72. OpenClaw: Analysing the hype

Jonathan: Welcome to The Difference Engine, the show for tech founders, investors and innovators. Paul: Hey Jono, what have we got coming up today? Jonathan: We're gonna highlight the Europeans leading the AI charge, Paul: and we'll sift through the debris of the so-called SaaS apocalypse. Jonathan: But first, let's wrestle with Open Claw, the latest piece of red Hot AI property.

But we have to ask the question, are we gonna get burned? So let's kick off this week with the rather thorny, or should we say snappy? In fact, question of. Open claw now.

Open claw. Is it a fashion, is it just a fad or now just a feature or frankly, something bigger than that. So it is a phenomenon, unlike many we've seen, um, you know, we've been around a long time. So that includes the internet, the growth of the iPhone, uh, even chat GPT in, in in the last year.

You know, it is. Is this thing around Open Claw Four seasons in one day? Yeah. I mean, frankly, it's more like three names in one fortnight.

What on earth is going on and why people speculating. It's just been snapped up by Sam Altman and also apparently a big run on the West Coast on little Minimax to run it. So perhaps you, you could, um, uh, enlighten me and presumably the listeners with, with just tell us. Paul, what is Open Claw?

Paul: Yeah, so Open Claw is some software that allows you to take tasks that you've been doing, sometimes very disjointed tasks in a business, and give it the keys to all the software for the various bits of the process you need and have it just, just go. Do that 24 7 without a break. Uh, when you're asleep. When you're awake and just grind away at the issues.

Uh, and people reporting that. It's been absolutely amazing for automating entire processes of work, depending on how much of the height you buy. This thing is the actual installation of what is gonna get rid of people. Um, let's start with the naming first.

Originally called Open Claude. The people at Anthropic didn't like that for, uh, obvious reasons. Um, uh, then we had, um, claw Book, then we had Malt book. Now we have Open Claw, which everybody's happy with.

Um, no, Jonathan: no. Nobody's been crabby about that though. Paul: No one's been crabby, although, um. Interestingly enough, the reason it was called, called Malt Book for just a, a couple of days, literally was, um, lobsters, obviously malt.

Um, so let's look at the hype here. You know, there's this hype good and there's hype bad. Um, you get a lot of very excited, um, sort of I would say semi programmers talking about just how amazing this has been for them, how it's changed their life, revolutionized everything they do, and they're gonna be rich before bedtime. All of that stuff.

Until now, chat GT was the, um, fastest growing. Tech and this thing's come along and completely tranced it. Um, someone else is saying that there's a 2 million weekly visitors to the OpenCL website. 2 million weekly visitors, uh, makes it one of the most heavily trafficked open source, uh, AI projects.

So if you haven't. You probably should start, uh, looking at this, however, we've got some friends who's looked at in a bit more detail. Jonathan: Oh, the great Jason Lemkin perhaps always worth the read. Paul: Yeah.

Jason Lemkin of sra. We like him, we love his, uh, event. Um, he wrote, uh, a long form writeup 'cause he is been, uh, on this from the start. Actually not so much just open claw, but also.

Vibe coding. I believe he uses Rept. Jonathan: Yeah. When, when they, every, every new Tech comes on here, they, they seem to use it at Esra just to try it out.

It's really quite amazing that they do that. Paul: I mean, yeah, he's a former entrepreneur. You can understand that. Um, I think it was, uh, I forget which company he sold to, uh, Adobe.

Um, one of the, one of those, um, contract reading things. Um. So he, but he's running currently over 20 agents, uh, to replace his go-to market team at Sutra, which is sad for them. I suppose.

I, Jonathan: I, I bet his sales team are a bit concerned about this, aren't Paul: they? Well, it's no longer their concern. He, um, he did have a 10 person sales team. He now has.

10, 20 plus agents doing their jobs. And in his podcast he says, we replaced our sales team with 20 AI agents. Uh, and then he goes on to talk about what happens. Um, and he, he re he repeats the current model is, um, 1.

2 humans or full-time equivalents, I guess, and 20 ai, a AI agents replacing 10 humans. Jonathan: Right? That's a bit, it's all a bit spicy. I mean, perhaps we should just break down what.

Are the concerns about this? I mean, this is, this is very, very paradigmatic. Paul: This is very early stage, right? So I, I spoke to an expert from one of Thehy cybersecurity firms.

His line was very simple, use open claw when, when Apple has a version of it and not before. And this guy's very, very experienced at, at cyber. So there, there's your warning. Number one, uh, the first thing you gotta figure out if you want to do something with, um, open Chlor is, do you want it?

Run it yourself. Um, or do you want to get a one of these hosting services? Jonathan: Well, well a AI delivered as software. As a service?

Paul: Yes. Yeah, exactly. Jonathan: Oh, that's interesting. Given the current arguments.

Paul: Well, yeah. What are you handing over there? 'cause you have to give it the keys to the kingdom. You have to give it the keys to your software for it to do anything.

AG agentic. Um, and of course all these hosting companies are brand new. If the tech's brand new. The guys that are hosting it are also brand new, so very unproven.

Um, you know, and then people, uh, there's been a run, as you said on Minimax. Um, some people are going the whole hog and getting the really top of the range. Uh, apple workstations to do this are very serious. Uh, but certainly it's caused a little blip.

In the, uh, hardware sales. I also hear from other people, all you need is an old knackered, old laptop. It's good enough. You don't need to go for the, the, definitely Jonathan: got one of those.

Paul: There you go. Um, and, um, the other thing you need to know, because part of the advantage of this thing is, you know, it doesn't go to the loo, it doesn't sleep, doesn't get sick. Are you, have you got somewhere if you've hosted yourself, where you can run reliably some hardware 24 7. Uh, I don't think a lot of us have, um, even though we, we might think we have, and, um, you may run into a couple of, um.

Issues with paying the bill. Jonathan: So another concern might be the environmental impact on this. Absolutely. I mean, clearly everybody's worried about, um, the environmental impact of AI data centers.

Paul: You've got your own one in your Jonathan: house, all these little knackered old machines, you know, running away 24 7, probably heating people's houses. Um, then I think that's something to be concerned about, but it, it strikes me the number one concern on this, uh, is, is security. What, what do we think about that at the moment? Paul: Yeah, so there's a, a big paper that caused a lot, I mean, this thing's moving so fast.

There was a, a paper called The Agents of Chaos. Jonathan: Oh, that's Natalie Shapira. Is it the, the, the MIT guys? Paul: Yes, there's a bunch of them actually.

Uh, that thing had 6,000 reads in 48 hours. And I know you've written a few academic treaties yourself. 6,000. Oh, did I have.

6,000 reads, that's like, what is that? Four, 400 an hour and a 40 an hour. Jonathan: Oh, I, I would've been very, very happy if people had read at that speed for, um, cable television in the uk. Issues in the formulation of the technology policy.

You can still get it on the internet too. It's Paul: number one. It sounds riveting. In, Jonathan: in 1981.

It was. Paul: Alright, so, um, the Agents of Chaos paper came out, um, ages ago now, February 24th, um, ages ago. Uh, but it's caused quiet stir. And, um, you know, some of the things here are really quite scary.

Um, I'll just read a little bit here. Observed behaviors include unauthorized compliance with non-owners, disclosure of sensitive information, execution of destructive system level actions, denial of service conditions. Uncontrolled resource consumption per your point. Identity spoofing, vulnerabilities, cross agent propagation of unsafe practices, and partial system takeover.

Jonathan: Oh, nothing to worry about there. Paul: Other than that, Jonathan: other Paul: than that, we're all good. Jonathan: It's fine. Paul: Good?

Yeah. Yeah, it's all good. It's all good. It's all good.

Go for it people. Um, then there's the, the, the smallest issue of cost. Right, because these things are burning tokens. Jonathan: Yeah.

Yeah. Let, let's not pretend they're free compared to employees. Paul: No. And this is where, it's interesting because a lot of people say AI will replace folks.

Right? Jonathan: Right. But your agents are gonna cost you money. Paul: They also cost you money.

Yes. Yes. It may not be sick pay. Uh, it may not be bonuses, but it's Jonathan: insurance.

No. Paul: No, not even, not even Rachel Reeves touches these guys, but they do cost money. Oh, Jonathan: she'll find a way. Paul: She'll find a way.

Um, you know, we've got Jason here saying that with our agents, we hit $300 a day per agent. Now bear in mind he's got 20 agents. Right. That was, you know, that's a lot.

Jonathan: That's a sort of a hundred KA year per agent, isn't it? Paul: Yeah. You could get a decent person for a hundred K year. Even in the States, Jonathan: you could get a very decent person.

You could probably get. Well, at least two maybe. Paul: Yeah. The guys, um, from the All In pod, they're actually thinking about this as they would do a little bit more strategically and talking about token budgets, Jonathan: right?

Or for the best developers, presumably, oh, well, look, we've got some token budgets here. Ooh, how good are you as a developer? Ooh, you can have more of this budget. Paul: Well, you can or you can have less if it doesn't work out.

The point is, you are, you are managing, uh, an agent. An AI agent workforce as if you are managing a workforce. We all know that's quite difficult. They reckon that to, for an AI agent to be worth it, they need to be twice as productive as other employees.

Now, in some cases they will be, but if they're not. It ain't worth it. And then I'm, I'm looking at this, uh, uh, I saw something on, uh, LinkedIn chap called Richard Lie. Uh, he's a senior full stack engineer.

So you'd think he'd know what he's talking about. Uh, and he's asking the question quite validly, open claw users. What's your monthly API cost looking like? Jonathan: Yeah, I'm, I'm, I'm, I'm also just, yeah.

Wondering. You know, inflation in the cost of tokens could easily out outpace the salary of employees. It's a sort of reverse union effect. Paul: Yes, absolutely.

And of course, we've got not enough history to know what to, what prices at which tokens are going to go up, although we do know. Most of the LLMs are not making money. Right. Jonathan: I I think we do know the basis of capitalism, which is once you, once you've got enough presence in the marketplace and everybody's locked into your solution, then you drive up the cost of it.

Paul: Yeah. So, so talking of cost, Richard Lies said he's seen reports of light q and a and simple tasks. This, that's open clause performing those 10 to 30 bucks a month. Yeah.

Okay. Acceptable. For heavier daily use. Um, this is for coding.

Um, the email automation that everyone is a use case. Everyone seems to like, please find everybody that's asked me this and send them an automatic response back. That's 24 7 working obviously. And that comes, he thinks between six and $15 a day, which, you know, adds up couple hundred dollars a month.

Um, and then runaway automations. This is where it gets creepy or scary. They can be 200 bucks a day and, and so he concludes that some people spend more on their AI assistant than their car payments. It's an American way of framing things, but it's sort of interesting.

Jonathan: Well, hopefully, hopefully the AI agents won't spend most of their time sitting on. On public property doing absolutely nothing, which is, let's face it, what cars mostly do Paul: they do? Right. Uh, Jason's back to Jason's top three issues, and I think these are really crystallized what the problem is here.

Jonathan: Oh, orchestration. I bet orchestration comes into Paul: this. Yeah. Multi-agent orchestration is messy.

So he says, uh, and I think it's incredibly honest of him, but you know, it's a picture into the future for the rest of us. He spends most of his time orchestrating and managing. Agent. Jonathan: Oh, agent hr.

That's what this is. Paul: Ar, I don't know, a IR air. Jonathan: So, so what's the difference between, between that and running a company full of people you spend most of, most of your time orchestrating those people. Paul: That's exactly right.

Yeah. Um, the second thing, and this also anthropomorphize is beautifully, is you need to do constant training, monitoring, and, uh, quality assessment. Jonathan: Yeah. Uh, do they get sort of bitchy when, when, when each agent steps on the other's territory?

They're gonna, they're gonna start interfering in other agents'. Agents' work, aren't they? Paul: They're not very clever, right? So they will just keep doing the same thing unless you continually refine it.

Check for bad outputs. Say that doesn't work. Try again. I mean, you, you literally have to guide them and as he says this, you know, you can't drop an untrained agent in and just expect magic.

He says it takes months of iteration on ongoing human oversight, uh, to reach the performance. They advertise. I mean, sort of, I get that. But at the same time, if you, if you believe the height, you wouldn't think this was an issue.

Jonathan: I, I can't wait for the hierarchical management agents. As, as better agents manage poorer agents, and the better agents have to report to human being. Paul: It's all coming. And the last thing he mentions of the three, so, uh, in addition to the constant training and the multi-agent orchestration is customer experience, uh, and risk control.

So. It's preventing agents. Of course, in his case, he's using this to literally speak to, uh, potential visitors to the, to his event. Um, going off script.

Jonathan: So sorry, when he, very quickly, his agents be speaking to the agents of the people going to the Paul: event. Yeah, of course, yes. And their Jonathan: agents, Paul: but we're not at this, we're not at that stage yet. So that's Jonathan: a loop.

Paul: That's, well, it could be an efficiency loop. So, but, but he's worried, and, and very specifically this rings so true from, from his business, sending the wrong message to sponsors. That's gonna annoy them. Uh, mishandling high value prospects, you know, uh, without being horribly cynical, you know, prospects have different values and the top, top ones you may not want, uh, an automaton to, or even a sacious automaton to speak to.

Um, you know, and that could involve, you know, them making promises that the business just can't keep. Uh, that, does that sound like a junior employee? It certainly does to me. Um, and he frames it all as a trust and brand risk problem, uh, saying the agents can scale.

But one or two bad interactions with key accounts and all of the gains are gone. I mean, that's, that's a stressful situation, right? So open claw, what have we learned here? Jonathan: Well, this is, this is development at break and X speed, the like of which has never ever happened before in human history.

This is the ultimate in move fast and break things. Um, and you know, one wonders what is gonna be broken in this. You know, it, it will, it go far as basically destroying human relationships and trust and, and humans being unable, uh, to talk to each other and debate because all that has basically been taken over by agents. Um, and there's the perennial question here is.

So what is, what is Jason's 10 unemployed salespeople gonna do with themselves? Paul: Yeah, they're gonna need to learn to program the, the, uh, the next instances. Instances. Jonathan: Yeah.

Well, it, you know, that, that's, that's a very positive view. Or we could have a lot of very unemployed, very disaffected, uh, people. And then we won't have much of a society for these new technological, uh, solutions to operate within. Paul: Gosh, that got dark.

Well, let's be a little bit more positive here. Um, this is the part of a category. Cycle where fortunes are lost, but also fortunes are made. It's very high risk and there's clearly some awful mistakes being made right now, and also some great breakthroughs.

Right? Jonathan: Yeah. Yeah. It's, it's probably not revenge of the nerds.

It's probably a revenge on the nerds Paul: in some cases. You're Jonathan: right. You often think, and I think we're gonna touch on this later in this episode, it's, it's, the winners may actually be. A hybrid of pragmatists and techies and people and technology as, as it's always been because I think these, um, you know, these new developments are ultra, ultimately just more new tools and, you know, that's how we human beings apparently differentiate ourselves from other mammals.

By being great tool users, I mean, sorry, I've got, I've got a friendly local, uh, neighborhood, Raven, who would probably argue that actually they're pretty good with sticks Paul: if open is your tool. Um. You know, one way to look at that. This is, this is great news, right?

And it's great news because it's powering on ai. Maybe, um, maybe just, maybe, but, uh, I don't think you've got, I don't think most, uh, folks who wanna keep up have got a choice. They need to get on board. Jonathan: No, this is, this is happening.

It will surface. In daily use in some form. You know, this is a classic example. You can't reco the tech genie it, you know, it's here, it's been discovered, it exists.

Um, you know, it will have its uses. Um, the issue as ever is, is diffusion. And you know, how, when and why will this enter, um, the, essentially the engines of production in our economy. Paul: Indeed.

So get your mini Mac, get out there because fortune favors the brave. Jonathan: Yeah. Give it a go. We've been talking in, in recent weeks about.

How Europe is often misunderstood in terms of the contribution it's made to the global technology environment. Certain commentators are reverting to the really, really, really tired, you know, US innovates and Europe regulates memes, you know, stupid comments about. Ooh, the fact that's in Europe, you can actually have your plastic top attached to your plastic bottle for recycling. That's a pretty damn good thing, I think.

Um, but the reality is, you know, we've talked about where Europe succeeds in tech, um, but we haven't talked about so far is some actual category leaders that are emerging from Europe and there's a load of them. Paul: Yeah. And I think we probably have been guilty of being a little bit dismissive about Europe's role in, uh, you know, in the future of ai. And I think that's possibly because we were looking at, uh, in the wrong places at, at, you know, at the bottom of the stack, LLMs at the fundamental AI way, which is obviously US driven.

Um, you know, a lot of money being spent there. Um, and it looked for a long time as though that huge wave of, of cash of investment was a defensive moat. For the USA, I think that notion is being challenged now, and there seems to be a bit of a stuttering of faith in the US uh, itself. That, that, that its big tech is gonna take all of the pie.

Jonathan: Yeah. Well, of course our, our friends across Lamont, um, are being very hy about it and saying, well, you know, even. The basic enabling technologies of the office or their massive public sector, um, can be replaced by domestic products and supported by domestic companies. Anyway, as, as we've said before, Europe does well and it specializes building a defensive moat out of deep knowledge of the context of the application.

To build an application that you run an economy on, not one that you just scroll through. Paul: Well, scrolling through might be becoming more difficult because of European legislation ironically, uh, without doubt, right? The US infrastructure play, uh, and um, you know, the energy investments, the data center investments. These things look strong.

There's massive amounts of money, uh, circulating, uh, almost literally circulating between those who are breaking actual ground. Those who are renting data centers and rent renting, um, racks, uh, and those designing and installing the AI chips, who, uh, yet again with the NVIDIA numbers are making plenty of money. Jonathan: Yeah, I mean, you know, the history serves. Swell in this, there's, there's plenty of pie left to go for.

I mean, just, you know, checkout companies like 11 Labs, you know, based outta the uk but very global, um, you know, specializing in AI voice generation, which is, you know, a classic level where AI comes to towards a useful application. Um, and that raised, I think about a billion recently at a 4 billion valuation. But, but, and, and, and Sequoia. You know, who you, or you have to take Sequoia seriously in anything they do.

Um, they're one of the grand people of, of venture capital, uh, so they're putting their faith in, in this particular application. And there's a lot, there's a lot more out there. Paul: Yeah. I think it's a feather in all cap billion at 4 billion from, as you say, one of the absolute number one, number two, in VCs, um, then we've got lovable, um, like 'em or don't love, love 'em.

Um. With our friend, Harry, et cetera, Jonathan: vibe, coding for hot shots. I think we can describe that as, Paul: yeah, there may be trouble ahead, but that's okay. Jonathan: Yeah, that's fine.

And this is sort of more foundational level, uh, as is, as is misra, uh, ai. Um, and, you know, they're, they're leading a charge for, uh, foundational models. Whether they can scale to be as big as the Americans. Um, Paul: maybe they don't need to.

Jonathan: Who, who knows? And does it actually matter? Because Europe wins elsewhere. I mean, another example of that would be like Depot.

They're doing something that everybody needs, which is, which is solving the problem of language translation in a, in a very connected world. And, and certainly for Europe, because we speak a lot of languages. Paul: Exactly. Yeah.

Yeah. There's a alfalfa, we talked about those guys before in Germany. Um, maybe gone a little bit off the boil, but, um, they're looking at specializing in sovereign secure. And explainable AI for enterprises and, and, and governments.

Jonathan: Of course, let's not forget, deep minds Fly, fly that Union Jack and, uh, you know, while LF Alpha are looking at enterprise and governments, uh, deep mind are looking at scientific discovery. Paul: Yeah. Also in biology, uh, out of the Netherlands is a cradle. Uh, applying AI to biology, Jonathan: even the gnomes are at it.

So it, it's on a ai, uh, in Switzerland. Um, and guess what? Guess what? They're focused on industrial AI for manufacturing quality.

And if you want quality manufacturing, Switzerland's always been a good place to go and get it. Paul: Absolutely. And then, uh, if you want education, sun Labs out of Sweden, uh, that is specializing in AI for education and, um, enterprise search. Jonathan: Yeah.

And if you wanna get a little bit more consumer, how about photo room in France? And they're, they're. Pretty much the leader in ai, iPad, photo editing. Paul: This is all moving at pace, right?

So, um, last year's, um, 2025 sifted, uh, AI 100 showed at the time that the average European funding was around about a hundred million dollars. So nowhere near the level of, uh, the US and certainly nowhere near the 11 labs funding that we just talked about. Yeah, Jonathan: that's an outlier at the moment, Paul: at the moment. Um, but I guess it raises the question of how much do you need to raise, right?

The rise in solopreneurs, uh, literally folks, um, you know, vibe coding on their own, setting up businesses and uh, trying things out, and the explosion of vibe coding that drove, that means that you can create MVPs or minimal viable products very quickly. Roll them out, try them. Uh, so it could be that the reason, the very reason that you needed the funding at the start, which was to. To buy some kit, to hire some coders, et cetera, is not there anymore.

And so that would mean that big raises are not the defensive moat that many thought they were previously. Jonathan: It's sort of interesting that I, I've noticed in my daily conversations in mergers and acquisitions in, in recent months, uh, a phenomenon arising where we are seeing European companies, um, developing. Very compelling product propositions, um, and a client base on relatively little funding. And the funding they actually need is to break out of the European market into the US and far eastern markets.

So the, the, the money. It's going less into the early development of, of the technology because that has been st starved of funds in the past, and they have had to develop their companies in a very lean way. But if you're gonna try and conquer the North American market, you need serious funding. You can't do it unless you fund it.

So you have a choice. You either get the funded and go to the states. Or you sell out. And you know, we've talked before about that in the past being a phenomenon of European tech development selling out essentially too early.

Paul: Yeah. So, and certainly we've helped people like hyper exponential build their case, uh, for a 16 z when they got the funding, do exactly that. Uh, we've also seen, um, you know, how these things can twist and change a little bit, um, and AI's relatively new, but, um, people forget that back in the.com area, there was some great.

Businesses that were built not simply on tech, they were at the ne nexus of it and business consulting. You think of Accenture, multi-billion dollar, very successful business, and the big Indian players who built off the back of the technology, they weren't necessarily the technologies. Technologies themselves and certainly a friend of the pod, Hussein Kasai. Uh, think so.

He recently dropped out a LinkedIn post, um, I'll just read a little bit of this now. So he's talking about the, uh, 95% of pilot AI deployments fail. That's an MIT figure that everybody's looking at. Um, but this is the point I think is interesting.

Most teams, he says, are burning cash on the wrong bottleneck. Software is becoming a commodity. There's no shortage of agent builders. And LLM tools promising a hundred percent automation.

Yet projects still stall because companies ignore the real constraint, the people who know the business and the work of wiring that knowledge into the systems that actually run. I think he's onto something here. Jonathan: Yeah, I mean, ai, I mean, I think he says this and it should not come as a surprise to anybody who. Done a product to develop a project to develop anything is, is that things fail at the last mile.

You know, you've got 99% done and the last percentage takes forever, or it completely screws up. And, and in the case of ai, it's when it really, really hits the real world. So, you know, translating messy business logic into workflows, you know, the permissions, the evals, and, and the operating rhythms. You know, you, you can't buy your way through this with, you know, forward deployed engineers.

The, the mid-market just can't, I mean, it leaves a huge gap between buying a tool and seeing ORI. Paul: Yeah. And, and I think it's important, uh, to remember that history often rhymes, but it never just repeats. Um, and I think there's some lessons learned from, uh, for European category designers to, you know, not fold, just because the, the big tech, uh, uh, and the large LLMs.

Happen to be massively funded and based somewhere else, the long-term winners are yet to appear. Right. Um, and there's some early signs of course, that there's gonna be displacement of skills, like, you know, programming, basic enterprise, uh, sales. That's just.

Order taking and low level marketing roles, certainly, you know, why would anyone employ someone to do SEO? It's not a job for a human. I, I mean, and I think AI is not playing out as many assumed, uh, certainly many, uh, who wrote big checks have assumed, and some of them are looking a bit nervous now, uh, as to whether they're gonna get paid back, but it's all still so, so early. Jonathan: It's a bit of, you know, putting moats around data going on at the moment.

But, you know, trying to stay close to human decision makers, um, you know, studying the signals from the market. Yeah, I do. Do you really think this is time to passively stand back and watch? Paul: No, definitely not.

Now's the time for us all to start by coding, to, to test our hypotheses and to look for new opportunities and um, you know, get cooking. Jonathan: Yeah, let's get out there people. So we really couldn't, um, put an episode out. At this time, uh, in history without some reference again to the SAS apocalypse, you know, and, uh, what is it?

The SAS apocalypse? Well, it's about AI disruption. It could be the AI contagion. Is this filling you with excitement or trepidation?

You know, if you've, if you've been building a firm in SaaS and you know, you're, you're not in your first flush of youth. Do, do you think it's, it's time to bail out or are you going full throttle, uh, to, to reinvent your company, to take it to the next wave of value? Um, certainly there's a lot of debate and a lot of confusion. Um, it matters because for years, SaaS companies built huge organizations are in for fun and all while their revenues just grew inexorably.

Now things are changing and with recent falls in evaluation of. What we, we have to call these days old guard companies, I, non I native companies. Are we at the end of an era of SaaS dominance? Now, you, you know, it might be becoming important when the, the, the, the Revered Financial times, um, writes a very, um, learned article, uh, headed by the, uh, title.

Salesforce Chief dismisses SaaS, apocalypse fears of ai, overtaking business software, and, and, uh, that Salesforce chief, as we know, is one of our favorite human beings, uh, in the world of category and tech. The amazing Mark Benioff. Paul: Yeah. But do they have a point?

Um, so Benioff's coming out, uh, swinging here, um, saying that there is no SAS apocalypse. Uh, and, um, you know, he's dismissed it all as, um, talking about. You know, just not being relevant despite the fact that his sales, which are, you know, still 40 plus billion, um, 46.2 I think last time.

Um, and, um, you know, his point is don't write me off too soon. Um, but is he protesting too much because, um, his life's work. Salesforce is in trouble. I mean, he was the original OG category disruptor.

He showed Oracle and everyone else who was around at the time exactly how to do it. He bossed it. He bossed the move from client server to cloud, and now we're moving to AI and maybe he's got a target on his back. Jonathan: Well, all great empires eventually, uh, decline or have to reinvent themselves, um, as we are seeing in in other areas of society and the economy at the moment.

Um, you know, it's, there is that law of big numbers. It's so much harder to grow fast when you're big already, but, you know, he's certainly coming out fighting and, uh, very, very bolshy. Um, this is when you know. There's a problem.

Paul: Well, here's the Salesforce tweet. This is literally what it says, and it's, it's ballsy, right? Salesforce says, they said enterprise software had peaked. We have just posted our biggest year ever.

Funny how that works. Ouch. Wow. Yeah.

And then they've got a bunch of numbers, uh, you know, sorry, 41.5 billion is the actual revenue number. That's 10% year on year. Um, 20%, uh, gap operating margin.

Not bad at that size. Uh, 1.5 billion operating cash flow. I can think of a lot of PE companies who would die for that.

Uh, 29,000 agent force deals now. That's a product that's got agent in it. I'm not sure it's properly agent, but the point is they're coming out swinging here, right? Jonathan: Yeah, yeah, yeah.

I seem to remember, uh, certain, um, hierarchical, uh, databases when a relational came in, um, putting little, little, uh, relational front ends on. So just doing a little bit reporting and somehow hoping people would think they was, they were as good as the fully relational product and they were better to start with, but then the relational, uh, took over. But I think the key issue. Uh, here is, is whether you view AI as a massive disruptive wave sweeping all before it, and rendering today's approach to building and using systems instantly obsolete, or one that Benioff is currently arguing.

Now we know why he's arguing it, because, you know, as the Ari said, you know, how'd you get to to y? Well, I wouldn't start to X, but he is a X. Um, so, but he thinks, I mean, his, his basic argument is that AI has the potential to drive the performance utility of existing systems to new heights. A sort of SaaS AI hybrid Now.

Is he playing for time or simply reflecting reality? Right? So after all, you know your history, any technology has to diffuse into the economy and every wave of technology leaves its mark on the next. Now, again, simple analogy.

The appearance of the jet engine didn't instantly mean piston engines disappeared. They eventually found aircraft niches in, in like smaller aircraft, um, and propellers and jets combined to create turbo props, which are highly fuel efficient at low speeds at altitudes compared to pure jets, which makes them ideal. Power regional community aircraft and cargo planes and, and stuff like that. Meanwhile, pure Jets dominated the military fighter and medium to long haul domestic categories, right?

But the real issue is what is actually happening in the tech stacks. It, it's more what the market now thinks of those companies are worth given the, frankly, I believe, still imaginary. Impact of ar, Paul: hence the defensiveness. Right.

Jonathan: It's interesting that, um, our good friend Phil Robinson, founder of, of Board Wave and, uh, an sa uh, CEO, he, he has an interesting take on it. He, he wrote a recent, uh, post on LinkedIn. It's worth checking out entitled Cloud Birth, which we, we, we rather liked, you know, and he said, um, until recently there's been a significant speculation, uh, in why we haven't. AI bubble in terms of valuations that might burst at any moment?

Well, you know, we all know that may still be true. Um, but last, over the last few weeks as we're highlighting here, you've heard a lot about the, and this is my favorite to sca. So Phil's, you know, highlighting the saca of public cloud software valuations. Um, so you know why?

Well, investors are nervous clearly about headwinds generated by AI for traditional SaaS businesses as though they're somehow. One day where SAS is gonna exist and the next day when it's not, which is clearly nonsense. Paul: And it is a method to Phil's madness here, isn't it? Jonathan: Oh, Phil, Phil is nothing if not clever.

I think it comes down from, I think, studying computing originally at university. Um, so Phil, um, you know, put his money where his mouth is, uh, well at least intellectually and built two custom stock indices. Um, a SAS 20. An AI software 20, and what he is trying to do is test the thesis that the market is not just readjusting, it's structurally repricing traditional subscription software.

And it sort of came out as what you might. Thinks, but there's some details that are worth thinking about. So he's doing this at an early stage. It's very e experimental.

So his AI software, 20 firms had to be the net beneficiaries of AI because so far there's very few public AI companies outside of the hyperscalers. Whereas, you know, a, a much bigger basket of public SaaS companies, uh, are available. Um, but. That aside the answer to fill a Phil structural repricing question over the last 12 months is sort of shocking if, if not a bit brutal.

So if you look at his list, uh, only two of the 20 SaaS stocks are in positive territory. The medium SaaS stock in Phil's model is down over 30%. Paul: Whoa. Let's just take that, let's take that on board a sec.

The median middle. Average SAS stock down a third. Jonathan: So other people doing worse than that and other, and some doing better. Right?

But a median down a third, that's significant. So it's significant because these are well-known names, which take up a lot of the index. These is, this is sap, Salesforce, Adobe, ServiceNow, CrowdStrike. These are the companies are the backbone of enterprise software, right?

So meanwhile, Phil's AI index, and this may come as a surprise, well, this bit shouldn't, if you know anything about the industry. But the impact might, uh, the AI index is, is currently being carried by three stocks and that's symbiotic up, up about 125%. Alphabet up, 75% Palantir up. Nearly up a half is strip those out.

And this is the interesting thing. The medium AI stock is down 12%, right? Paul: Wow. Jonathan: Down.

Paul: So the winners and losers are being picked, right? Jonathan: The big question here is, is this a correction or a structural repricing, whatever. This really isn't, AI is winning, it's something else. Now we think, could it be that the market is losing faith in pursuit subscription economics?

Because that has been the basis of this for, for a dog's age and. If AI can generate a marketing campaign, draft a contract, or triage a support ticket, what's the moat for the SaaS tool that used to own that particular workflow? I mean, like, Phil, I mean, I, I, I don't think this means SaaS is dead, but I think the era, um, and I'm seeing this every day in m and a, the era of 15 to 20 x revenues. Uh, multiples, revenue multiples for, let's say 25% per annum growth in enterprise software is over and it's over forever.

Paul: Yeah. I, I, I wanna add to that, uh, the perceived pricing one is really interesting. A couple of quick anecdotes, if I may. Uh, one is, uh, the direct experience of, uh, with Salesforce actually, uh, who we, you know, we started, uh, this piece talking about, uh, related to me by the chief.

Operating officer of a very large firm, and let's just say an area related to the hospitality sector that'll do it. Now, he tried to negotiate, renegotiate with Salesforce for a temporary license suspension during COVID, and uh, he had to, this is a time when he had to lay off over 40% of his team, a lot of whom were personal friends. Guess what Salesforce said, we've got a contract and we're sticking to it. Yes.

Jonathan: We've got, we've got a contract, we've got shareholders. We need to keep our numbers up. Sorry. Paul: They just point blank.

Refused. Zero customer service. Uh, so as a consequence for the last couple of years, he's just blanked them, but he's never forgotten. Uh, and I know for, I know for a fact he's exploring, uh, options.

If he finds one of those ide uh, you know, your per pricing seat to Jonathan: get, did, did you think, do you think those, uh, agents we were talking about earlier, earlier would do that? Paul: Uh, yeah. Jonathan: I'm a, I'm a slighted agent. Paul: Better watch out, I think might be okay.

Retiring for a couple of days as well, as opposed to just saying, gimme the money. But anyway, uh, enough of the, uh, iniquitous practices of certain software companies. I also had supper. Well, actually a Soho za shouldn't, shouldn't dress that up, uh, too much Jonathan: for people who are not from the north or indeed from the uk.

A ZA is a curry Paul: spiced Indian style. Uh, Repas. Yes. Um, and, and with this is with the CCFO Over 20 year, uh, 20 plus year, uh, British tech firm, uh, valued north of a hundred million in revenues.

Um, you know, significant revenues too. Um, he was talking about. What he saw as, um, the still existent residual belly fat of many vendors and partners in the SAS space. Um, and, and he was referring to, um, you know, maybe some of the companies we've talked to earlier and about how quite clear it was, um, that his path forward for his business to be successful was.

Deeper and deeper customer engagement. Not, we're not, you know, we're not conceding on seats, not this is your price, like at a lump. His customers don't want to be on the bleeding edge of software. They don't.

They just want software that works at a reasonable price. And a low risk. And yes, that might not be the race evaluations of before his company, which is p PE-backed, is certainly going nowhere. Uh, it's got dominant, uh, it leads the category.

It has also decades of data, which any AI vendor would love to get their hands on, not for sale. Uh, and the point is, as you said just now, there will be some unexpected winners, uh, among, among all of the ai. Doomsters, some of those SaaS companies, which we talked about at the top of the, of this who grow off the back of, let's give you the next version. You need to upgrade.

Oh, you're not, you don't have the right package. Let's take you to the next package. The massive I'll do you a favor or the all you can eat deals that just sticks you with like shelfware that, that you end up paying for and, and, and don't really take control of. And the almost payola like relationship that some of those SaaS companies have with particularly with line of business buyers.

Who maybe are a bit more naive about what else is out there, those guys, sas, apocalypse, uh, sca, whatever you wanna call it, those guys are in trouble. Uh, and some of us would say they should be. Jonathan: Yeah. So this, this might, uh, teach the SaaS community form of a better word, a an overdue lesson.

Um, but I, I would, um, I wouldn't bet that, uh, the generation of AI native companies are gonna behave any better. Although they should. Paul: Yeah. So, so I think the, the, the way forward here is if they can learn those lessons and, and, and maybe not be as, um, transactional as the SaaS companies ended up.

Jonathan: Yeah. It's not gonna be about seats anymore, is it? Paul: Definitely not. It's about value.

Um, and, and that means really understanding, accepting and embracing what it is that AI can bring, uh, because that's gonna be. The new determinant of value. Jonathan: Thank you for listening. If you wanna learn more about category design, head to be categorical.

com. If you need help designing and dominating your category, then get in touch. Contact details are in the show notes.