Anthropic 的超级碗广告：谁赢谁输？| Sierra ARR 达到1.5亿美元：客户支持赛道是否过于拥挤？

Mike Cannon-Brookes

The idea that software as a category is dead is ludicrous to me. It's tough out there. It's going to be hard. Welcome to the technology industry.

Rory O'Driscoll

I just think we have to give up on TAM. We have to let the revenue show us the path to TAM. Every category that I know of outside of engineering and product is at existential risk of shrinking seats.

It's the venture capital equivalent of “You can't get fired for buying IBM,” right? You can't get fired for sticking money in with a 1× preference on the consensus winner. This is classic top-of-the-bubble stuff.

Ready to go. So excited for this, guys. We have a very special guest joining the trio, Mike Cannon-Brookes. Mike, thank you so much for joining such a phenomenal group of intellectuals.

Mike Cannon-Brookes

Thanks for having me.

Now, I want to start with Anthropic's release of its updates, which wiped billions off the stock market. Anthropic predicts $149 billion in ARR in 2029 under the most optimistic scenario. The question that we ask ourselves is: Is this just the collective budgets of enterprises all going to Anthropic? How do we think about that?

Mike Cannon-Brookes

Just to mention it quickly, $150 billion for Anthropic. Let's say, if they make that number, OpenAI's estimate for the same year—I checked—is $180 billion. Call it $350–380 billion between them. The total worldwide software business is $700 billion, so it would be a pretty big chunk.

Unless you expand the TAM—the same old discussion—if you don't believe in TAM expansion and it's a zero-sum game, then if 2 companies are going to do $400 billion, and we can probably assume Microsoft, who's already at $200 billion, is going to roll over and die, then, yeah, they're going to have to take the money away from the rest of us. But clearly, the bet is TAM expansion; otherwise, it gets really hard. That would be the first pass.

Hey, Mike, can I ask you a meta question? I'm sure you've thought about this: Does all this Anthropic revenue, at some point—companies only have so much budget, especially CIOs and down—compete with that budget for product engineering? I know it's a boost for you, but on the other hand, there's only so much budget to get extra seats and extra licenses. Do you feel any meta-competition from that, or none at all?

Mike Cannon-Brookes

Let me take that second. I think you also have to work out the revenue stack here, because we use Anthropic a lot. It's one of our biggest models. We use multiple models, which I think is what most good SaaS vendors are doing with their customers' choices, right? We use a lot of Gemini, a lot of Anthropic. We have a whole bunch of Llama and Mistral running internally, and a bunch of OpenAI.

Our job is to pick the best model for the best cost, with the best quality outcome and the best speed. Then we have a whole gateway that's constantly balancing that.

But when we spend money on Anthropic, don't forget we pay AWS, and then AWS pays Anthropic. So if someone spends $1 million on the Atlassian platform and we spend $200,000 on AI bills or something—pick a number, nowhere near that high proportionally—that goes to AWS revenue. And then AWS spends, what, $150,000 on Anthropic revenue. Don't forget, the individual revenue number is not necessarily counting the whole stack. They have to buy a bunch of chips and stuff. The revenue stacking here gets a bit complicated, because there is a whole set of layers underneath this.

There's no doubt that's a big number, though. Do we compete with the model providers? I think this is one of the big questions over the long term. I don't know that we've yet worked out what all the model providers are going to do, right? We use a lot of Gemini and deeply partner with Google, but obviously, you could argue Google Docs and Confluence are somewhat competitive—the 2 ways you can write a piece of text, collaborate on it, and change it.

Megacorps in tech compete on one level and partner on other levels. There's nothing new in that. I think we have to be good. The question underneath that is: Can they all afford to shut down their models? Can they get so good that they'll only use themselves? I don't believe they will, given the competition from the other 3 or 4, before we even get to China, on models or anything else.

I'm not particularly worried about that. We have to compete with everybody that's providing our customers with the same value that we're providing them, right? At some point, $150 billion for Anthropic is a lot of CIO budget. It does make you wonder where that money is really coming from.

But I think, to Rory's point, it depends on your question of whether it's zero-sum, right? I don't think any of this is going to be zero-sum. People say, “Oh, well, that's going to swap software.” And I'm like, “Well, is it really swapping for software?” If your business is getting more productive, you're building more products and services, your quality is going up, your speed is going up—whatever—you're going to spend money on different things.

The technology budget of most companies over the last 30 years has gone up massively, I would argue, since 30 years ago, when they were buying keyboards and mice and stuff like that. I don't think that's going to stop.

I think the interesting thing about that—there's a lot in that answer to unpack—but I just want to highlight the “you have to be good” sentence, because when we're going to talk about all the “what happens to SaaS?” and all that, I think Mike's big-picture comment is something we should remember every time. It turns out, if you want to make a lot of money, you have to be good, right? And you have to compete with everyone else for the same dollars.

My guess is that will be a recurring theme. It's not impossible to succeed. You just, quote-unquote, have to be good.

There's a lot of complexity here because you're right: You have the margin stacking. A classic example is, you look at Anthropic, it's doing $1 billion in revenue. You look at Cursor, it's doing $1 billion in revenue, but it's pretty much the same $1 billion. So you have to think about the margin stacking.

Then, on the other hand, you have to say to yourself: If Anthropic hypothetically was doing $100 billion and all of it was through third-party ISVs that were also adding value and getting more money on that, it would point to the end market having to expand a lot. If everyone's going to eat and grow, you're going to have to buy into big TAM expansion.

And you're right, Mike, software's grown nicely over the years. My guess is, for this kind of growth, you probably need more than the trend line of the last 20 years. You probably need some acceleration on top, just because the numbers are so big.

I mean, to go from $4.5 billion to $150 billion—and I'm not for a second saying I think they will; it's just their number—in a world where Microsoft does $200 billion, you're basically saying, to a rounding error, Anthropic is another Microsoft. OpenAI is another Microsoft. And that's 2 more Microsofts, which are pretty voracious mouths to feed out of the same pile of cash. There better be more money coming, or it's going to get pretty tight around here.

An obvious one is when you look at the IT spend: You get $600–700 billion of software. But the stunning thing is, you have $1 trillion, plus or minus, of consulting-services-type stuff. If some of the software can eat that, that's a pretty big slug of extra budget available for a more efficient product. So I kind of agree: Finding the extra money is the thing.

Do you not think that consulting-services pie actually goes up? I think one of the most profitable categories right now is your Accentures on the implementation and integration of some of these systems in the largest enterprises in the world. I tweeted this. I said I think they will be more revenue-generative than the foundation-model companies in the next 2–3 years, as they often are in the early stages of cycles. But actually, I think that would grow, not shrink.

Rory O'Driscoll

That kind of stuff, maybe. Yes. Consulting to help you implement AI, yes. Consulting to help you morph your codebase from Fortran to blah—the market we've been looking at a lot is all the consulting spend around systems integration: SAP, Oracle, NetSuite.

A lot of that is pretty rote, and it makes coding look hard. There's a lot of automation that we think you'll see there. So again, it's the same thing: swings and roundabouts—places where it grows and places where you're going to get scrunched.

Listen, we all need these consultants, whether they're FDEs or Accenture. We all need them. I think we need them more in the age of AI. The only thing that I find lost in this simplistic version of the argument is that they have to be smarter than they used to be. They really do.

The average agency—I don't know. Maybe I've lost a little bit of track of this at Atlassian, but traditionally, in the more enterprise, Mike, there were a lot of partner agencies that were helping do big deployments of Jira and Confluence. I'm sure they had to be smart, but not quite “train the agent from scratch” smart.

Maybe Atlassian's a bad example, but I can tell you only a handful of folks that we work with are really smart.

This is why customer success is dead: all these B2B companies are trying to convert their CS teams to FDEs, and I don't think even 10% of them can do it. They can't really be product engineers; they don't need to code. So I wonder if we have enough consultants out there to really do this.

Mike Cannon-Brookes

I like the argument that we're heading into the death of consulting because we're going to run out of consultants.

The ones that will do it—the ones that are smart enough—

Mike Cannon-Brookes

We'll have to make a new factory.

Mike's taking no prisoners here. Okay, it's open season. But the thing that's going to stop that is, look, their economic incentive with the software provider is to make this adoptable by ordinary people. If the only people who can roll out your shit are wizards, then you're going to run out of wizards, and then you're going to run out of revenue. So, for mass adoption, you're going to need mass simplification.

We will build those guardrails, right? For most of our customers deploying agents, the agents that they deploy themselves when they start are relatively simplistic. They are small automations and pieces of software that make some business process go a little quicker. And you could argue, well, it's a very minor improvement. I say, no, but it's a major leap in their learning, right?

They're learning and they're controlling and they're letting them out slowly, and they're getting an understanding of what they can and can't do. As the technology improves and as they find more business processes that they can automate with software, we can call it an agent, we can call it whatever we want, but that's a pretty normal, I would say, corporate adoption pattern. And we're super early in that cycle.

These companies will tell you some sort of fantastic AI story, but they're not having millions of agents run right over their stack. So how do you genuinely respond to that? Because it all makes sense to me, and I've said this to guys before: I think one of my weaknesses as a VC sometimes is I can be too incrementalist. That makes sense to me—that's how I'd adopt it, right?

How do you square that with the Wall Street VC buzz that the entire SaaS industry was meant to implode in the last 2 weeks and new company agents are going to take over everywhere? How do you—I don't want to put you on the spot on that last theme in particular, but how do you—

Rory O'Driscoll

On the spot.

Okay, come on. Rory, put him on the spot. Rory, come on.

Rory O'Driscoll

24 years of being on the spot.

Mike Cannon-Brookes

Okay.

I just didn't want to be rude to our guest, but I'm game. Yeah. Genuinely, how do you respond to people? How do you respond to the madness of the last 3 weeks where people were saying the entire software industry is going away and being replaced by AI? Because my LPs are asking me, how do you think about SaaS today? What are your thoughts on that?

Mike Cannon-Brookes

Look, I've said a few times: I don't think software is dead. I think software is alive and kicking. I think it's confusing to a lot of people in the software industry, and maybe that's normal.

To take the most generous view, will every single software and SaaS company make it through the next 5 to 10 years? Absolutely not. Will a lot of them continue to grow and prosper through that 5 to 10 years? Absolutely. Is that any different from the last 10 years? No.

I keep saying internally, the competitors we had in 2000, the competitors we had in 2005, 2010, and 2015—I went and pulled all these old documents—a whole bunch of those companies don't exist. They've been merged, acquired, or gone out of business. We have a set of competitors today that are new and upcoming, and we have to compete with those. That is the history of the technology industry, and that's going to keep going.

The idea that software as a category is dead is ludicrous to me. I'm like, wait a second: it's very efficient for businesses to buy prepackaged technology solutions. They don't write everything with assembly, and they probably still won't.

But can I throw in one micro thought? I view Atlassian—and I'm just a student; I didn't found it—as above the fold. And what I mean is, you're not seeing seat reduction. Product and engineering is a beneficiary of AI spend, right? Whether it's vibe coding or however we move, we are creating radically more software than 6 months ago. Like any curve, if you look at it, for me, it's exploding, right?

At least for a while, I think that benefits Atlassian because I need more tools, right? I need to track more issues, more tickets, more everything. I think there are plenty of other spaces that are being decimated. I mean, if your traditional support software is something like Zendesk, the old Zendesk is growing 0%. We're cutting humans.

To go to Rory's point, I think there are some folks that are above the fold, and there's a few, like Anthropic, that are so far above the fold we need to figure out how to scroll above the top of the monitor. But I think more are below the fold. More are seeing hits.

I actually think product and engineering is an island of stability today. No one is—the idea that you see teams with fewer engineers, maybe. But when I walk into leaders, you know, I was at Replit the other day at $300 million in revenue, 300 people. There are like 11 people in go-to-market. It's probably like the old days at Atlassian, right? But a lot of engineers. They're not shrinking the size of their engineering team, but everything else looks pretty lean.

Mike Cannon-Brookes

Look, product and engineering and software teams are a significant part of our business, right? 40%—it's a big part of our business. Service in general is a big part of our business. So your argument about Zendesk—the Service Collection is a very large business. It's our largest at-scale business, growing very fast, faster than the overall company.

IT service management and HR service management are our fastest functional areas, right? Going inside there: financial service management, just general employee service management. We just shipped a customer service offering in October. So we're wading into those waters.

This is crazy, right? This doesn't make any sense. No, the business is growing really, really well, right? We have to provide that software layer that makes that cheaper. There's no doubt about that, right? That's our job to do.

I think the generalized view that customers are going away in these times—certainly for some businesses, yes, you can say that. The industry-wide phenomenon, which I think is what Rory was asking, doesn't make any logical sense to me. Again, we're either an example of the whole industry, or we are somehow unique, and people argue one or the other depending on the point they're trying to make. So it's very difficult.

The one point I would make is, again, we're accelerating our cloud revenue, which is a 26% quarter growth rate, at well north of $5.6 billion at the moment, and we grew our RPO by 44%. That's also accelerated for the 3rd quarter in a row.

What is RPO? That is customers who are buying $3 million worth of software for 3 years, something like this. $2 million of that goes into our backlog of the future, right? We book a billion dollars in revenue. Ultra-simplistic CFOs and accountants—I understand how it works, but that's a close enough proxy. Those are customers making 3-year bets.

And so now we have people arguing, ah, but in year 4 it's all going to fall apart. And I'm like, really? Like—

I think—

Mike Cannon-Brookes

These customers aren't stupid.

Comments. One is because of the AI—OpenAI and Anthropic—we now all are experts on RPO. So everyone knows RPO now. Turns out once someone signs $100 billion in RPO, everyone has to figure it out.

But going back to your comment, first, I want to say the software word is overused because you're right. People say it's the death of software. But on the other hand, we're going to talk to Harvey in a while. Harvey just raised a wonderful sum of money at a high price, and that's clearly software. So “software is dead” is a stupid statement, right? We should discard it, right?

On the other hand, your comment—yes, there's always an element of change. If you look at your competitors every 5 years, some folks go to the wall. The argument here would be somewhere in the middle. It would say when technology doesn't do a step-function change, you probably have an organic death rate of 5% to 10% a year, because that's the way capitalism works, and the winners like you guys come to the top.

But every 10 or 15 years there's an architectural shift in software, right? Where it just gets so new that instead of being a 5% death rate, you go through a 2-year period where there's a 50% death rate. I'm just picking numbers out of my ass, right? And that at least is a credible discussion.

So, to make it tactical to you, do you think the software you guys write with LLMs is quantum different from what you did before, or do you see it as a step-function change in the kind of software you make, or just an organic progression? Does that make sense?

Mike Cannon-Brookes

It's both. It's both. So, the way that we are building software—and I think the irony is often forgotten here—is that apparently we're some sort of cavemen sitting around just banging on clicky-clacky keyboards and writing assembly code, and we haven't figured out that LLMs are pretty good at writing code and can speed us up.

We've got 10,000 people in R&D. This is amazing, right? I'm like, sweet, we can build a lot more stuff, a lot better. Do we have some pockets of the company where people just sit and Claude Code all day, every day, and write cool stuff? Yes.

But I don't know if there are many software companies still sitting around in their R&D departments working the old way, right? Some won't make it fast enough. Totally agree with you. You have to build new things. Some of those will work, and some of those probably won't work.

I don't think it gets as simplified in people's minds as, “You're just bolting on AI features.” No, I'm building—I have a very, very large team at this stage building an entire substrate of context, search, and chat, not to sell, but because I have to. I also have a very large infrastructure team at this point managing AI costs.

You guys all went on about inference costs: “It's going to crush us,” et cetera. I'm like, actually, it's going the other way, right? Our inference costs are going down, and we're very good at adopting new models. We're also good at knowing which features need the old models that get very cheap. Some of our features are 1,000 times cheaper to run than when we introduced them, and the feature still works really well as an AI feature.

So guess what? We're good at optimizing infrastructure, right? Our COGS has gone up—or rather, COGS has gone down, right? Our gross margin has gone up in the last 6 or 7 quarters, and we've deployed a lot of AI, right?

Jason Lemkin

My question is then, are we just throwing the babies out with the bathwater? We've got your Monday.coms, your Duolingos, your Workdays, your HubSpots. This is real. I'm just trying to understand: should I just call it quits, sell my Monday.com and my Duolingo, and accept defeat? Are they not coming back, or are they coming back?

Mike, we have a brilliant saying on this show. Rory always says it wonderfully. Harry says, “That's great, but what about me?” I'm like, help me out here. I'm looking at all my positions and managing this portfolio. It's a living.

That's right. Well, apparently AI can just listen to this podcast repeatedly, and it'll do the trading for you. You'll be fine.

Look, there's no doubt we have a secular rotation out of software, right? I've been on many, many therapy calls with many software CEOs and other people. It's tricky to understand how that changes and rerates and everything else. In my view, that's definitively fear-driven at the moment, and fear overrides it.

Generally, fear comes from somewhere real as well, though, right? Will every single one of those names be bigger in 5 years than today? It's probably pretty hard to say. If you list a basket of 20 companies, they will all do that. Will the overall average be higher? I would argue yes. The fear-driven average software price at the moment is there.

You also get all sorts of financial effects, like people buying chips. The software-versus-semiconductor chart has been around a lot, right? It's this thing. And guess where those chips are going? They're going into data centers to make AI inference, to do all sorts of wonderful APIs, and eventually that's delivered through applications at some layer. We're working out where the profit stack and the revenue stack get redistributed, as we talked about beforehand.

Jason Lemkin

Maybe I've got my Atlassian numbers wrong. Forgive me, but Atlassian had a pretty good quarter. I know Mike probably doesn't want to talk about this, but somehow we got onto it. Atlassian's growing 23%, and RPO is up 44%. That's pretty good.

More importantly—and I don't want to put Mike on the spot—it's accelerating. Atlassian is doing better. The problem with so many peer companies—and these are the best founder companies—is that almost all the public SaaS and B2B companies are founder-led, right? Very few are reaccelerating; they're still decelerating.

The LLMs are open to everybody. You have a year. Claude 3.5 and 3.7 were really good. You've had 7 months since then. Do you get 12 months? Do you get 18 months? How long do we give you to recapture some growth and some revenue from the AI revolution?

Atlassian has this layer, but it also has growth. If you've spent the last year with 50,000 engineers building all this great stuff and now your growth is in the teens, maybe heading down to 10%—I mean, you have better comps than me—but that's the one where, Rory, I think you should tell your LPs you're worried.

More of the public folks have seen no reacceleration; more of them have seen persistent deceleration. We're going to see more this quarter, and I wish it wasn't the case. My life would be better if it wasn't. My life would be better if SaaS was on autopilot like it was through 2022, but you have to show me the money in 2026 at some point.

All that infrastructure spend that goes to inference has to go to software. It has to go down, and you've got to show a connection to that spend, or you may not have a reason.

Presumably, if you look at history, some of that goes to new companies that arrive, right?

You nailed it. I was actually going to say, you're exactly right. The problem with looking at the overall SaaS growth rate now is we haven't added any good ones to the top of the funnel, right? So gradually, the survivor growth rates are going down because we looked at this once, and it was 30% for about a decade and a half.

When you dug into it, that was because anyone who goes to your point, Jason, when you go below 10%, you get the red card. You're sent off. You're told, you know, you go off DCF and P/E land. But they were dropping another one in at 60% growth on the top, right? So it stayed at roughly a 30% median growth rate for a decade and a half, and now there's been nothing good coming in. You're just seeing the bad side.

Jason Lemkin

And in fact, you've had a lot more. I would argue there's a whole series of phenomena. Maybe it's a great post to break apart. Software IPOs for the last 5 years have not been a category of new companies being introduced, right?

To be fair, if you go talk to a Figma employee, my friend Dylan's having a fun time over there. They're a good example of a company that's going to do great in the current environment, right? But you're probably not setting an example where a lot of other private-company CEOs are motivated. You had Cliff Weitzman on a little while ago, another good friend of mine who actually lives about 2 blocks down that way. He's a little less motivated having seen what's going on, right?

Rory O'Driscoll

I wouldn't be rushing today to IPO there.

Jason Lemkin

Well, the number 1 factor is you don't have those 60%, 70%, 80% growers that take the average up, right? You do have this one. Number 2, you've had private equity—and, again, a totally valid part of the capital markets—come in and hoover up a ton of companies over the last 5 years, right?

I'll give you an example. In 2020, we formed a group of SaaS CEOs. I think there were 16 of us who got together. We'd meet on Zoom every single week, right? At one stage, 10 or 11 of us were public, maybe more. We all had earnings calls, and it was a really good therapy group for a while.

I think it's me, Eric from Zoom, Aaron from Box, and Toby from Shopify. That's it. That's all that's left, right? Almost all of them have been bought by private equity or big tech. Big tech is getting a lot better at buying stuff, historically.

All of these phenomena lead to this weird average system. You have companies staying private a lot longer, right? They've got a lot more capital. Poor Cliff Weitzman would have gone public a lot earlier in the old era. They don't need to nowadays as much, right? The Stripe guys are the same thing.

When you look at a public company, you're almost seeing the survivor bias of not being small enough to get bought by private equity, not being large enough to get bought by big tech, and no new companies coming in. It's hard to work out what that average should be, actually, right?

That's not saying it's not happening. I think this is where all these forces are happening simultaneously. We'll see how much this causes it, because there's one more effect that we're not talking about, which is that some of those SaaS companies—and, again, people have argued it's us, and we're like, wait a second—that 44% RPO growth is not coming from people paying $20 on Google Ads.

Some of those have also been affected by changes in marketing channels and other things that AI has brought about. If you're a multidecade company, you get used to this, right? You've moved your marketing to TikTok or some other way of the business model, and you get these chasms in marketing that have opened up. Some of those have also become a lot more challenged.

We mentioned new companies bringing enterprise value and expanding budgets. Harvey raised $200 million in December at an $11 billion valuation. It must be quite nice, though, to fundraise every month. It's just a nice hit of ego, isn't it? Another $200 million from venture investors.

Rory O'Driscoll

A nice bunch of secondary.

Nothing wrong with that. A lot of this is secondary. It drives a lot of it.

Okay, let's talk about this deal. How do we think about this deal? They're going to end the year supposedly at close to $600 million. The growth is pretty phenomenal from $200 million today. How do we think about this?

First of all, let's pick the obvious point. It was fun that it happened this week because, if you think of what also happened this week, Claude Code produced 13 plugins, and one of the plugins was the legal plugin. That's one of the things that's alleged to have caused the SaaS apocalypse.

Oh my God, the old-school vendors, Thomson Reuters and some of these guys, declined, right? At the same time, a company that, if you squint one way, is just, quote unquote, a GPT wrapper, raised $200 million at $11 billion. Literally, the kind of whiplash in the thinking here is hilarious: “I think all software is dead except this software,” which is pretty close to the plugin from Claude, but at the same time it’s worth $11 billion. So it just works? There’s a fair amount of inconsistency in the worldview, right?

I don’t think so. That’s just worth pausing on. Other than that, look, we’ve discussed it. It’s actually very similar to the discussion we had on OpenEvidence. This is the killer; this is the lead company in the AI-for-law-firms space. Obviously, Legora is number 2. They’re both killing it. You’re seeing mass adoption.

The million-dollar question—and you cited what you think they’ll do at year-end—is that all you know for sure is what they’re doing today, at $190 million a year. The question here is total time, right? In other words, can you compound? If you can compound for 3 or more years at this kind of rate, you’ll do amazing. If you top out based on the number of Am Law lawyers in the U.S., which isn’t as large as you’d think, you could find yourself time-constrained, and that’s the bet.

They’re clearly going to get an arbitrary number—60% of the market share, with Legora getting 40%. I don’t know, 50-something like that. So it’s really a question of how big the TAM is and when they hit it.

Mike Speiser

If they’re going from $200 million to $600 million in 1 year, that’s tripling. If they’re magically able to double the next year to $1.2 billion, we’re paying 10 times 2-year-forward revenue, right? It’s an asset that, unlike Atlassian, I can’t go buy. So there’s scarcity on top of it, right?

Jason Lemkin

I just think we have to give up on TAM. I think we just have to let the revenue show us the path to TAM. But if the numbers are there, and Harvey does $200 million to $600 million, it showed us the money on TAM.

Right?

Mike Cannon-Brookes

So why do you think we should give up on TAM? Just because it goes from $200 million to $600 million doesn’t mean we should give up on TAM.

Because it’s a reason to say no. You’ve got to—because if you’re generating— it’s a reason to say no to Databricks growing 56% at $5.5 billion, or 65% at $5.5 billion, right? Sometimes you’ve got to lean into it if you want to make money, at least.

Mike Cannon-Brookes

I don’t know. TAM is a concept I’ve always struggled with, because our whole job in technology is to change TAM, actually. If you think about it, the TAM of technology as a whole has changed for 50 years.

I think Harvey is a fantastic business, but let’s not forget: if anyone’s been involved in legal proceedings or court cases, or even just dealt with a lot of lawyers, they’re not an efficient part of an industry, right? And they deal with a lot of text. So you’d argue that there’s a ripeness to do that.

Credit to this company. Maybe Harvey is a GPT wrapper, but you know what? They’re a GPT wrapper that’s got a product their customers like. They’ve got a go-to-market model that’s clearly working in a market that probably doesn’t have a lot of competition to start with. Good job building a piece of technology that customers want, and hence a valuable business, whether it’s worth $11 billion or whatever.

It’s growing 300% a year at $190 million ARR. Fantastic business.

Totally agree. I want to come back to 2 things. One is because I realize, as I replay it, I was using the word “GPT wrapper” as a term. I didn’t—that’s not a phrase I would use. It was a phrase that was applied to the sector, and then people simultaneously were doing this, and I’m pointing out the divergence. You can’t—it can’t be both at the same time.

I think that if you’re doing a deal like this—and I think it’s an amazing deal, and we have a deal in a related space—I don’t buy the GPT-wrapper appellation, is my point. I don’t think it’s a useful phrase. My point was it was wonderful and funny that it happened the same week that Claude was deemed to be able to annex the entire legal market just by issuing a plugin. Those were 2 different, contradictory ways of looking at the world.

Jason Lemkin

Oh, entirely agree. I was trying to poke fun at the GPT wrapper, right? If Atlassian is just an orchestrator of humans, all software is just a process wrapper. We take some business process and deliver a better UI to do that, or a way to repeatably do the process, and we make money doing that, right?

Does Harvey have a sustainable competitive advantage that will just be eaten up? That is entirely, I assume, what their leadership team is sitting around thinking: We’ve got customers and distribution, and that is in itself an advantage, right? Those customers are presumably pretty sticky right now.

Do they keep adding value to their customers with more GPT wrapping, or do they add more and more bits of the legal process into the UI, into the technology? I suspect that’s exactly what they’re doing, and that’s what they’re getting paid for, right?

Any law firm can pick up Anthropic and go build their own Harvey. Are they? No.

But I think the important question to ask is: risk-adjusted and opportunity-cost-adjusted, is this a good place to put my dollars? And when you go back to Jason’s argument, I’m totally with you. I’m not shitting on them at all, Mike. Fuck me, I’ve created a podcast. These guys have created $600 million by the end of this year. So who the fuck am I to throw shade?

Ultimately, I can put my money in Atlassian today, which, respectfully, as you know, is down to very low levels, unfairly priced, brilliant business. I can put my money there today, or I can put my money in Harvey, which needs to triple and then double, and then I pay 10 times that.

Mike Cannon-Brookes

And I totally agree. I’m not arguing whether you should put a dollar in A or B. There’ll be mispricings up and down the stack, right? I’ll give you an example from history, which I’m sure the Accel guys wouldn’t mind.

When the Accel guys invested in Atlassian, it was all secondary. They put $60 million in 2010, and they bid the highest of the 5 VCs. Can I just say, they’re also the nicest and had the 1-page term sheet and all sorts of other things.

But when they put that in, we now have their model. We’ve known them 15 years. Rich just stepped off the board in December after 15 years of amazing service, well beyond Accel’s investment. They’re legendary people as far as I’m concerned.

Their expectation was that Jira and Confluence collectively would grow, I think, 30% the year after they invested, then 20%, then sort of peter out at 15%. So by 2013, instead we did 50%, 60%, 70% or something, and they were like, “What?” And we’re still growing Jira at faster than their 3-year terminal rate 15 years later.

Now, are they stupid and did they get the TAM wrong? No. The business just was a lot better than even they understood. They were trying to get 2 to 3 times their money. They made 100 times their money. But that’s a success. You’re looking back at a success example, right? They probably did a bunch of those deals that didn’t work around us, right?

It’s very hard to predict what the management team at Harvey are going to do in building out their business over the next few years. So I assume the smart people putting this money in believe that the management team can build a business that gets to, at current SaaS rates, $11 billion. They’d have to be, what, 4 times forward revenue—to, like, $4 billion in revenue or something, $3 billion in revenue.

Jason Lemkin

Because, just to take your 2 points, I think your comment is correct as the entrepreneur, but I also think Harry’s comment was correct as the investor. Because, look, when you’re talking about companies that are raising at $11 billion, there’s no useful sentence that says, “This is a bad company.” They’re freaking amazing companies.

But as Harry points out, we’re not here to say, “You’re amazing because you built a big company.” The Harvey team are amazing because they built a big company. I didn’t build a big company. I’m not amazing on that dimension. But in terms of capital allocation, you have to say to yourself and compare the 2 scenarios.

You were a profitable company. Accel put in $60 million. My guess is the revenue multiple was well sub-10, right? Growth rate—thinking back, I can tell you what it wasn’t. What have we got? Harvey’s at $190 million ARR now, $11 billion. That’s 50 times. I can tell you it wasn’t 50 times, Mike.

Mike Cannon-Brookes

No, no. I was going to say it was probably 10 to 12 times forward revenue, but we were an unusual company. We were, like, $50 million in revenue with $20 million in profit or something.

Jason Lemkin

And that’s a deal that outperformed on the growth rate. As a result, as you say, a 3x became a 50 to 100x, right? When you look at another deal where you’re paying 50 times run-rate revenue, it’s growing much faster than you were growing at the time. How do you, to Harry’s point, think about the risk-return on that?

It boils down to this: you were able to beat their model by just growing 30% instead of 20%, right? What do you have to do to beat the model when you’re paying 50 times? The answer is you have to peg that growth rate at well north of 300% or 400% a year for 3 or 4 years.

Mike Cannon-Brookes

Yeah, I can’t agree with the math.

Jason Lemkin

If you do, then it works.

That's the point. Go. Sorry, just on the TAM thing. What I mean is that when we started this podcast, I was actually fairly skeptical of AI replacing humans. Not because I didn't think it would happen when I started vibe coding, but because I thought too many VCs talked out of their rear. They'd say, “AI is going to replace humans. AI is going to destroy all jobs,” and I thought, “This is a bunch of people who haven't done any work. Are they just reading Twitter all day?”

My mind has changed over time. What I mean is, when you look at it, it's not going to be true in all categories. But when you say, “Listen, there's $200 billion a year spent in the US on non-partner lawyers, associates, and paralegals”—$200 billion a year—you could have said 6 months ago that it'd be ridiculous to think Harvey's going to make a dent in that. Maybe it's not ridiculous now.

If that growth stays—if it can literally grow at the pace it is—what you're seeing is happening. It has to come, to Rory's prior points, from somewhere; it can't come out of the legacy software budgets. It can't come out of the old TAM for Bitbucket or whatever, or legal software of the old days. My point is, it's not all going to prove itself. Some of these will flame out, but that amount of growth is spectacular. It's spectacular.

At some point, like Anthropic, you prove it: The growth is so stellar, you have proven that you have captured a massive amount of TAM that was uncapturable before. So, as an investor, I would just put my money in at some point, right?

Rory O'Driscoll

I agree. Let me make the argument against—not because I think it's correct, but because it is the other argument. I sat there as an investor thinking about it: What if you look at it this way? They have $190 million in revenue, right, in ARR, and 100,000 active users. That's $2K per active user, right? That's under what Westlaw gets from a lawyer. That's not replacing the labor revenue; that's nice next-generation tool revenue. Do you understand me?

If you're only getting $2K per lawyer, and there are 1.3 million lawyers in America, some of them are in-house and some of them do weird things like patents and can't buy Harvey's product. There are 400,000 or 500,000 lawyers in Am Law—big law people—who would be buyers for this. It's a $1 billion, $1.2 billion marketplace. If you're only getting 2 grand per head, the math won't work.

Jen Abel

Now, Mike's right. In the end, my point is this, and this is why I get back to Harry saying: To make this a big-ass company, you have to take that $2K and turn it into $5K and $10K per lawyer by eating the work, right? Then you TAM-expand, and it's doable. Mike's right: The whole last 50 years has been software doing more.

Jason Lemkin

I don't think it's going to be true, Jen. I think it's because this is what I'm already seeing. It's not that; it's $50K to $100K because you get rid of the humans. Listen, we're in a weird niche, but I can tell you, because I know the GTM agent category really well, that they're all $50K to $100K because you get rid of the people.

If you compare Pipedrive or a Trello clone for CRM in the old days, at $8 or $12, to an agentic product that gets rid of humans, I'm not saying for sure this is what happened to Harvey, but $50K to $100K—if you can get rid of half of your associates or two-thirds of your associates, it's dirt cheap, and you'll just pay it because they won't quit. They don't want raises. They don't complain about weekends. They don't have all the issues. Our agent closed a $100K deal the other day on a Saturday night. It doesn't complain.

These agents want family lives. These associates in law firms, you just get rid of them. $100K is nothing.

Jen Abel

There are two arguments against what you're saying. One, Mike's already made, which is this: If we can make more software, we're not going to get rid of those humans and the coordination problem. We're just going to make more software, have the same number of humans, have the same coordination problem, and keep buying Atlassian stuff. So, the same argument applies in law here. Maybe you just automate more.

Mike Cannon-Brookes

No, it's totally different. We are building more software than ever, and we need the humans, and no one is reducing their engineering team. The world is so competitive today. You can't wait 4 years for a release like the old days. You're pushing code. So that category—every other category may shrink.

Okay, that's coherent, at least. What you're saying is, if you're selling to engineers—

Jason Lemkin

No, seriously, because we've talked enough that we know what we're going to say. You're right. What you're saying is every engineering category in software won't be cut; it will use the tools to just make more software. But every non-engineering area, like law and SDRs, could be automated away, so the headcount goes down there.

I think every category that I know of outside of engineering and product is at existential risk of shrinking seats. Workday said it. Everyone else is saying it. We're seeing it. I mean, Workday's founder had to go back. They said, “Even at Workday, we're seeing headwinds on seats,” because people just aren't hiring across the Fortune 500 like they used to.

But we are in a renaissance of software. We are building so much software. It is unbelievable.

I think it's fascinating that on this podcast, the product and engineering firms are going to win. So we're a winner. We're above the fold, which I love as well. This is, in itself, an old-school term, Jason. The fold doesn't exist anymore.

Jason Lemkin

I know. But I think—listen—but I think Zoom, I mean HubSpot and Monday, even Zoom, are below the fold.

Mike still has the floor.

Mike Cannon-Brookes

Well, a few things. Firstly, your holiday party with your lack of belief in humans is going to be amazing, right? You're sitting around drinking wine with all your agents. That's going to be a super-fun holiday party, right? I'm really looking forward to it. I want to see the pictures.

Jason Lemkin

They'll get drunk and make an ass of themselves if they're all agents. At least we won't have managers there.

Mike Cannon-Brookes

I like humans. Humans are fun to hang out with, right? We're going to need a lot of these people. I think there's a lot in Harvey that's fascinating.

The question is two things. One, legal firms generally bill time, so it's hard for them to increase their profit margin by billing more for less time. I don't know how that affects this particular industry. I'm not an expert at all on legal, but I'm just saying there's something interesting about their pricing, right? I don't know how legal services are going to be value-based priced or something. They're literally the ultimate selling of hours—photocopying and all that sort of stuff. So we'll see how that works for the billing model, and maybe Harvey helps them with that or doesn't help them with that.

Secondly, I like to think about things that are demand- or supply-constrained, right? Is a particular function input- or output-constrained? That makes a big difference in how I think AI will be applied to it. It's one of the interesting schemas to put to it.

You've talked about customer service or legal. I would argue those are input-constrained domains. I have a whole lot of fantastic lawyers at Atlassian in our in-house counsel group. They're input-constrained because if I make them more efficient, or if Harvey makes them more efficient, they can't create more legal problems. There is a certain set of problems.

It's just like customer service: Your customers ask 100 questions a day. You need a certain amount of service to help them out. If you have twice as many customers, you'll probably have 200 questions a day. It's based on some other ratio.

The difference with something like software or engineering is that it's creation. It's how much you can make. The roadmap is never finished. You can keep creating more. So some domains in a company have a constraint where the input is fixed and you're doing a certain amount.

Sales and marketing are kind of in this middle ground, which is fascinating, because ostensibly sales is input-constrained by a TAM, but you can build new products and new ways of going to market. So I don't know. We sometimes jump across these categories, and it's not super helpful because they're very different.

I think, genuinely, that's super helpful, because I remember when we were looking at the systems-integration automation category. One of the things that we realized—and you got to in a second, and it took us a few days—is that if we automate the process of installing SAP, installing Salesforce, or installing NetSuite, it's not like the company's going to buy 2 helpings of NetSuite because it got cheaper. They're just going to spend less money doing it.

Those are inherently less magical markets than the stuff where the intelligence allows whole new things to come into existence. I think that input-output thing is actually a very useful way to think about where the magic upside might be.

Mike Cannon-Brookes

But AI may also let us have more inputs, right? For example, we're one of the few organizations of our size that uses Agentforce. We ran an experiment. Marc said, “Use Agentforce to go after the leads. It's not worth anybody's time to do,” right?

So we did that. We unleashed Agentforce on all of our leads, inviting everyone to come to an event, sponsor us, or do anything that wasn't worth a human's time to follow up on, and it worked.

And that’s like radically more inputs. That’s like 5 to 6 times more inputs. There are a lot of legal things that aren’t worth taking to your lawyer. I can’t wait 4 days for a sales contract to be reviewed; I can’t do this. There may be Jevons’ paradox or whatever in every category—it may explode.

We’ve all had terrible customer support so often, but when you have a magical customer-support experience, with AI or a human, you want more of it. You’ll consume more. If I could get help all day long with a product, I would be in that agent all day long, helping me—not just vibe-code, but vibe-use my product. So we made Jevons’ paradox apply to every category. Go, Jen.

Customer support is one I just cannot get my head around. I know that sounds ridiculous and very basic, but when you look at the sheer number of players that have been funded to huge extremes—and we saw Sierra’s $50 million-plus quarter, taking it to over $150 million—it kind of made me think of that straight away when we’re thinking about net additive versus replacing things.

Is this eating Zendesk’s lunch? Is this adding new users that never had customer support before? Is this taking on jobs that weren’t being done before? How do we think about that? I’m waiting for Mike because I didn’t know it, but he has a customer-support product coming out.

Mike Cannon-Brookes

It’s already out. It’s doing very well. I can’t tell you how well it’s doing, but it’s doing very well. Our service management business could go public by itself, right? It’s a big business. It’s growing. It would have great numbers. Jason would be writing great posts on it.

Jason says it needs $4 billion to go public.

Mike Cannon-Brookes

For a banger. For a banger, Harry.

Banger IPO. For a banger IPO—get back to the question, Mike. Let’s stop.

Mike Cannon-Brookes

It’s a text-based field, and I say that with the broadest definition of text, in that voice and other things can quickly be approximated to text now with AI, and vice versa. Let’s assume it’s a conversation-based area. It’s probably 3% to 7% of every business in the world in service and support. Your bank, your insurance company, your car company—every company has this.

Then you get internal support and service, right? Most of the service management revenue is on the internal basis, right? Twitter/X, or whatever, has 300 or 400 service desks running on service management, right? But that’s not all internal service. So the idea of any team helping some other team in a business, or any team helping their customers, is ripe for making it efficient.

All the things you talked about—Jevons’ paradox—is totally true. But it’s also a measurable input and efficiency play. So I can say, I spend X; your product can very easily make X less. Is the less more, or is the money I’m saving more or less than I’m paying for your product? There’s high ROI. It’s a sellable thing. We go around the loop, right?

I think it’s a great category. Whether the category will grow or not is a different question. The hard part is in year 2 and year 3. The savings are baked into the business, so you have to be delivering continual value to the company. That’s going to be really interesting when we get through this set of changes in the service market most broadly.

You’ve got ServiceNow and Atlassian and others at the top end. You’ve got a lot of legacy providers out there. There are still tons of legacy providers of service. Then you’ve got lots of customer-service companies. Mark has Service Cloud over there. I think some people still use that. There are a lot of tools in this space. Zendesk probably went private for a reason.

I think Zendesk says they have around $300 million-plus of AI support revenue. It’s just the legacy business. That category is tough, right? They’ve got 2 streams going. The agentic piece is exploding.

The only thing I would say on why the category is exploding is that there are 2 threads helping support overall for so many of the players. If you look at Pave—I think that’s how you pronounce it—it’s an HR startup that tracks around 300,000 startups. No category has been more decimated in hiring than support. No one’s hiring; it’s radically down. That money is going into agentic products.

Traditional support—and a great AI support tool can do radically more than a legacy tool. I think software in the AI age, and this is why so many mature SaaS leaders are struggling today, is that there’s no more aha moment. There’s no more moment when you walk in and—yeah, maybe Sierra spent a month configuring your demo—but it literally answers questions like you’ve never seen answered before in your enterprise. You want to buy that product, and most SaaS products didn’t change for a decade. We fell out of the aha moment.

So there’s budget, there’s an aha moment, and we’re cutting more humans than ever. It is a great moment. The final point, not to ramble, is that the super-interesting thing about support is that it becomes a Trojan horse to do everything else.

When that’s the main agent—if that is the main agent you’re working with—you’ll start using it for sales help, research, marketing, or other things. These become more horizontal agents. Man, hiring is over in the space.

And bringing it back to Harry’s question, Harry, I think the reason we’re all saying, bluntly, that you’re wrong is that you were doing the “I don’t get this market; I don’t know why people have funded 10 companies” thing. I think what you’re hearing, and I think it’s true, is that it’s not a single market.

At a high level, you can say it’s all the same: people helping people. But internal help desk is different from external help desk. One thing we saw—we had a pre-LLM AI company—B2B help desk, B2B external support, has a different rhythm than B2C.

B2C is lots and lots of small interactions. B2B, you might have a long-lived interaction with multiple people internally. If one of Mike’s best customers has a P1 bug, you don’t just have 1 person involved. You have 6 or 7 people involved. You’re reaching across departments.

The internal workflow to manage that support is very different from thousands of people ringing about, “How do I change a password?” internally, or externally, “How do I get my ticket to the whatever concert?” So there are differences there, and there are differences in company size.

I think there’s also a difference in modalities, because you have people who lead with chat, people who lead with voice—and, yeah, they’re converging—and people who lead with video. I think there are a lot of different submarkets rather than 1 big mega-market, is my gut.

And I think it’s just like Harry, though: there are some interesting things that it tells us about how AI is changing the world. There’s no doubt that the overall service category is probably not going to have a lot of growth in humans, let’s just say, working in that field.

However, I would guess—and I don’t have any stats to hand—it’s a very short-term field anyway. My sister worked in telephone banking in the mid-’90s for a bank here, right? Everyone was there for a year or 2. They train you to do it and you move on. It was not a career job. It already had high turnover. That’s different from lawyers, where they’ve done many years of training and they’ve got 40 years of career, you know.

So it’s an interesting job field. It’s pretty different, firstly. Secondly, one of the interesting things that we see, both internally—I have a few examples of agents internally—but I’ve got a number of customers who’ve seen this phenomenon as well, is that your AI and your agents that are helping people, in the case of answering a question in service, need to be thought about much more broadly than just customer service, like “I lost my wallet in the Uber.”

The AI that’s answering the question is as good as the things you have written down. Some teams you see are deploying agents and then writing a lot more documentation, and their documentation is getting a lot more stepwise, because it’s less about humans reading it and more about the agent reading that documentation and then answering the question.

So you may end up with, in some areas, more writers and fewer people answering questions to get an overall better support response. We have a lot of examples of this in service collection. The more knowledge you hook it up to, the better answers it gives. It’s sort of logical. Some businesses don’t have all this knowledge documented because it was in the heads of the people who did the service. So that’s 1 interesting effect of AI, and it probably applies elsewhere.

The last thing I will say is that agent service collection is 1 of the areas where most agents are deployed, right, in terms of AI agents. It’s confusing in service because we talk about human agents that answer questions, and then we have AI agents that either answer the question or help the human answer the question faster in the complex scenarios.

In B2B especially, as they’ve already talked about, agents are far more highly deployed in that category. For us, of any of our collections, I think it’s the highest proportional deployment. The reason is because those agents are taking actions. So traditionally, someone’s coming in to ask a question.

If you think about HR service management, it’s, “Can I take these people out for lunch when I’m in the Indian office? I want to apply for leave or parental leave,” whatever it is, right? You’re asking some sort of HR-based question. There are answers to the question: yes, you can. And then there are actions taken: “Hey, I’m going to go and file this expense claim,” or “I’m going to go and do that.”

The AI in service that’s most interesting to customers often isn’t the cost-saving aspect, which is what we always talk about. It’s the ability for it to actually go and do things to make my business quicker. If it goes and files the parental application or resets your password—the canonical IT service thing—the actions that AI can take more accurately, the automations that we used to call them, now agentic automations and workflows, expand massively. That means you not only get your question answered; you get your job done, and that business moves a little quicker.

So, your TAM argument—what’s the TAM on that? It’s not really a thing we had before, right, beyond human service?

Rory O'Driscoll

It could be better than the humans. It enables that business.

If the whole form gets filled out for you for parental leave, and I don’t have to do anything and it’s all automated, it’s better.

Totally understand that. Rory, let me punch back at you, my friend. We, as we said, live in a resource-allocation world, or mindset, where we have the choice to invest in categories and companies. There have been 14 companies created in the last 2 years that have now raised over $100 million.

You then have traditional players like ServiceNow, Atlassian, Salesforce—the list goes on—Zendesk, Intercom. I can go on and on and on. And then we have the new generation of very fast and public companies, à la Navan or à la Airwallex, who are building their own systems. There is a generation of technology-first companies that are building their own customer support systems to deal with very intricate and complex needs.

When we live in this risk-adjusted opportunity-cost world, I don’t feel that’s a good place to put your dollars. When I’m competing against Mike, Mark, and the 14 companies in the last 2 years that raised over $100 million, with Bret from Sierra, I don’t think that’s a good place to be investing.

Rory O'Driscoll

And I agree. I think, look, we looked at a number of companies that were behind Sierra and Decagon and came to the conclusion that a direct head-to-head against those companies, when you’re an AI startup, is just a little bit too hard. You kind of go, there’s all the air out of the room with Sierra at the high end. I think Decagon have executed well.

It’s that classic: it’s a big market, but in the subsegment, you never want to find yourself third in a subsegment. You can be winning in an adjacent section, and then you can get unlucky because the big dog can come in and decide to annex your adjacent section. That’s just stuff that happens.

But agreed, there is going to be overfunding in all these sectors. A lot of it is that we’re all trying to pick our points. If I was to fast-forward—in fact, I said the same thing—if you look back in 5 years from now, you’ll probably find that 1 of the biggest sources of errors was, “Oh, we invested in the number 1 or 2 player in this little space here, but it turns out that this little space wasn’t a thing. This big space was a thing, and the adjacent competitor just says, ‘I now bundle this in and do it as part of what I do.’”

Suddenly, Mike decides he wants to be king of Service Cloud, and he just rolls over us because he adds it to every 1 of his customers, right? So, I hear you. I think that you’re trying to pick it.

By the way, it probably explains—I want to go back to it—it probably explains the Harvey financing, because when you have all the dynamics that you just outlined, it’s arguable, despite the TAM comment I made, that the best risk-adjusted bet—and we’re seeing it happen—is, “Screw it, I’m just going to pay the winner at any price.”

If you’re in the winner, you can’t—it’s the venture-capital equivalent of, “You can’t get fired for buying IBM,” right? You can’t get fired for sticking money in with a 1× preference on the consensus winner, which is why this market is way more consensus up and down the stack than it’s ever been in terms of venture. There’s not a whole ton of people trying to be contrarian out there.

There’s a whole bunch—I mean, the statistic I tell my partners that blows me away is, and I probably repeated it on this podcast, of the newly minted unicorns in Q1 of last year, by Q4, 40% of them already had 1 or more up rounds. In other words, the money is saying, once you are a winner: kingmaker, double kingmaker, treble kingmaker, pile on top. It’s too hard to pick. It’s too hard to pick the guy at $50 million pre that might make it. Just pile into the guy who’s made it, even if it’s at $12 billion. Worst case is a 1×.

I think what you’re going to be amazed by is that every time we get one of these new technology disruptions, everyone gets very frothy and yells and screams a lot, “It’s going to change the world.” It usually does. That means a lot of capital is raised and deployed, and a whole bunch of it turns out to be torched because VCs are very individual thinkers. They pile into the 2 winners, then the next 10 companies, and probably the next 15, and a whole bunch of those don’t really work very well. It takes 5 or 6 years for us to figure that out.

Technology just changed the world. A handful of winners are made. A lot of money is lost; net money is made—more than lost—in the transition to said new technology. And this has never happened before. This is a totally new phenomenon. We’re in the era of AI. Everything’s new and different, and they’re not a herd of sheep following the potential next winner.

And will that Harvey round pay off? To Rory’s point, 1 thing that they don’t say is that a lot of VCs want the logo on the wall. So when you walk into the office, on the wall is a nice logo that says, “Oh, you invested in OpenAI. You invested in Anthropic.”

Well, if you put money in at $350 billion, you’re probably not doing the VC thing of getting on the $4 million round. But you get the logo on your wall. That’s important, I think, from the VC marketing point of view, isn’t it?

Oh, we can put “early investor” on our bios if we buy some Atlassian today, can’t we? Is that not—

That’s right. Yeah.

Mike Vernal

Early in 2026.

I feel vindicated, by the way. Thank you for that, Mike and Rory. I appreciate that you conceded that I was right on the customer-support argument. It’s awesome. But I would love to finish on 1 thing, which is the Super Bowl ad. I’m sorry, we’ve got to finish on this.

Anthropic got the essay back from Sam, and then the CMO of OpenAI was like, “Betrayal, treacherous, epic disaster,” whatever the words were. How did we think about this and the response?

Maybe I’m just going to say what I think. Let’s confirm the facts. As I understand it, Anthropic ran a series of ads. I think OpenAI was going to do 1 ad—I’m not quite sure what that was, so remind me, please—and then Anthropic ran an ad trail saying, basically, “We don’t have ads in our AI. We’re good; you’re bad.” They ran versions of that ad, and then OpenAI kind of pivoted and ran a very “You can just build things” ad in the actual Super Bowl.

Is that the kind of chain of events here, Harry?

Absolutely. The chain of events really is: Anthropic ran a series of ads laughing at OpenAI for having ads, which is slightly ironic in that way. Then Sam and their CMO basically came back with these epic tweets. I think everyone getting angry is a sure sign of tension.

But actually, Brian Kim from Andreessen Horowitz just did a really good piece that basically said, “Of course they’re going to run ads. It’s a consumer subscription product, and less than 5% of consumers ever pony up for subscription products. So if you’re going to be the consumer product, they’re going to run ads, and that’s ChatGPT. And if you’re going to be the enterprise product, you’re not going to run ads, and that’s Claude.”

95% of people watching that game must have gone, “What are these people talking about, and why are we taking away from Bad Bunny for these morons to talk about something I don’t even know what’s going on?”

Back to the prior discussion, when these technology revolutions happen and everyone pours in capital and torches a whole bunch of it, 1 thing that’s again unique about this era is the purchasing of Super Bowl ads to try to, I guess, justify ego. I think that’s largely what’s going on here.

Let’s face it: it’s human beings. They’ve got a lot of capital, so they make Super Bowl ads. Again, this has never happened in the history of technology before, except Pets.com. I thought you were talking about AI.com—the guy that bought the domain for $70 million, spent $8 million on an ad, and then is selling OpenClaw. And I’m like, that is also a weird business, but maybe they’re going to—

Rory’s comment: what do they call them online nowadays? The normies, the regular people who just wanted to watch the Super Bowl are like, “What the hell is this about? Next, give me the Doritos commercial.”

And 1 thing I’m getting to know you, Mike, is that your tone doesn’t change when you move from fact-based comments to acute sarcasm with literally no variation in tone. It takes a while to adjust to, but you’re so right, man. It’s like when you said it’s never happened before.

Rory O'Driscoll

We all flash back to ’99. We flash back to crypto. This is classic top-of-the-bubble stuff. You have $200 million burning a hole in your pocket and you're like, “I can spend $5 million and be in the national agenda. I'm in the Super Bowl.”

You know why this ad was weird, though? I'll tell you what was weird about it, for what it's worth. Anthropic is a generational company. It changed coding. It changed everything. But the only place Anthropic is weak is consumer. I'm one of the 7 people that uses Claude instead of ChatGPT as a consumer app.

Of all the things they could have done in the Super Bowl—say, the engineer's pal, the product guy's buddy, all the things they could say about how he revolutionized product creation—they talk about the one area they fail, which is, “We won't put ads in an app nobody uses for this use case.” So it was some sort of meta-ad, because it's advertising for its worst product. It's nothing but a dig. It makes no sense.

I saw a smart tweet that made the point: the redeeming argument on this is, “Never thought I'd say so, we're actually not talking to the 95%.” It was 2 people exchanging signals to each other and each other's employees in the Valley, using $5 million of money to send a signal, right? This is Anthropic reinforcing their “we are the consumer, we are the enterprise good guys here” message.

They're really messaging to maybe 10,000 engineers they might want to hire, and literally the other 325 million people involved are just—plus our head of safety quit this week, so let's mask that over with the Super Bowl ad.

Rory O'Driscoll

It's all just posturing. It would be so much cheaper to buy a billboard on the 101.

I think—but it does go back to the original discussion, Harry. I think what's interesting there is it's a sign of the times. I'm not sure we can call it a sign of the top. I don't think that's probably true at all. But it's a sign of the times.

Look deeper at it, right? If you're burning $1 billion a week and people are going to keep funding you, the ads are cheap. They don't make any difference, right? Your CMO doesn't have a constrained, capital-efficient, profit-EPS calculation right now. One day they will. Right now, they do not.

So right now, everyone is maximalist. They're trying to be in every single category all the time. What's fascinating is you go to the SaaS and software world, you have financialization going on a lot because they can't afford to be maximalists, because those pesky public-market investors are asking valid questions around, “Hey, profitability and growth and all these things.”

You have a lot of categories that are raising tons of money and don't have that problem. Hence, you get these ads as one example of that. That will normalize. It does every time. I think we're automatically assuming that Super Bowl ads are, I'll say, frivolous, a sign of the top, not actually effective.

Like Wix, for example, which is obviously incredibly hot in the public markets, I spent an hour and a half with their CMO the other day, who's bought 6 different Super Bowl adverts over the years. This year he bought 2: 1 for Base44 and 1 for Wix. He's like, “The impressions that you get from the Super Bowl—the effectiveness of this advertising channel—is immense.” I back it entirely.

I don't think either of them were, like, top of market—“We got a billion.” I think they were brilliant uses of capital.

Okay, I'll answer that, and then I want to come back to Mike's point because I think there's something more important in that point. But on your point, if it was a brilliant use of capital, you'd see it happen consistently across the cycle. But in fact, you only see tech companies when they don't have capital to burn. Marketing is the first thing that gets cut. It's not a brilliant use of capital in general.

Therefore, what it is is a good way to spend the marginal dollar when you've already spent it on everything else. But I want to come back to Mike's point because I actually think there's a more serious point in there. You made a comment that these guys are not financialized and the public SaaS companies are financialized, and I think there's a bigger comment in that—and maybe it applies to you.

When I was thinking about these fights between a public-company incumbent and a private-company next-generation player, I think one of the biggest things is the public companies have this constraint of having to make EPS, having to grow, whereas the private companies are, as you say, with no marginal costing taking place. Do you think that impacts public companies' general ability to compete, or do you think it makes them more efficient?

Mike Cannon-Brookes

Man, so many great topics. Firstly, I think Super Bowl ads—just very quickly—are priced that way because that's what people pay. It's an efficient market in Super Bowl ads, surely, right? Doritos pay $8 million because it sells a bunch of Doritos. It's a good investment. Is it the best use of capital? Probably.

Why do Doritos and beer companies advertise all the time? For some businesses, like Wix, I would argue it's a totally sensible investment. For others, it's completely ego. The CEO wants a Super Bowl ad, so they go bake one. That ego stuff happens when capital is not being efficiently spent in a business, right? Both of these things happen in every Super Bowl, probably in some crypto, whatever.

There's an interesting argument about public companies, and maybe why—I mean, I know things I shouldn't know, I guess—but the good public-company CEOs are often founders, as you've pointed out a few times. You're trying to take a long-term view and manage the short term. You are trying to do both.

We have massive investments in AI. We have to deliver public-company results, and that's the job: to do both. If we just focus on the public-company-results part in the very short term, we will not have a good business 5 years from now because we won't be investing in building AI. But that necessarily takes away from something you could otherwise be doing today. And that's the balance of capital allocation in any of these businesses, right? Microsoft has to do it. Everybody has to do it, right?

Capital allocation is not easy. But a good leader, I think, will allocate capital to those investment areas and allocate enough capital, and then tell a story about why that capital is being invested, right? We spend a huge amount of money on R&D. We spend a huge amount on AI. I would argue, if you're a public company today and you're not spending a lot of money on fundamental, new R&D around AI and your product category, you're probably in trouble.

You can't see that in any public-company accounts, right? It is very hard to actually see where they are allocating their R&D dollars. But I would guarantee you a lot of them are moving more and more and more toward thinking, “How is this going to change my world? How am I going to hire in this era? How am I going to do this?”

You have to tell a story to your investors, your shareholders, your stakeholders, and your employees about why that's going to make a difference. Why are you going to be relevant and competitive in the AI era? It doesn't mean you need to make a foundational model, but you have to be relevant and competitive, and you have to actually do both.

Now, Rory, I think you're trying to make the point that maybe there's a world where the free capital in the private markets makes it impossible for those spends in the public markets. I don't think that's the case. You've had Adobe and Microsoft be public for an awful long time. You have to make tough choices. You have to make capital-allocation choices. You have to bet on the future to some extent. Not every company's going to get that right. But I don't think that's impossible at all. I think it's what you have to do.

Great answer. Reading the commentary after your last result, listing the things that people bust your chops on, they're giving you grief about stock-based compensation, and then you're looking on the private side and you're like, “These guys are doling out SBC like it's candy, and no one even calculates it,” right?

It does feel—I mean, I'm glad to see that it's not doing your head in, as we'd say in England—but it does feel like it's two separate games. Almost like both teams are playing the same game, but one of them has to play it with different rules, which must be a little irksome.

Mike Cannon-Brookes

Yes, but the difference in rules—again, I've said this to a few people—we're a better company because we're a public company. We've become a better business, right? We're better at forecasting. We're better at planning. We're better at executing.

What we have to do is not replace forecasting, planning, and executing with strategy. I think that's where it gets wrong: where you just focus on that piece of the business. You still have to strategically compete and grow and go into new areas, invest and build things, right? You absolutely have to do both of those things.

I think managing the finances of that is very challenging at the moment, right, for all the reasons you pointed out. Totally agree with the challenge that's going on. Secondly, it's very hard for investors to see winners in that process right now, right? If I say, as generically as possible, I think that's the challenge, right?

Where is that money going? What are those people building? What is the talent? How is that likely to change your strategic positioning with customers over the next few years so that you have sustainability and durability? That's very hard to see at this current time, right, for investors.

If you look at any list of AI winners and losers that's above the fold, let's call it, every list is different.

That alone tells you people are like, “I don’t know,” somewhere over here. That’s our job. That’s my job. I’m not complaining about that for a second. My job is to tell a great story about why we’re a winner, then make us a winner, and do both at the same time. And I have to fund that.

You’ve got to enjoy that. You should be doing a different job. I love you, and I’m so grateful to you for being a friend, being awesome, being so good. The numbers are so good, but the stock price is down so much. Do you think you’re doing a good job telling that story?

Mike Cannon-Brookes

This gets to the CEO scorecard, right? We had a fantastic quarter. Our numbers in almost every category are great: delivery of our goals. The external, current-time judgment of the delivery of those goals, leading to future value, is not in a good spot. There’s no doubt about that, and we’re working hard to change that.

We have the same business we had a year ago. We will still have a great business in a year and 2 years, and my job is to make sure we have a great business in 5 years’ time. Look, there’s the mental-health aspect of it for staff and things like this, and then there’s, “We just got a lot of work to do. Let’s get back to work and just go do that,” right?

If we keep delivering great results, we’re not short in the Bank of England here, but at some point you’re like, “I don’t know. We’ve got to do what we’ve got to do,” right? And you’ve got to also adjust. I said this in my shareholder letter. We spent a lot of time writing our shareholder letter. You’ve got to accept reality as well.

Sometimes I see people and, as I said, I do a lot of SaaS therapy at the moment. One of the things I’ve told a few people is, “Dude, you’ve just got to accept reality and then go build something. Go do something,” right? You can’t pontificate about the technology journey, about creation and everything else. Guess what? Part of creation is destruction. You’ve got to go build value for customers. You’ve got to go build new products and technologies and services.

The way I see it, we’re going to create our way out of this problem. We’re not going to hide in a hole and wait for it to go away. Part of that is accepting the reality of what’s changed and also what’s changed in a positive way. That always seems very negative. We can build things way faster and way better than we ever have been able to before.

I think we talked a year ago, whenever that was, about how if we refounded Atlassian today with 350,000 customers, 50,000 enterprise customers, a couple of billion bucks in the bank, 10,000 people in R&D, and a great distribution engine to get to all those customers and deliver new value, you’d be like, “It’s a pretty good starting point to start a business, right? Let’s go build some stuff,” right?

You’re right, Mike. When you talk about creating value, we as 3 VCs know exactly how you feel. Yeah, we are the value creators.

Rory O'Driscoll

Just to be clear, it’s clear when Harry’s being sarcastic.

I would say—I don’t know what you think of your SaaS group, which is pretty elite, Mike—but maybe only the best SaaS B2B CEOs are really up for that. A lot of them are introspecting. I talked to one that quit at the end of the year at hundreds of millions of revenue and said, “I was told AI wasn’t important in our space.” That was the goodbye message.

Not everyone has it in them to push through this next wave. Maybe only the best do. Maybe only the best do. Maybe—I mean, you’re one of the best, right? We don’t have to rank you.

Mike Cannon-Brookes

And that’s okay, right? I think we shouldn’t pass judgment on that.

Rory O'Driscoll

I totally agree. It’s not even about being the best in the world. Look, it’s the same in our industry. Some people decide, “I’ve done enough. I’m 50.” Some people say, “I’ve done enough. I’m 40.” You only get 1 life, right? Some people really want to go to ballet at 50. I’m totally with you, Mike, on that one. It’s actually a better call if you don’t have it in you to do it.

I actually admire more the person who says, “You guys should get someone else,” than hanging on, waiting to get the head shot, and just calling it. I actually agree. That’s a harder call to make, right? That’s a very difficult decision in our business.

One partner said to me years ago, “Our business is a learning business, and the day you stop learning is the day you should say to your partners, ‘In about a year and a half, you should be replacing me,’” right? You can’t just be sitting here. You can’t take the money and not play the game.

And I agree. So you should tell those guys who aren’t feeling it—it’s really simple: “Congratulations. You’ve made life-changing money. You don’t have to do it anymore.” How’s your number 2? Because he’s in charge now. I’ll bring back that founder who left. Are you working harder, Mike, than you have in the last few years?

Mike Cannon-Brookes

Am I working harder than I have in the last 3 years? Yeah.

My co-founder, God love him—we had a whole bottle of wine the night before last—retired a year and a half ago, so that’s a big 23 years. He is full of gratitude, which is really nice. He understands, right, that creates more work. There’s no doubt. At the moment, I start working at 5 a.m. every day.

Yes. People are working harder now than they were before. That’s 100% true, right? Because of the disruption and the speed and the change and everything. I don’t think that’s a bad thing. That’s a normal thing. That’s an okay thing, right? The question is, do you enjoy it? And do you have enough balance?

Again, I am the worst person to be a therapist to any of these CEOs and SaaS founders, but I do tell them what you’re saying: those people are making the choice. It’s balance, right? Do you have enough time with your kids? Do you get out in the trees? Do you walk around? Do you do some exercise? Whatever it is that’s your jam in terms of balance, stop doing that for a year or 2. You’re going to be in trouble. I believe—I’ve done this long enough—you can’t shortcut those things.

Secondly, are you enjoying it? And don’t just tell me yes in 1 second. Wait. Think. Go away, take a weekend, drink a bottle of wine, drink a green tea, whatever is your thing. Are you enjoying what you’re doing? Would you choose this job again today? Because it’s tough out there. It’s going to be hard. Welcome to the technology industry.

If you’re not enjoying it, no harm, no foul. Go do something you do enjoy, right? Move to a different arena, if you want to use the phrase we throw around all the time in the tech industry. The enjoyment is sometimes forgotten in the ego, the need to win, to disrupt, to grow, those sorts of things.

Like you said, if you’ve built a great business—a couple hundred million ARR, whatever it is—that’s not a reason to just do it decently. Give your team time, give your company time. Make no mistake, there’s a thoughtful way to do any of these sorts of things. I think it should be seen as a mark of amazing human achievement to do that.

The problem is, if Mike said that to most of the people I know, they’d quit. If you said, “Do you really enjoy—do you love every minute of your job as founder and CEO? Are you getting enough time with your kids? Are you walking enough? If not, you should…” I’m being a little facetious, but not completely, right? Not completely.

Mike Cannon-Brookes

I’m not being polite at all, right? There are parts—there are days that suck. There are times that suck. But over some average—90 days, 1 year—Rory’s point, you’ve got 1 life, right?

We got into this because we enjoyed creating things. We enjoyed building teams and businesses and delivering value to customers—not in the traditional way, but literally sitting down with customers and seeing what they do with your [expletive] is awesome. And you’re like, “Man, that’s so cool. Hey, let’s go do some more of that,” right?

So you’ve got to enjoy that part of the job. It doesn’t mean every part of the day or the job is going to be easy. It doesn’t mean you don’t sit there with your team and go, “How the [expletive] are we going to deal with this?” right? These guys are coming over the wall.

But if you don’t enjoy that, I don’t know. There’s some analogy that every era is going to be different. The echoes are the same, but there are different ways, to your point, of learning. You’ve got to learn. You’ve got to think about how you’re going to compete. You’re not sitting there with bows and arrows if people are coming with machine guns or whatever.

Exactly. But I do think you can be balanced. I do think it’s important.

You think all founders are a little crazy?

Mike Cannon-Brookes

Yes.

Yeah. How can you be balanced if you’re crazy? Like, there’s no B—because you’re going to be miserable or bored, or you’re not building. I don’t think Elon looks too happy to me. I think all the good founders are crazy.

Mike Cannon-Brookes

There are some people…

I think all the good founders are crazy.

Mike Cannon-Brookes

…who have unique choice sets.

Rory O'Driscoll

That was well done.

Mike Cannon-Brookes

I think, in whatever way you enjoy what it is that you do, you choose to be intentional with your own time, right? I spend a ton of time with my kids because it’s super fun and it’s super different, and they’re awesome little human beings. I’m trying to grow them into awesome big human beings, right, and help them on whatever journey they’re on.

That is enjoyment. I think it makes me a better CEO to do that. I don’t spend 60 hours a week with my kids, you know what I mean? They’re sleeping and this and that and other things, but I don’t spend 100 hours a week at work.

Right now, when we were younger, we did, and that's the evolution. But you have to learn to do that. I think if you spend 100% of your time at work, you make worse decisions, right? But you have to do the job, right? The balance is really tricky. Intentional time management is really tricky.

Do I spend as much time with my mates as I used to? No. We catch up once a quarter. We do this and that, but there are things you have to trade off. All I'm saying is, if you don't do any of that outside work because you're like, “This era, I've got to knuckle down and do 100% of this right now,” I don't think that's sustainable. Maybe that's just not for me.

It's not. Every time I give that speech to my wife—“This era, I've got to knuckle down and do 100% of this right now”—she just looks at me and says, “Stop. You're just going to go insane, and then you're going to become irrational, and then you're going to make bad decisions. So we're going for a hike for the next hour and a half.” And I just say, “Okay, Matt, I'm going.”

Mike Cannon-Brookes

And Harry's moving in for the kill to tell us we're too old and we're going to get rolled over by him. Well, I do think that, but that's a little bit rude. And we have a guest, Rory, so I wouldn't say that with our guest, obviously.

No, I actually have a secret for you, Mike: just work with your great mates. Yeah, I have become great mates with Rory and Jason, and I now get dedicated time with them every single week. What a joy. Well, I mean, how lovely.

Rory O'Driscoll

Okay.

Mike Cannon-Brookes

On any team, the best things that we've done have involved a series of fantastic leadership teams—really good friends that I still hang out with. We had a whole bunch of them over the Christmas period: old-time Atlassian leadership-team people who came to join us. They're amazing people. We've been in various trenches and scuffles and fights together. We've had some victories and some losses.

Do you get enjoyment from that part of it? You're winning more than you're losing—great. And you go back and you can have people that you can have a cup of tea with 10 years later. Man, we did a good job there. You know what I mean? There's a part of the human tribal getting together that your leadership team, wherever you are, in whatever business—

Rory O'Driscoll

Has to do.

Mike Cannon-Brookes

That has to be part of your enjoyment, right? And you have to create that, and it's hard to create. Don't get me wrong, it's tricky. You can treat people as employees and mercenaries, or you can say, “Hey, we've got a challenge to do, and we're a group of people in a room. We're trying to deal with this challenge. We need help. We all need help to do that from each other.”