloomberg terminal. What is the Bloomberg Terminal? It’s 15,000 screens that have all been hand-coded, a specific front end, and a specific back end.

These screens are probably using anywhere between 5,000 to 10,000 lines of code on average. Collectively, across all of them, it’s billions of lines of code. But already with the technology we’re building, I think about 20 to 30% of those screens can basically be hand-generated on the fly as the user wants it.

I think, honestly, in the next three months we’ll be able to do about 80 to 90% of it. And you know, by year-end, you’re talking like the majority of the terminal. And I think it’ll be quite obviously a better experience because with the terminal, you have to learn the exact screen, you have to learn the UI. It’s this thing that’s been around since the 80s.

Systems like ours will be able to generate the screen you care about. It’ll learn who you are as a personality, figure out what you care about, and generate on the fly. And it’ll be an intuitive experience because it’ll all be the exact same thing. You wouldn’t even think about it. You just kind of show up, and there’ll be a UI with whatever you want to be generated.

And so today, you’re starting to see that technology is applied for the product managers or for the engineers. There aren’t too many that are going out to the end users yet, but it’s really, really close. Literally, I think in the next few months. And by the way, I’m describing what’s going to happen in our system, I don’t think it’s very long before you start seeing OpenAI and Anthropic and systems like that basically generating software on the fly for everyone.

Again, they already do it a little bit, right? If I go into Anthropic, it’ll generate a little piece of code, and it’ll generate a little UI. They create those COD little artifacts to give it a few more months, and I suspect all of those are going to generate whatever code you care about whenever you want, and just host it.

And yeah, I see that basically consuming the entire software ecosystem in some way, shape, or form. Again, there will be all kinds of little cool things that’ll come out of it. But I fundamentally agree SaaS in the form that it existed before is dead. And we almost see it kind of going in a different direction. Service as software is totally different from what came before.

Mark MacLeod:

That is truly fascinating. I don’t know what the aggregate market cap of just publicly traded SaaS companies is, but it’s in the trillions. They better figure this stuff out. That’s really, really fascinating.

Josh Pantony:

Some are, by the way. There are definitely some that are really intuitive and ahead, but I think mostly it’s going to be startups disrupting incumbents because startups can just move so much faster.

Mark MacLeod:

You know, I had Tobi from Shopify on the podcast last year, and a couple of things come to mind. One was that you talked about how much you just love playing with technology. And Tobi told me there are days when I wish I were still just a CTO. It’s just like the thing that he loves so much. And then I had Glenn Coates on the podcast recently. I had helped Glenn sell his business to Shopify, and now he leads product for Shopify Plus.

And he told me that Toby is just spending inordinate amounts of time in AI. You know, any new model comes out, he’s figured it out, and then he’s like, the very next day, hey, why haven’t we implemented this? Like, I guess that’s just what a SaaS CEO needs to do these days, is become very intimately familiar with AI and figure out how to disrupt themselves.

Josh Pantony:

I actually, against the advice of many people, started coding one week a quarter.

Mark MacLeod:

Oh, amazing.

Josh Pantony:

Primarily to understand what was possible, and then in some cases to build prototypes to sort of show my team what was possible. I think it must be very challenging being a non-technical CEO, navigating the environment right now because the pace of change is so high and like I said, when things are moving as quickly, exponentially as they are, when they’re accelerating as fast as they are, a single quarter misstep can fundamentally ruin your business.

It’s challenging for me as an AI native who’s been operating in the space for years. I can’t imagine someone who’s not a technology native trying to keep up with it. It feels like it’d be very hard.

Mark MacLeod:

I know you’re vertically specific, so I think I know the answer to this question, but I’ll ask anyway. What is your frame of mind when OpenAI or Anthropic releases a new model? Are you nervous, excited, or indifferent?

Josh Pantony:

I think it’s really important for anyone building a business right now to set themselves up in such a way that 99% of the time when Anthropic or OpenAI or someone else releases a model, you’re excited. If you’re getting existential dread every time there’s an improvement to their system, you will probably not have a durable business.

And so, to the best of our abilities, we try to understand where we think the technology is going over the next 3 to 12 months. I actually don’t believe anyone, including OpenAI and other leaders, knows where it’s going to be a year from now because all of us were seeing crazy disruptions we didn’t expect.

But I think you can kind of understand where it’s going over the next 9 to 12 months. And I think it’s really important that when you’re building a product, I always ask my product team this: Does this still make sense when LLMs are 10 times more intelligent than they are today? And so I get excited mostly for us, we’ve got a reasonably good mapping of what our user needs are, we’ve got a reasonably good mapping of where the gaps are, we’ve got a reasonably good mapping of what kind of intelligence we need to unlock those needs.

And when a new model comes out, there’s a new set of user needs that have been unlocked, there’s a new set of things we can do, and we’re able to sort of step into that. The way I like to describe our technology is, fundamentally, we believe that the people in finance using the best technology are going to win, and we provide the best technology in finance.

So at all times, we need to make sure that when a new model comes out, we’re able to sort of surf that wave and encapsulate it in some way, shape, or form. There are ways to build moats. I don’t think they’re going to be as durable as they were 10 years ago, but there are ways to build moats here. And there are ways to think about things in a way that’s fundamentally different from what the big providers are doing. And I think we’ve been reasonably successful at doing that.

Mark MacLeod:

Amazing. So let’s take that statement. The players in your vertical will win based on the best tech you have. So what’s your vision for Boosted? When you’re done and you look back in the rearview mirror, what impact have you had? What have you built?

Josh Pantony:

Okay, that’s a big question. I think the easiest way to unpack it is to actually ask a question about what Wall Street is ultimately. What is the stock market? And to me, the stock market, if it’s functioning well, is actually one of the most important collective intelligence that’s been built in human history. It’s the system that decides the most efficient allocation of CA capital. It’s the system that decides what technologies, products, and companies should be developed in order to effectively create the maximum amount of wealth for society as a whole.

And so, fundamentally, when I started this company, the idea was that if we could make Wall Street more efficient, we could make the capital markets more efficient and have a fundamental effect on all of society. So 10 years from now, I would like to be a company that is incorporated into every professional manager’s workflow. I’d like to be a part of the decision-making system in which everything is made.

And the sort of net effect on that society as a whole is an even further acceleration of, effectively, wealth creation. I want to be the sort of cyborg element of the brain that is guiding the best decisions on where to allocate capital and how to grow the economy.

Mark MacLeod:

Wow. I love it. Now, let’s say you have that impact. We talked about moats earlier. You talked about moats. In a world where every investment manager has you, does that mean alpha goes away, or does it mean they get more alpha? Or is it just like a race? I got to use the alpha before it disappears?

Josh Pantony:

Yeah, yeah, yeah, yeah. So I love that question. And probably one of the earliest questions I was asked from the very first time I had the idea. And the way I’ve always thought about it is that the early users of technology in this space gain an alpha advantage. Eventually, though, it reaches a point where it’s no longer an alpha advantage. It just becomes a high watermark. You need it to be a professional.

If we were to again use the Bloomberg terminal as an analogy, I did work there over the years, so I use it for a lot of analogies. In the early days, the Bloomberg terminal gave unprecedented visibility into the bond market. It gave you an alpha advantage because you could better understand the pricing of the kinds of things that you were buying versus your peers. Today, nobody thinks of the Bloomberg terminal as an alpha advantage. It’s just fundamental to how you do your job.

Mark MacLeod:

Stable stakes.

Josh Pantony:

Stable stakes, exactly. I see our technology going the same way. Ten years from now, it’ll be so ubiquitous, so common, and how everyone thinks about the world that you just won’t even consider doing your job without it. I think the kind of advantage you’ll get will be less than the early movers. The first movers get a huge advantage over everybody else. But I think when it becomes ubiquitous, you just need it to even function.

Mark MacLeod:

So that’s a giant vision, right? Giant opportunity because you have a very large, rich TAM. So, what do you think about fundraising for this company? You know, it used to be again, SaaS businesses, it was sort of meat and potatoes, right? There was some correlation between the amount of capital you raised and your valuation and your revenue, and it was set, functioning, but not exponential. Whereas with some AI companies, it’s almost in the early days, no correlation between their revenue and the capital that they raise, and it’s just an absolute arms race. So, what do you think about that?

Josh Pantony:

Yeah, I actually like the analogy of an arms race here. So I talked earlier about how I see software getting 80 or 90% cheaper. And I also talked earlier about how if you look at something specific like the Bloomberg terminal, I think there’s going to be agentic systems. They’re going to be able to reproduce like 98% of what the terminal can do in the next six months. Like it’s happening imminently. Our system will be able to do it imminently, and I think we’ll build a new system. There’ll be other systems that can do it too.

So that’s observation number one: the incumbents are going to get massively disrupted. Observation number two is that I’m very convicted it is going to be a startup that does it. The reason I’m convicted it’s going to be a startup that does it is because we do releases twice a week. We are constantly iterating, constantly Learning from our users, and so we can grow exponentially.

When I was at Bloomberg, it was like a one-year development cycle. We built something in a cave and showed it to users. Sometimes we had good releases, sometimes we had bad releases. We’d go back and iterate on it and then build it again. There were tons of projects that went five or six years without getting released when I was there.

I just don’t think companies like that are going to keep up with the pace of technology that you’re seeing. So, observation number one, the income is going to get smashed. Observation number two, it’s probably going to be a startup. So then, how do I think about my company in relation to that? The third observation is we noticed that when people get used to using our product, it’s really hard to get them to switch. But the same is true of a competitor. If someone is using a competing product, we have to be exponentially better to get them to use our product.

This is probably one of the most conservative spaces in the world. People do not like changing their workflows. And so that leads me to the conclusion that there’s going to be sort of one company that’s going to be a leader and capture the entire space, and that company will be one of the most important companies in the world.

And then the second company might get 13%, the third company might get 10%. And by the time you get to company number four, number five, they’re just eating crumbs. This is kind of how it works today with Bloomberg, S&P, etc.

And so for me to earn the right to raise large amounts of money, I need to show a credible path to being that leader, to being that one who’s going to capture the entire space. And of course, revenue is a part of that story. But I actually think user adoption, growth, the ability to capture as many people, and the ability to actually build differentiated offerings, those matter just as much, if not more.

I think the situations in which you’ve seen these really large evaluations, in some cases, and a pre-revenue situation, it’s because they were able to sort of capture the imagination of the VC about this is what the future can look like, and this is how we are that future.

One of my favorite examples is actually Devon AI, where at the outset, I think a lot of people in my space were a little bit critical of the amount of money they raised. But the reason I think Devon raised so much capital was because they showed what an agentic system could look like. They were one of the first ones to sort of unlock that UX and credibly show, here how an agentic system could credibly do some of the work that a coder does.

I think there are going to be lots of other products like that where the power of the product, the ability to sort of capture the imagination, the ability to capture an entire space, will create a massive incentive to buy at a discount relative to future valuations. And that’s how you raise money right now.

Mark MacLeod:

Amazing. I wanted to touch on that conservative aspect of your market, so maybe this is a great final question. And by the way, I feel like this could have been a two-hour conversation, so we may have to have you back another time.

I realize there are many different ways to slice and dice your market, but you know, at the biggest end, the JP Morgans, the Bank of Americas, the BlackRock’s huge trusted brands, like what is their readiness to embrace this kind of innovation?

Josh Pantony:

Yeah, so it varies. I’d say that the really interesting thing about the space is that they can be a little bit conservative on the starting gun, but when they start to move, they tend to move all at once, all very quickly.

And that’s something we’ve seen. I’d say in Canada actually, ironically, all the big banks seem to be moving all at the same time in a very big way. You know, in the US it really…

Mark MacLeod:

Very Un-Canadian.

Josh Pantony:

It is, and it’s something I find interesting. In the US, I think there have been very large hedge funds, pod shops, and groups like that. But I think what happens really quickly is that hedge funds have a huge amount of pressure to move fast, be ahead, think about information asymmetry, and other things like that. And then once you get ubiquitous enough, it creates more momentum and pressure on the rest of the professional managers.

So we saw, kind of in the beginning, a lot of these internal AI committees pop up. We saw a lot of extensions to the kind of procurement process and things that you would need. I’d say kind of over the past six months a lot of those firms have finalized those processes and if anything we’ve started to see kind of the opposite where there’s now top-down pressure to adopt and it’s the top kind of trying to force the more conservative portfolio managers and analysts to use it.

And you know, where before, if you were a sort of tech-forward portfolio manager, tech-forward analyst, that was seen as a sort of risky thing to do. Now it’s the opposite. If you are someone who’s more, let’s say, tech reticent and AI reticent, that is seen as the more risky move.

And so I think that’s really fundamentally shifted the adoption curve and what we’re seeing in the market.

Mark MacLeod:

Fascinating. So do you think if I’m a young business grad looking for an analyst job that’s going to go away in a few years or…

Josh Pantony:

I think the jobs are just going to get way more fun. Way more fun. You know, I’ll pick on a few things here. But like, if you’re starting as a young investment banker, you’re spending like 80 hours of your week generating, let’s say, not the most interesting presentations that have been made in history. That kind of work’s going to be like 80% automated.

If you’re a young analyst, the kind of job of writing a squad analysis reading through an 8K, or trying to build a financial model. These are all examples of things that are going to be automated. So then what’s left of those jobs? Well, there’s a huge part that can’t be automated, and that’s the human-to-human interaction.

A machine will not be able to have a conversation with management as well as an analyst will. A machine will not be able to sell or pitch a deck to clients as well as a junior investment banker will be able to. A machine will not be able to eat the latest pizza from the company you’re covering. Computer AI will not be able to attend conferences.

There’s this huge amount of information that doesn’t exist in written form that really is like human-to-human communication. And that’s going to be the new bottleneck. And so I’m actually super excited. If I were a young grad, I’d absolutely get into this.

Just be the first guy to start using AI in your field because AI will do all the boring work that you didn’t want to do before. And the things that are going to be left are the most delightful, awesome parts of the job. And people go into IB banking because they love meeting executives, they love flying to different locations. That’s the glamorous part of the job, and that’s the part that’s going to be the main focus. So, yeah, if anything, I double down on going into it.

Mark MacLeod:

Very cool. Well, Josh, this has been fascinating. I’ve learned a lot, and like I said, it could have been a lot longer. Thanks for educating me on some of the technical aspects, and I’m sure that our listeners are really going to enjoy this episode. Thank you so much. Awesome.

Josh Pantony:

Awesome. Great to hear.

Hey, thanks for listening to the Startup CEO Show. If you’d like to connect with me, be sure to visit my website at markmacLeod.me, or follow me on LinkedIn at The Mark MacLeod, or X account @markmacleod_, and if you want to tune in again next week, be sure to subscribe on YouTube, Spotify, Apple, or wherever you get your podcasts. We’ll see you next time!