Jared Sleeper on Which Software Companies Will Survive the SaaSpocalypse
By Joe Weisenthal and Tracy Alloway, Odd Lots, Bloomberg Podcasts
Timeline
00:00:00 – Cold Open
00:00:56 – Software Economics Fundamentals
00:04:48 – Guest Introduction: Jared Sleeper
00:07:11 – Historical Software Development Model
00:08:38 – Enterprise Software Complexity
00:11:01 – What Software Companies Actually Sell
00:14:37 – The Scapegoat Value Proposition
00:15:32 – Terminal Value Concerns vs Current Performance
00:18:37 – Data and Context in AI Era
00:21:28 – Platform vs Niche Software Risk Assessment
00:24:03 – DocuSign Case Study
00:25:03 – Bull Case Scenarios for Software
00:27:28 – New Pricing Models and Growth
00:29:48 – Economic Dynamics of AI Reselling
00:30:51 – Model Vendor Competition Risk
00:32:28 – Enterprise vs Consumer AI Adoption
00:37:57 – Need for Layoffs and Cost Discipline
00:39:59 – Career Implications and Adaptability
00:43:05 – Social Skills Premium
00:43:23 – Practical AI Coding Experience
00:45:47 – Hedge Fund Trading Dynamics
00:47:21 – Closing Thoughts
Transcript
Cold Open
Jared Sleeper: The idea of software companies’ value lying in being a scapegoat—essentially, for when things go wrong—is kind of funny and dystopian, I think, in many ways.
Joe Weisenthal: Yeah. I mean, I think it’s a real fear, right? And the way I think about it is: there are two arguments against software right now. One is, the world is going to stay the same, but software is just going to get a lot cheaper over time now that it’s cheaper to build. And I don’t think anyone would argue that it hasn’t gotten dramatically cheaper to build, for reasons we laid out in our deck—and we can talk through it more. We don’t buy that argument.
Jared: I don’t buy that argument. But the second is: the world’s about to get really weird, and the way that knowledge work happens is going to change. And if we think out three or four or five years—who knows if there will even be customer support reps, or sales reps, or software engineers? And I think that’s what’s causing the hit to share prices lately.
Software Economics Fundamentals
Joe: Hello and welcome to another episode of the Odd Lots podcast. I’m Joe Weisenthal.
Tracy Alloway: And I’m Tracy Alloway.
Joe: Tracy, we’re recording this February 11th, and IGV—the software ETF—is down another 3% today. It has been ugly in software. Everyone’s throwing around the term “SaaSpocalypse.”
Tracy: I mean, the great thing about SaaS is there are a lot of things that rhyme with it—lots of puns. “SaaS crash.”
Joe: Yeah, exactly. “SaaS is trash.” Whatever. But I’m looking at the share price of Salesforce — in particular, because I always think of Salesforce as sort of emblematic. The poster child.
Tracy: Yeah.
Joe: Poster child of a software company that I’m not really sure what they do.
Tracy: But yeah, it’s ugly. It’s basically been cut in half, hasn’t it, since its peak in early 2025?
Joe: Right now it’s 184.84.
Tracy: That’s your fault.
Joe: That’s right. Because earlier in the year—after we got back from Christmas vacation or Christmas break—I’d seen everyone playing around with cloud code, and I had to do it. We did an episode, and so people were like, “Oh, if Joe Weisenthal can figure out cloud code, there must not be any value to any of these companies at all.”
Tracy: You know, you mentioned Salesforce—that’s far from the ugly one. I’m looking at Atlassian, which makes a lot of workforce productivity stuff—
Joe: Yeah, like Slack competitors and stuff.
Tracy: That was a $450 stock back in 2021. That’s an $86 stock. So yeah, it’s ugly. And yeah, as you said, everyone is realizing that if any old fool can write software, maybe these companies don’t have much value.
Joe: I mean, I will just say: it’s not just software right now. We’re seeing this rolling series of concerns where every time AI does something—or creates some new product—it hits a particular industry. So on Monday it was the insurance industry, insurance brokers. And today—Wednesday, February 11th—I think it hit some of the broker firms, stock broker firms. And all you have to do is just say “AI” plus “industry,” and there’s a lot of anxiety.
But there’s something that doesn’t make sense to me, or the thing I’m wrapping my head around: sure, any of us could easily write some software, but writing software is a cost center for these companies, right? Like, if you’re Salesforce and you can trivially reduce the cost of building software, that should also benefit you.
And there’s a lot more to a software company than just code generation, because there are all kinds of network effects and linkages. A software company is clearly more than just code. So the fact that code can be generated a lot cheaper does not scream to me, “Oh, these companies are worth less than they used to.”
Tracy: Sure. But at the same time, their pricing is based on that assumption, right? Like there is no competitor for what they’re doing. And suddenly you might have an in-house editor.
Joe: Absolutely. But it’s like—network effects. And do companies want to start building their own payroll software?
Anyway, I have a lot of questions about this selloff. And to your point—no, no, no—this is you doing penance first for causing the selloff.
All right, let’s talk to someone who might actually be able to answer some of these questions for us. We’re going to be speaking to someone who’s been in the software space—an investor in the software space—for a long time, and who recently put out a great deck really diving into the SaaSpocalypse, and what kinds of companies are thriving, and what kinds of companies were struggling even before everyone started talking about AI code generation and all that.
We’re going to be speaking with Jared Sleeper. He is a partner at Avenir, which does growth investing in private companies. Jared, thanks for coming on.
Jared Sleeper: Yeah, my pleasure. Excited to be here.
Joe: Why are we talking to you? Just, you know, for our listeners—apparently this is your first time on a podcast, which is crazy—but give us a little bit about your background investing in software and understanding the space.
Guest Introduction: Jared Sleeper
Jared: Yeah, my pleasure. I think one thing that makes me a little bit different in the investor world is that I’ve spent time investing in early-stage startups, public companies, and everything in between.
I spent a chunk of my career at an early-stage venture fund in Boston called Matrix Partners, working with an OG SaaS investor named David Scott. And then I was also at Coatue, where I ran public software. So I have this experience across the spectrum—from ground-floor startups to looking at the big public companies, which I’ve done for the last ten years.
Tracy: Perfect guest. So give us a sense—give us some color on the mood in software at the moment. Are people, like, hunkering down in their bunkers? How bad is it?
Jared: Yeah, I get texted constantly from folks on the buy side—just retrenching. “I can’t believe this is happening. Can it go lower?” I keep saying, “This is the hundredth time I’ve bought the dip.”
You know, you use “SaaSpocalypse”—like, “SaaS atrophy.” Guys, it’s definitely one of those moments.
And we were talking about this a little bit earlier before starting, but one of the things about software that’s really fascinating is: there are very few folks, even on the buy side, who really understand how software works. It’s one of those Rorschach-test kinds of sectors where no one’s logged into Salesforce and clicked around—much less been a Salesforce admin and understood the full complexity.
And so when there’s panic, there’s not a lot of support for the stocks, and people get scared very easily.
Joe: Explain what this means. So, for example, in a lot of companies: you’re saying the people who invest or trade these stocks, they just know them as financial tables, basically?
Jared: Yeah. They have some idea of their financials and some idea of their customer base, etc., but they don’t have a great intuition for the product—unlike, say, people who use Instagram and therefore might have a feel for Meta, for example.
If you’re an investor in Lululemon, you have a pretty solid conception of what that business is. You can go into the store.
Joe: Exactly.
Jared: You can buy the product, try it yourself. If you’re an investor in Veeva—which makes CRM software for pharmaceutical reps—I bet there are almost no investors in Veeva who have ever been inside the product even once, much less used it day to day and understood how it works.
Historical Software Development Model
Joe: Got it. So I’m going to go way back in time and start at the beginning. Why is it that software like this—payment management systems, whatever—why were they historically not developed in-house? How did we get this model where we have these huge software companies that today have been really integral to a lot of businesses?
Jared: Yeah, it’s a great question. Back in the very early days of software—like in the ’70s or ’80s—there was a lot done in-house. And we’ve seen a very clear shift over time toward using third-party software.
And what it comes down to is: software was expensive to build and maintain, and there’s this need for an ecosystem of integrations around it, which are also expensive to build and maintain.
So if you look at a software company, you can afford to have one, two, three thousand engineers plus partnership teams, etc., all working to build the perfect piece of software for a given application.
And what’s striking—and this will come up a lot more in this conversation—is: they’re not selling it for that much money, right?
A lot of software companies report a stat, which is the share of our customers that pay us more than $100,000 a year—and $100,000 a year is less than half of the fully loaded cost of a software engineer.
So the software model was: build a product that can be applied to thousands of customers, it’s the same product for every customer, and then sell it to them for way cheaper than they could ever hope to build it themselves—even less than the cost of one employee.
Tracy: We could—I’d love to talk long-term software history, even before SaaS and the startups and stuff like that. But like: a lot of the big companies we think of in software, especially pre-Salesforce—whether it’s SAP, Oracle, Microsoft—don’t they have, like, a bunch of third-party companies whose job is just to help install it for you? Like “SAP install.” And that’s a totally separate company because it’s so big and unwieldy and complex that you can’t just install it yourself, or it has to be customized, or whatever.
Enterprise Software Complexity
Jared: One hundred percent. And there are two parts to that which I think are important.
One is the integrations into your existing systems. A lot of big old companies have old databases, old applications, and it’s important for everything to be stitched together. So you need software engineers and consultants to go in, understand those existing systems, and link them up to the new systems.
But the other one—which is probably bigger—is people management and change management. Any software system is a combination of the code and all of the individual users who have learned how to use it.
If you’re trying to change out your CRM at a company, that means training every single sales rep on how to use the new CRM, and getting it right. And if they get it wrong, then you lose deals that quarter.
And so one of the tropes in investing is: if you see a company that’s doing an ERP transition—ERP stands for enterprise resource planning; it’s that core software for accounting, supply chain, etc.—that company is probably going to miss its earnings over the next one or two quarters, because those transitions are so painful.
So yes, there’s a big consulting complex around it that does its best to parachute in the talent that’s required to make those transitions smooth.
Joe: And that tells you something about what makes software so sticky—or at least historically. It’s third-party agents all the way down, I feel.
But actually, on this note: we hear the integration point brought up a lot. And I think the very first episode we did on cloud code, we talked about it as well. But if you have something like cloud code where you can just give it permissions to make changes to your computer—does some of that integration expertise start to go away? Because presumably we’re going to get AI that, at some point—given how fast it’s developing—will be able to plug itself into various systems.
What Software Companies Actually Sell
Jared: Yeah, one hundred percent. I think the challenge of writing the code for the integrations is going away.
That’s not the bulk of the challenge for the majority of integrations. It’s about really deeply understanding the prior system and how it maps to the new system.
And the reality is: within most organizations, that’s a human problem. It’s, “Hey, this column says status 2004. What does that mean? How does that map to the new system that we’re building?” So I have to go talk to someone and understand it.
So there are certain types of integrations where I think they’re effectively solved problems now—you can write in the chat, in the cloud code, and get a perfectly written piece of software to make it happen. And then there are others that are fundamentally human problems because the data doesn’t exist in digital space.
Tracy: Let’s talk more about that, because it really is extraordinary—the degree to which… I don’t know, it’s working code. I don’t know if it’s high-quality code, but certainly these models can generate working code. It blows my mind whenever I use it.
But talk to us a little bit more—like, from the perspective of various software vendors. I’m sure there’s a range in what they’re selling. How much is it code versus how much is it other stuff? And which ones are more exposed to pure code generation?
Jared: Yeah, it’s a great question. And you’re one hundred percent right—it’s producing working code. Frankly, it has been for the last year or so. I built my first Lovable app that was working in production about a year ago, and it’s even intensified in the last three months.
I think when people buy software, there’s a set of things they’re buying.
One thing that’s important for everyone to understand is: open-source software has been a thing, and there have been free, open-source versions of almost any software you could buy for basically all of recorded history.
There are actually public companies that built their businesses by packaging that open-source software, adding a few custom features, and then layering support on top of it. Because when a company is reliant on an open-source database—or a company like Elastic with Elasticsearch, which is an infrastructure tool—and it breaks, they need someone to call: both for security reasons and because it can be very complex and technical, and they need to quickly understand it. So that has been a big part of the story historically: the need to have support.
Another thing you sell as a software vendor is what I call “herd familiarity,” which means everyone on Earth knows how to use your software, which simplifies training and onboarding.
I’ll give a few examples, because I’m sure it’s a new term for listeners since I made it up. Zoom is a great business. Microsoft has been giving away a free version of the product forever in Teams. Why do people use Zoom? Because in certain industries, almost everyone knows how to use Zoom. They have their Zoom set up, they have their virtual background chosen. They’re not going to fumble around for the first minute or two on the call. And that’s well worth the $20 a month for a Zoom plan.
But that applies to lots of other areas, too. Think about Microsoft Excel. You might be able to use Google Sheets to do the same thing, but do you really want to retrain every person who comes in on Google Sheets shortcuts versus Excel shortcuts? It’s not a good use of time, especially when the software is already so cheap.
So that’s another plank in what people are buying when they buy software: standardization and the knowledge that they’ll be able to hire employees who already know how to use it.
And then there are things like brand—again, the ecosystem that comes around it. So it really is more than just the raw code.
The Scapegoat Value Proposition
Tracy: We’ve been joking about this, but the idea of software companies’ value lying in being a scapegoat for when things go wrong is kind of funny and dystopian, I think.
Jared: Yeah. I mean, it’s a real fear, right? And the way I think about it is: there are two arguments against software right now.
One is the world is going to stay the same, but software is going to get a lot cheaper over time now that it’s cheaper to build. And I don’t think anyone would argue it hasn’t gotten dramatically cheaper to build, for reasons we laid out in our deck—and we can talk through it more. We don’t buy that argument.
But the second is the world’s about to get really weird, and the way that knowledge work happens is going to change. And if we think out three or four or five years, who knows if there will even be customer support reps, or sales reps, or software engineers?
And I think that’s what’s causing the hit to share prices lately: this terminal value concern.
Terminal Value Concerns vs Current Performance
Joe: Yeah, it was interesting. One of the companies that’s sort of been caught up in this—would you say “catastrophe”?—I read through their conference call and their CEO was like: not only do we not see red light—do we not even see yellow light—we actually see a lot of green lights. Which I think is really interesting because it fits with this idea of: this year could be fine, next year could be fine, and then the year after that could be zero. Or at least that’s the anxiety—that there’s this terminal value cliff.
Jared: Yeah, I think it’s really helpful. This is our second iteration of the deck, and so we forced ourselves to recenter on what actually happened since the last deck.
There’s a very clear pattern in software. Over the last five years: the pandemic. People freaked out at the beginning, but it rapidly became clear it was an accelerant for SaaS as everyone tried to digitize their companies. So you had a spike in growth rate and net retention. It peaked at just over 40% in 2021 for the median software company—really nice annualized growth.
Then there was a hangover. That slowed down. And we wrote 18 months ago that it reflected the sector maturing. Adoption slowed because most folks had adopted the software they needed under the pressure of the pandemic.
So for the last few years after that, we saw degradation in growth rates across the sector. By the beginning of last year, the median company was growing 18% instead of 40%. So you saw a pretty significant drawdown.
What’s fascinating is: if you look at the actual financial performance of the companies in the last year, it’s been pretty good. That growth rate has held—it was 18% again in Q3. Net retention has also been consistent at about 110%. That’s revenue from existing customers compared to revenue from those customers the prior year. So there’s not a churn issue developing, or a lack of expansion within the customer base.
And a lot of companies are actually accelerating growth, or guiding to accelerating growth. We have a chart showing the number of those companies has increased each quarter over the last three successive quarters.
So there’s a lot going on right now with terminal value. But it’s very hard to argue this is something happening today and showing up in the numbers.
Tracy: The thing is, investors are sharp, right? They’re constantly looking.
Joe: Yeah. I mean, look at Chegg. It went down very quickly in the aftermath of ChatGPT coming out, and that was completely correct. Investors were ahead of that. And of course, for the first few quarters, Chegg management had their heads in the sand. But then it became clear it really was existential to their business.
Tracy: That’s a fun chart. I thought I was looking at a typo because—wow. Chegg was nearly a $100 stock in February 2021. It is now a $0.01 stock.
Joe: That’s right. And you have to give markets credit: the second ChatGPT came out, people were like, “This company’s in big trouble.” They didn’t wait for it to hit the financial results. So there is signal in what people think.
Data and Context in the AI Era
Tracy: I have a bunch more questions, but just briefly: where does data actually fit into all of this? Because the other thing we hear about AI is maybe the models don’t matter that much, but it’s the actual data you have access to.
And I imagine the customers themselves—the SaaS customers—they have their own data. Do the SaaS companies have their own data as well? Can they build off that?
Jared: Yeah, it’s a great question. And we’re here at one of the world’s biggest data companies, so very apt.
Full disclosure: data definitely becomes more valuable in this world.
If you think about a stylized AI model, it could have PhD-level intelligence in a domain. But if you hired a PhD into your company and sat her down on her first day, she wouldn’t be very useful, right? She would have to understand how the organization functions, where things live, “Do I trust this chart or that chart?” “I need access to the Google Drive.” “I need access to Slack.” She needs time to read up.
So what we call this is “context.” It’s all the extra information an AI needs to get something done, no matter how intelligent it is.
And we wrote about this in the deck, but there’s a real question of who becomes that system of context.
And you’re right: a lot of software companies do sit on a pool of very important data. Take Salesforce. CRM is where you track records of every customer you have, every prospect in your pipeline, all your historic interactions with them, notes from sales reps, the status of their account, their customer support requests. It’s an incredibly complex piece of software for a large enterprise.
And obviously, if you’re an AI agent working within a company, you’d need access to that to get almost anything done. But you need more than what’s there. You don’t know what happened at the sales dinner last night unless the rep took really detailed notes—and I can tell you: one common learning in software is they do not take very detailed notes.
Joe: So you had a salesperson, right?
Jared: Yeah, exactly. People assume software management teams know exactly what’s going on, but they’re looking through really messy Salesforce data and doing their very best.
Tracy: And now I’m imagining a sales agent being like, “The Cabernet was exquisite at last night’s party,” and putting in all these irrelevant diary entries.
Jared: Exactly. But a lot of that context lives in human brains. A sales rep meets someone at dinner, gets to know their kids, figures out what sports team they root for, and they’re not automatically pumping all of that into the CRM.
So there’s this race to collect the information that an AI agent would need in order to actually take proactive action. And the software companies have a position there. But there’s also a set of AI-native startups coming in—building actual agents—who are doing their own work to collect that context.
And that’s one of the battles we highlighted in our deck: whoever wins that has a chance to be a really valuable company.
Joe: You know what I think about—and I think you talk about this in your deck—when I think about software, I have this spectrum in mind. On the one hand, Salesforce.com is a platform, and there are third-party developers that build on top of Salesforce. And then I think about something niche—like a company that makes point-of-sale software for dentists’ offices.
They went around giving free payment terminals, joined Y Combinator, signed up 10,000 dentists’ offices, and then those offices pay them, like, $20 a month forever for access to the software.
Well—no—I’m just making it up. But things like that. Is there a side of the spectrum that’s more at risk here? Is that spectrum a legitimate way to think about the industry, or is there a threat everywhere you look?
Jared Sleeper: Yeah, it’s a great question. I mean, certainly in the “world gets really weird” scenario, it’s not clear there’s anywhere immune from threats. But it’s important to think through what it looks like.
I think what’s most at threat is companies that serve enterprises with very customized software already, or software that takes a very heavy implementation. And the reason is: if anyone is going to take advantage of this wave of technology to really advance and replace a core system of software, it’s going to be the enterprises that have the resources and the customization needs.
If you think about SMBs—my dad runs our family’s grocery store. It’s been in the family for 100 years. And he just changed his point-of-sale system for the first time in a few decades, and it was a really messy process. Took a long time.
We love grocery. We love independent grocery. And he’s certainly not going to sit down and vibe-code himself a point-of-sale system and put the store on it. I can guarantee you that. Nor will any dentist.
There’s a chance that someone comes along with a cheaper version, but I don’t think that’s something he’s going to switch to anytime soon. He’s not going to want to go through that pain for another few decades to come.
So it really is kind of company by company.
I’m doing this exercise right now on X where every day I look at a different software company and think hard about: what will it look like for this company?
And it’s really interesting when you press. I’ll give you an example: DocuSign, which I think to most investors would seem like an incredibly simple, easy piece of software. It’s e-signature software—we’ve all experienced it.
DocuSign has more employees today than OpenAI and Anthropic combined, which is a crazy stat and probably reflects that labor is inefficiently allocated across the market. But when you actually double-click into what DocuSign does, there are ways in which it’s very complicated.
Understanding signature regulations in every country around the world—what does it take for a signature to be legally valid? Most of its signatures are done as an API, so folks are integrating it into their own applications.
And there’s a benefit to using DocuSign, which is the brand. People have been giving away free e-signature software for a very long time. But if you’re a company of a certain esteem, you want to make sure your customers trust what they’re signing. If they’re getting a contract from you, you’d rather say “DocuSign” than “XYZ Sign” that someone vibe-coded.
So I think it’s really important to look company by company. It’s definitely a stock-picker’s market, where there are some that are either relatively immune or have a chance to benefit, and there are others that could be in real trouble.
Joe: So is the bull case—or at least the non–sudden-death case for software—this idea that if you have a software company producing something like DocuSign, you’re able to sign documents digitally and track them and share them and all of that… you can build more quickly and efficiently off that base model and provide new versions, new customizations for customers?
So I could do “DocuSign for dentists,” just to stick with that example. I don’t know what specific needs dentists would have—maybe marking up teeth or something. And then I can do “DocuSign for doctors” and “DocuSign for sales agents” or whatever, and just keep going?
Jared: Yeah, I think that’s right. I actually kind of think of it as three cases: there’s the “software gets wiped out” case, there’s the “not much happens to software” case, and then there’s the bull case where the software companies capture a lot of value.
I think it’s a little different than them adding a lot of features and functionality. Frankly, I think a lot of software products today are pretty mature. There are a thousand engineers working on them for ten years, and they’ve built—not all, but most—of the things you’d want to build with today’s technology.
But with agents, there are ways to automate a big chunk of the work.
One software company that’s done this very well is Intercom. Intercom sells customer support software—it’s those little widgets in the bottom right-hand corner of websites. They were the creators of that. They had a nice business, but then they got very aggressive about building out an AI product called Fin, which answers customer support queries on its own.
And they’ve mentioned that it’s almost $100 million of—now on a base that was, you know, like $300 million a year or something like that. And so they’ve really accelerated their business by building an AI-native tool that actually solves the work, not just exists as a tool that humans use.
So yeah, I think that’s the mega bull case.
I think about it almost like the transition from brick-and-mortar retail to e-commerce, where you have a brand new way of doing business and you have a bunch of legacy companies. Some of them will probably just exist as they always have. Others can benefit from the change and add new business lines.
You look at Walmart’s share price—it’s done amazingly well at incorporating e-commerce into its business. And then there are going to be some that are like Sears, and go away.
Joe: That’s funny—Sears reminds me of my dad. He loved Sears because he always said the parking lot was empty when he went to the shopping mall, so he always went through Sears anyway.
In that world, though—so I understand the cost argument: it brings down the cost of code. Maybe you have fewer employees or whatever. But where does growth actually come from in that world? How are you expanding your customer base?
Jared: Yeah, you’re really going to them and saying: we’re replacing human labor, and there’s a different pricing paradigm now.
You used to think of us as something you paid—$20, $30, $40, $50 per seat per month—as a tool for your employees, almost as if your employees are artisans and they’re getting a toolkit to work with.
And now we’re just selling you an employee—or the results of an employee.
So we will sell you customer support tickets getting closed out for $0.50 or a dollar per ticket. And you can do the math of what it would cost you for the human to do that, or what it would cost you for AI to do that. And we’ll be cheaper—but we’re also dramatically increasing what you pay us because we’re cutting into a completely different stream.
And that’s what I think it looks like. We see a lot of exciting examples in the startup space of companies getting much, much higher pricing. And this is a totally new pricing model for software.
Tracy: If you just recorded another episode and the guest teased at that—but talk to us about it. So it’s results-based pricing?
Jared: Yeah. It’s results-based pricing. There are a lot of questions about how it will ultimately shake out.
Fundamentally, what these companies are doing is they are reselling intelligence. The core model vendors—OpenAI, Anthropic, Google—have created a way to get elastic intelligence. And if you have the right data and you can put the right harness around it, you can now sell that to your customers.
What’s an open question is: how do you price that relative to the intelligence?
I was talking to someone this morning who said they think 50% gross margins on intelligence are about right, but we see a lot of variance in how startups are doing it today. Some are getting 80% gross margins on top of the model vendors. Others are getting 20%.
But what’s absolutely true in any case is: if you’re able to do that, you get much, much higher pricing in total dollars than you did before—orders of magnitude in some cases.
Joe: But just to be clear: it can’t be priced so high that the company using the software to produce these outcomes isn’t saving money, right? That’s the balancing act.
Jared: One hundred percent. But think about where software pricing already was.
Think about Salesforce at the elite tier—$80, $90, $100 per user per month. So for round numbers, say $1,000 per user per year—for sales reps who could be making, on average, $250,000 to $300,000 per year.
If you have a technology that can come in and replace a sales rep, you can charge $50,000, still give the customer a 5x ROI, and then you’ve effectively—your take rate on that revenue is much higher.
And so that’s an exciting opportunity. That has people excited in startup land, for sure. If you talk to folks in Silicon Valley, they are foaming at the mouth about the opportunity to really expand tech spend in this way.
And that’s also the opportunity for the software companies that get it right.
Tracy: There must be another risk, too—if you can resell intelligence at, say, an 80% gross margin, then for the model makers themselves they’re like, “Why do we want to be the dumb intelligence?” Like, “We don’t want to be the dumb pipe.”
We saw that in the cloud era: Azure, Google Cloud, Amazon—they didn’t want to just be commodity cloud. They started building features, they wanted to differentiate themselves.
So it must be a risk for the companies reselling intelligence that it’s so lucrative. How are you thinking about the core model makers themselves, and how they’re thinking about expanding into some of these fields rather than just piping in intelligence?
Jared: Look, like in any situation, they’re going to have to make decisions.
Amazon built AWS—they had to decide: where are we going to press, and where are we not? Are we going to sell database software, or are we going to let other vendors do that on top of us? And they made those decisions as they went.
What’s really interesting is: if you look at the foundation model vendors, they have been racing toward the application layer.
Both Cloud Code and Co-Work and OpenAI Codex are applications that people download and use. And I think that reflects an understanding that there is value in getting users used to using your application. Otherwise, you risk being an API that’s commoditized—people switch back and forth between you, and the application vendor has control.
So one of the advantages that software has is this network effect—comfort—software as a security blanket for management.
Joe: Right. But at the same time, people are getting really comfortable with AI—telling it everything. And I keep thinking: if part of the sales pitch for software is that sense of comfort, but then AI is rapidly becoming the thing you talk to for everything… does it eventually become a portal for doing all these different things?
Jared: It’s a really interesting question. And this is where there’s probably the biggest disparity between how enterprise buyers think and how humans think.
I’m sure you guys have seen Claude Bot and the rise of these open-source agents that people deploy for themselves—giving it access to everything, their whole computer, etc.
Joe: That’s Joe. No, I didn’t install Claude.
Tracy: Oh, you didn’t?
Joe: I didn’t install Claude.
Tracy: I’m getting mine set up. Wait—the one with a hammer next to it. I’m really curious. Why not?
Joe: Because of this.
Joe: And it just seemed like a potential waste of tokens and stuff. And then it turned out that for a while, on “my book”—which was the social media for… like, all the APIs were available in a public-facing database that anyone could go read. So it was like a completely open system that had to get fixed.
Jared: And enterprises really do worry about this stuff, and they worry about it for good reason.
Another really interesting example is: there are a bunch of startups that help you record Zoom calls and transcribe them. All of those Zoom calls then become legally discoverable because they’re transcribed somewhere.
So you have VCs in Silicon Valley who will refuse to use them, and you have other firms that are all-in—recording everything that happens across the board so they can upload that into AI as context.
Tracy: I think that’s a really great point. And one thing that makes me wonder is: companies that are willing to skirt the rules, play fast and loose, will be moving much faster over the next two or three years.
And one reason big incumbents struggle is because they actually do have to care about this stuff. They have things to protect. They don’t want to be sued. They can’t handle a major breach. And startups are able to move faster.
Given that: every time software stocks sell off, people say, “Oh, they might go bargain hunting—what’s cheap, what baby is being thrown out with the bathwater?”
And someone—always—says: “Yes, they look cheap, but have you considered stock-based compensation?” And it turns out these companies are not nearly as profitable once you factor that in.
There was a very interesting note from Barclays—I think it was Barclays—that said: European investors are always asking about SBC; American investors only ask when there’s a crisis. I think that tells you something about the difference between Europeans and Americans. I thought that was a fascinating sociological observation.
But tell us: how should we think about the cost side? Because again, if code generation is a cost base, presumably these software companies don’t need as many employees either, and they could pare back.
So talk to us about how we’re thinking about costs inside the software company.
Jared: Great question. And yeah, certainly theoretically true. But aside from Elon cutting 80% of Twitter/X’s headcount, we really haven’t seen any companies take the pill and realize the benefits of that.
The SBC debate has been going on for a long time. I’ve had it ad nauseam over the course of my career. It’s a real expense. You’re issuing your employees stock. They value it like cash. Many of them auto-sell it the day it vests.
And I think the problem it creates for software companies is: management teams are addicted to reporting non-GAAP, which excludes the impact of SBC.
So if you’re an entrepreneur who founded the software business—technical, hasn’t really cared that much about the financial side—you’re a product person, you may think you’ve been doing a good job being profitable because your CFO is telling you, “We’re at a 25% non-GAAP operating margin.” That’s pretty good—when the reality is you’re running break-even. Which is a very common state of affairs.
We looked at the whole universe, and the median public software company has a 5% GAAP net income margin—which is not enough to value companies on.
So it creates this dynamic where, yes, there’s this terminal value concern—which is by far the most important thing—but there’s also no floor.
I was looking at the earnings report from Freshworks, which is a mid-market seller of customer support and IT management software. It trades at one-and-a-half times EV to sales. If it ran at even a 10% GAAP margin, it’d be trading at 15 times earnings—which is a pretty attractive place to be. You could get some value investors, maybe some European investors, interested in buying it there.
But it doesn’t have material GAAP earnings. And on their earnings call there was no real sense of trajectory toward that.
Tracy: And you see the share price now—it’s down over 16%.
Jared: Exactly. And the top-line results were actually pretty good. So there’s a real issue here on the financial side as well.
It’s incredibly disappointing to me that management teams—having embraced this as a way to cut costs themselves—I expect they will. Yeah.
****
Jared Sleeper: One of the advantages that software has is this network effect—comfort—software as a security blanket for management.
Host: Right. But at the same time, people are getting really comfortable with AI—telling it everything. And I keep thinking: if part of the sales pitch for software is this sense of comfort, but then AI is rapidly becoming the thing you talk to for everything… does it eventually just become a portal for doing all these different things?
Jared: It’s a really interesting question. And this is where there’s probably the biggest disparity between how enterprise buyers think and how humans think.
I’m sure you guys have seen Claude Bot and the rise of these open-source agents that people are deploying for themselves—giving them access to everything, their whole computer, etc.
Host (Joe, likely): That’s Joe. Yeah—no, I didn’t install Claude.
Host (Tracy, likely): Oh, you didn’t know? I’m getting mine set up. Wait—the one with a hammer next to it. I’m really curious: why not?
Host (Joe, likely): Because of this.
Host (Tracy, likely): Yeah, because of that.
Host (Joe, likely): And it just seemed like a potential waste of tokens and stuff. And then it turned out that for a while on “my book”—which was the social media for… like, all the APIs were available in a public-facing database that anyone could go read. And so it was like a completely open system that had to get fixed.
Jared: And enterprises really do worry about this stuff, and they worry about it for good reason.
Another really interesting example: there are a bunch of startups that help you record Zoom calls and transcribe them. All of those Zoom calls then become legally discoverable because they’re transcribed somewhere.
So you have VCs in Silicon Valley who will refuse to use them, and you have other firms that are all-in—recording everything that happens across the board so that they can upload that into AI as context.
Host: I think it’s a really great point. And one of the things that makes me wonder is: companies that are willing to skirt the rules, or play fast and loose, will be moving much faster over the next two or three years.
And one of the reasons big incumbents struggle is because they actually do have to care about this stuff. They have things to protect. They don’t want to be sued. They can’t handle a major breach. And startups are able to just move faster.
Given that, every time software stocks sell off with this, people say, “Oh, they might go bargain hunting.” They say: what’s cheap, and what baby is being thrown out with the bathwater?
And then someone—always—people say: “Yes, they look cheap, but have you considered stock-based compensation?” And it turns out these companies are not nearly as profitable once you factor that in.
It was a very interesting note from Barclays—I think it was Barclays—and it said: European investors are always asking about SBC; American investors only ask when there’s a crisis. I think that tells you something about the difference between Europeans and Americans. I thought that was a fascinating sociological observation.
But tell us: how should we think about the costs? Because again, if code generation is a cost base, presumably these software companies don’t need as many employees either, and they could pare back.
So talk to us about how we’re thinking about the costs inside the software company.
Jared: Great question. And yeah, certainly theoretically true.
But aside from Elon cutting 80% of Twitter/X’s headcount, we really haven’t seen any companies take the pill and realize the benefits of that.
The SBC debate has been going on for a long time. I’ve had it ad nauseam over the course of my career. It’s a real expense. You’re issuing your employees stock. They value it like cash. Many of them auto-sell it the day it vests.
And I think the problem that it creates for software companies is: management teams are addicted to reporting non-GAAP, which excludes the impact of SBC.
So if you’re an entrepreneur who founded the software business—technical, hasn’t really ever cared that much about the financial side—you’re a product person, you may think you’ve been doing a good job of being a profitable company because your CFO is telling you, “Well, we’re at a 25% non-GAAP operating margin.” That’s pretty good—when the reality is you’re running breakeven.
Which is a very common state of affairs. We looked at the whole universe, and the median public software company has a 5% GAAP net income margin, which is not enough to value the companies on.
So it creates this dynamic where, yes, there’s this terminal value concern—which is by far the most important thing—but there’s also no floor.
I was looking at the earnings report from Freshworks, which is a mid-market seller of customer support and IT management software. It trades at one-and-a-half times EV to sales.
If it ran at even a 10% GAAP margin, it’d be trading at 15 times earnings, which is a pretty attractive place to be. You could get some value investors, maybe some European investors, interested in buying it there.
But it doesn’t have material GAAP earnings. And on their earnings call there was no real sense of trajectory towards that.
Host: And you see the share price now—it’s down over 16%.
Jared: Exactly. And the top-line results were actually pretty good. And so there’s a real issue here on the financial side as well.
It’s incredibly disappointing to me that management teams, having embraced this as a way to cut costs themselves… I expect they will.
Host: Yeah. Talk to us about this specifically: do companies need—are we going to see big layoffs across the SaaS space in the near term? And what do you think is the time frame for that?
Jared: Great question. I think we will. I think we’ve seen that management teams respond to price signals.
If you look at the history of the sector, it was in 2023 when there was a round of layoffs, and companies showed margin—and then they’ve kind of resisted it for the last two years.
The thing about it is: layoffs can help you move faster, right? I think if you look within any company today, unfortunately there is this spectrum of employees and how fast they’ve adopted AI—whether they’re still doing things the old way, or they’re on Cloud Code / Cloud Co-work and changing the way that they work.
And the employees who are on the lower end are actually slowing you down as a company. They’re not even zero marginal product—they’re negative marginal product. There’s just been such a change in how you work, especially in software development.
So I think management teams are going to realize there are two benefits to layoffs. In addition to the obvious pain of it and the human cost—which I never forget to discuss—one is saving money and showing shareholders you’re financially disciplined, probably seeing your share price stabilize, especially if you’re trading at some very low multiple.
And the second is moving faster.
And also, almost as importantly, being able to pay your top-performing employees. The war for talent in Silicon Valley has never been more intense.
I was talking to a private company we invested in, and they’re losing employees left and right to these high-growth AI companies who can afford to pay huge comp packages in both equity and stock. And you want to keep your good people. You don’t want these AI companies to pluck away all your best people and leave you with folks who are relative Luddites.
So I do think we’ll see this. It’s very sad that it will have to happen, but it’s the obvious path forward for the sector. And I think if done right, it accelerates innovation.
Host: I have a tangential question on that note. Whenever we talk about technological disruption, people bring out examples like: remember when actuaries—remember when Excel—was basically actual people sitting with papers in front of them doing the math. Those people didn’t disappear when Excel got created, but they started doing new things.
I imagine a lot of people are very interested right now in alternative careers for basic, commoditized coders. What do you think it actually looks like? I feel like you might have some insight here.
Jared: The alternative? Well, I think there are two ways to answer the question. There’s: what do you do if you want to stay a coder, and then there’s: what are the careers that are going to still exist over time?
If you think about what’s happening to coding, it reminds me a lot of civil engineering. It’s kind of a funky example, but civil engineers used to work pen-and-paper doing calculus—will this bridge hold up or not? That’s been obsolete for a very long time. All those calculations are done by a computer.
They’re kind of clicking and moving. They go on site, they collect some data, they talk to stakeholders, and they’re effectively project-managing this computer that can do the physics part of their job for them. It’s important they understand the physics in case something looks strange, but they’re not doing much physics.
That’s clearly where software engineering is heading in the near term. In a lot of companies, it’s already there.
And these companies are still hiring software engineers because that job is valuable. In fact, each individual software engineer is way more productive than they were before.
And there’s happily elastic demand for software—we’re still undersupplied with software in the world. So there’s quite a bit of room to add those.
So I’m not necessarily bearish on demand for software engineers, at least for the next three to five years. Beyond that—if things get weird—hard to tell.
But for people more broadly, I think the best advice is adaptability: constantly trying and testing these tools, making sure you’re staying at the cutting edge, and being aware of what’s human.
In my work in venture investing, there’s a lot of data that comes out of human relationships that an AI wouldn’t have access to. I can’t call a friend at another fund and ask how a company’s doing—not yet at least. You have to make some friends first.
Host: They’re talking to each other on what—Book?
Host (Joe/Tracy banter): Right, they’re talking on their book. So maybe if there’s an AI agent from Sequoia and an AI agent from Andreessen…
Jared: It’s intriguing for about five minutes.
Host: Yeah. It was evocative, but—also there was that Wired article about the guy who infiltrated as a bot and pretended to be a writer. It was pretty obvious. They were like, “Oh, are we? Let’s create a new language just for us.” Like they’re not making new languages, right?
Jared: But yeah, I think the rough mental model is: if there was any effort to outsource your job to India, that’s risk—because that tells you the job can be done by someone who’s not physically present in a space.
And if you like working on problems in isolation, not socially with other people—grinding out math problems or little coding assignments—that’s probably also a pretty tough place to be.
Host: Yeah. It’s going to be a more social world. This is something we’ve touched on before, which makes me kind of sad: the edge in the AI world becomes sociability, right?
And to some extent, we talked about this in the context of… look, Max, I know you love it, Joe, I do not—can I say—
Two little observations from my time vibe-coding in 2026 that are interesting.
One: I have zero technical background, and I’ve been surprised by the speed with which I can build intuitions about when it’s going off the rails—like when it’s doing something that doesn’t seem right.
Like, I joked that vibe-coding is just typing “Make it better,” pressing enter over and over again, and then hitting yes when it offers to do something you actually should do. But you can start to build an intuition fairly quickly for when something doesn’t make sense.
And the other thing—this relates to trusting the AI—I’m having a lot of documents get annotated, and I do that through the cloud API, which runs up the bills a little bit.
One API run was going to cost like [amount unclear]. And I was like, “Is this a good thing?” And I stupidly ask it a lot: “Is this a good thing?” It’s like, “Well, when you’re done, you’re going to have this annotated asset that no one else has done and that will be very—” But it was sort of useless what I did.
So you shouldn’t always—like it’s selling yourself. So I was like, “Oh yeah, here’s the API, Joe. Run this—annotate all these documents.” It wasn’t actually a good use of my time. So you can’t really always… they’re just going to sell—
There’s one other set of sectors I’m interested in. You see companies like Moody’s or Fair Isaac… [unclear phrase] global… index.
And they’re selling off too. And it’s not like—this is another area where, you know, fund managers for a long time, unless things get really weird, are going to benchmark themselves off the S&P 500 for a long time. Or lenders are going to be using the FICO score for a very long time, etc.
Intuitively, this would strike me as a very hard thing for AI to replace.
Jared: I share your intuition. I can’t say I fully understand the sell-off in these companies. I wonder if there are parts of their businesses that are more services or consulting.
Often there are combinations where—like, I don’t think anyone suspects the S&P 500 is going to be displaced as an index. Yeah—maybe.
But look, we’re in a world where folks are very happy to shoot first and ask questions later once AI risk comes out.
Host: Actually, going back to your hedge fund time: how much is it just the nature of hedge fund traders right now, where there’s very little stomach to take downside risk and appear to look stupid for missing it, holding the bag?
How much do you think that’s contributing to some of these moves?
Jared: It’s a great question. I won’t speak to my firm, because I think “tiger cubs” like it are a fairly small share of the overall market in dollars.
But if you look at trading volumes, the pod shops—Citadel, Balyasny, Millennium—are a very large share of the volume. And yeah, those people can’t afford drawdowns.
And the scary thing about this for them is: because it’s not fundamental—because the companies aren’t struggling themselves—they have no idea when it will stop.
So you’re left predicting this thing, and you’re like: “Well, I bet my career that people are going to feel better about software companies in three to six months than they do right now.”
And you’re one OpenAI model release or Anthropic model release away from more fear.
So I do think there’s a lot of short-termism right now. And again, come back to the SBC point: there’s also no valuation support—no real valuation support.
In normal times, if companies were like this, they’d be buying back stock, shrinking the share count, issuing dividends.
I have a friend who works at a mutual fund where there are a lot of dividend funds that would love to buy dividend-paying companies growing 10% to 15%—like a lot of software companies—but they’re not. And so you’ve kind of lost the ability to put an actual floor underneath valuations as a result.
Host: Jared Sleeper, thank you so much for coming on and explaining how software works.
Jared: My pleasure. That was super fun. Thanks, Tracy.
Host (Joe): I thought that was really interesting. I’m fascinated by this idea that in the short term, most of these businesses are doing fine. In the long term, they might go to zero. But also in the short term, they’re not really doing fine because they’re not making much money. I guess that makes sense that they’re all selling off right now.
Host: Yeah. I keep thinking this is probably a stretch, but I keep thinking back to that book Bullshit Jobs, remember? The argument there is a lot of jobs exist not because people are doing anything specific, but because they provide some sort of social value.
So for instance, you have a person who is essentially the designated scapegoat for senior management.
And I keep thinking business is basically an ecosystem of different players. So it might be that in the new AI world, the role of software companies changes—their social role changes. And I don’t know what the price or valuation looks like on that.
It still feels like to me… I know, I know, I have no idea how businesses are going to buy software in the future.
I did think that was really helpful. I really don’t know anything about how the software business works generally, so I found that very helpful.
One other interesting thing: even high-end software is not that much money, right? If you have a salesperson making $250,000, what is $1,000 a year from Salesforce to do their job?
And also, given that free and open-source software has existed for a long time—still, you want to pay an implementer, a company that manages the transition, getting from here to there. That really changes the nature of software buying.
Does it feel like you have to get it to—this is weird territory—but maybe things are good? I think things probably are going to get really weird.
Host: Yeah, I think that’s a pretty good bet. Like, if you bet on weirdness—there is a weirdness index. Someone should build that weirdness index. That would be a pretty good investment.
Shall we leave it there?
Let’s leave it there. This has been another episode of the Odd Lots podcast. I’m Tracy Alloway. You can follow me @tracyalloway.
And I’m Joe Weisenthal. You can follow me @thestalwart.
Follow our guest Jared Sleeper—he’s @JaredSleeper. Follow our producers Carmen Rodriguez @carmenarmen, Dashiel Bennett @dashbot, and Cale Brooks @calebrooks.
For more Odd Lots content, you can check out our daily newsletter. You can find that at bloomberg.com/oddlots.
And if you want to chat with fellow listeners about all of these topics, including AI, check out our Discord: discord.gg/oddlots.
And if you enjoyed this conversation, then please like or leave a comment on the video. Or better yet, subscribe.
Thanks for watching.
