20VC：Atlassian CEO Mike Cannon-Brookes谈代理为何不会取代界面、企业为何不会信任中国模型，以及专用模型的价值为何仍未有定论

Mike, dude, I love our conversations. They are always my favorite, so thank you so much for joining me today.

Thanks for having me back, man. It's appreciated. If I get invited back, I must be giving interesting answers. I saw your little rant the other day. I loved it.

You give honest answers, which I don't always get. The thing I really want to start on is that we see a world of agents, and agents remove a lot of human need and interaction with data providers or data sources. How do you think about a world where value shifts as agents retrieve and do tasks for us, and the potential for an Atlassian to be a data provider, not a doer?

1. Agents Will Not Replace Interfaces

I think there's a premise of the question that's slightly incorrect.

I can see that being the case for certain tasks. Where the tasks are very functional, that is the case. Where the tasks involve a lot of taste, interaction, and collaboration, I don't think that will be the case, certainly not in the same way. So it's very hard to give an answer across all types of tasks.

One might ask about certain types of shopping, if you take a consumer task—not something that we do, obviously, so probably a safe example. There are certain types of shopping where I just want to get the thing and move on. If I'm buying a charging cable off Amazon, just give me the charging cable. I don't care. I don't need anything else.

There are other types of shopping where I may enjoy the experience. I may want to go to the store, do a lot of research, and do it myself. There's going to be a whole range of things. So yes, for some tasks, it's going to get rid of the pedestrian work—I just want to get the thing done.

But in the enterprise collaboration world and in the business workflow world, agents are going to help, and are already helping, in a lot of cases, move the work forward and let people honestly focus on the higher-value, creative parts of the job. I think that will change a lot of jobs, but I don't think it's necessarily negative, nor do I think it's bad for Atlassian. I think it's great for Atlassian.

We want to help be a part of that agentic world. But the experiences of how we do that are a long way from being solved and settled in enterprise software.

If I were to ask you, on the flip side of that, what is the core strategic decision that will determine the success of Atlassian in the next 12 months, what would your answer be?

2. Headless Is Brainless

I'm well on record: headless is brainless by definition.

You've said this before, Mike. You stand by that?

Absolutely. When people talk about headless in that way, if you're giving up on your user interface, if you're giving up on evolving your experience for a customer to solve the problems that you set out to solve for them, and if you are not a significant part of that experience, then yes, I think that's a brainless bet.

That does not mean we're not big on MCP or CLIs. I think we have one of the top 5, top 3 MCP servers in the world. We're hugely invested in MCP and its ability to allow agents and harnesses to contribute data to our platform and to the Teamwork Graph, and to read data from the Teamwork Graph.

MCP and harnesses show off the results of our graph—how great the quality of our context graph is—in solving all manner of problems for our customers. That's great. But if our interfaces and our design are not evolving to solve some problems, or some significant part of the problem set, better for the customer, then yes, I think you're going to be in trouble.

So you fundamentally disagree that, in a world of agents, interface and design become less important?

Absolutely. I think they become more important. I think design and experience are among the keys to success in the AI era.

We are massively invested in design and experience at the moment in a way that we never have been before, and we've always been quite a large investor in design and experience. We have continued to upgrade our experience in a way that I would say few enterprise software vendors have done over the last 20 years.

It's really easy to make an app—I always think—better than 90% of apps out there today. That's easy. But in 5 years' time, that app will be better than 20%. Experience degrades as we have higher and higher standards.

To upgrade app experiences is really hard. It takes a lot of work, a lot of R&D, and a lot of difficult craft work to continue doing that. We're in an era where we're all going to need to uplevel our experiences to handle the AI era, but I do think design and experience are even more important in the AI era. I think they will be the differentiator.

If you argue that technology creation is increasingly cheap, and that technology creation is abundant and verdant, then the differentiators, to me, clearly include design and experience. They will be what makes you stand out as an application, product, or service.

So you don't think we'll have horizontal agents that stretch across a number of different vertical providers and do functional work for us? You think the design and experience of those vertical providers will remain incredibly strong? This is not a leading question, by the way, sir.

It's too simplistic a view of the world. I think we will certainly have some horizontal agents that handle a breadth of tasks. I think we will also have vertical agents that go very deep in a singular task. I don't think agents will replace all the interfaces. I think all 3 of these can be true at the same time.

As an example, we have lots of agents that run on a lot of platforms—from Claude, Codex, Cursor, and places like this, to Teams and Slack—that access the Atlassian platform and contribute data back to the Atlassian platform. That does not mean that's the best interface to do all the jobs on the Atlassian platform. It can be very useful for a lot of different things, and vice versa.

We're announcing a lot of features with our partner Figma, for example, and another set of features with our partner Databricks. In both cases, allowing you to get to Figma functionality and Databricks functionality from within the Atlassian platform is really powerful.

If I'm reading a Confluence page and I want to select a piece of text and say, "Can you check this number for me?" it goes off to Databricks or Snowflake through the Teamwork Graph's data context, gets you back the answer, or gives you a small chart. That's great.

That's not saying Confluence is a BI tool. It's not going to have the power you need, but sometimes, when you want a quick answer retrieved from somewhere else, it's a very different experience from living in that world and doing that thing.

The same applies to Figma. You can interact with the Figma agent inside the Atlassian platform in lots of places, and you can interact with the Atlassian platform from inside the Figma interface in a lot of places. It's the definition of A2A, I think. It's one of the best examples of A2A across multi-craft teams.

But both those interfaces have strengths in their own areas, right? I don't think there'll be one über-agent that just does everything magically. I'm a big bear on the magic-button thesis of AI—that there'll be one button we press and it does it all for us. I don't think that's how it's actually going to work.

So if I go back to that question for you, which is the core strategic decision that will determine the success of Atlassian in the next 12 months, what would you say it is?

Well, one of the top handful would certainly be continuing to invest in and have the best enterprise context graph in the world, continuing to stay ahead. I do think what we talked about before—continuing to evolve the design and experience in the agentic era—is going to be a really interesting challenge.

I love spending time with our design team at the moment. For the first time in a long time, we're solving foundational design challenges again. I would say the mobile era was the last time we had to think about completely new design paradigms. A lot of the features we're announcing here in Amsterdam are really foundational design breakthroughs.

How should agents appear in a broad collaboration platform? How should I interact with them? How do I deal with the problems of humans interacting with agents, agents operating on behalf of humans, and knowing what is AI-touched content and what is not? Is that a good thing or a bad thing? It's one thing for AI to generate a piece of text 100 percent. It's another thing for it to help you write something.

How do we explain all this to somebody, and where does it matter? Provenance and accountability are incredibly important. Speed and understanding. Do I want to see everyone's agent traces? No. But do I need to understand what they are and learn from them? Yes.

There are lots of foundational design problems that I think will improve customers' results. It's a really interesting time for that.

What is the foundational design problem with agents that we have not solved yet?

3. Agentic Interfaces Need Better Design

I mean, text boxes suck, firstly. Giving you a little text box and saying, "Here is a one-line piece of text that you can type characters into, and it has unlimited power," and it's just a blank white line—you're like, "Well, that's a terrible user interface to start with," right?

We're seeing lots of attempts at changing this problem. Agents, chat in general, and LLMs are so powerful, and most users just stare at them and use them like Google Search. That's the reality of what they do. But how do you explain the power to them? How do you give them the right power at the right times?

There's a lot of talk about agentic UI—the sort of Word ribbon that changes which functions it gives you at which times—and how it takes that into account as those functions evolve. There's a lot of this that's a challenge, firstly.

Secondly, agents doing things on behalf of users is an increasingly common pattern. I think software teams are really at the forefront of a lot of this, but I think it'll come to legal teams, HR teams, and marketing teams. It's all going to be the same.

If I'm using Codex and it does something for me—for example, if I'm just writing code and Codex creates a pull request—did Harry do that, or did Codex do that? Or was that Codex on behalf of Harry? How much of that was Harry?

Why does it matter?

Because if you've been heavily involved, it changes the nature of the content, and we need to understand where that comes from. People are going to want to see, "I just want the human comments here. I want just the agentic comments." There are valid reasons to see both.

This is a really interesting design challenge. The provenance of where content has come from is a non-simple question. We see this with Anthropic's watermarking features. It implies that watermarked text in AI is bad. That's not true.

The question is: did I contribute 1 or 2 ideas? Did I say, "Write me an essay about agentic design," and it wrote me 3,000 words? Or did I say, "Here's my essay on agentic design. Help me make it better," and it just helped me refine it? Completely different, same set of text, same watermark. How do we help you solve this problem?

Then you have a lot of business processes. Every company is just a system of processes put together, a big organic system that's constantly trying to improve its processes. How are we thinking about improving the process versus working on what is flowing through the process? How do we evolve how we run that?

We see a lot of this in the software factory thinking that is very prevalent at the moment. It's not always a great analogy, I think, because a factory is a very mechanistic, defined-edge thing, and software is fundamentally very creative. But the idea is that I'm working on the factory as much as I am watching what's going through the factory. I'm not stirring or chopping or doing the things anymore or stamping out steel. Machines do that, but I still need to operate the factory.

I still need to initiate. I still need to handle exceptions. I still need to steer, co-create, and co-guide often. The interfaces for when I jump in and when I don't are a real challenge.

As there are more things going on in a company, how do I have confidence that when things go wrong, I will be notified, and when they're going right, I'm not notified? If I need to get into the detail, I can, but I don't have to see all the detail because there's just a sea of things going on.

There's so much of this that's really exciting—a field of change in design and experience right now.

You spoke about the watermarking of different content types that we've seen. I think in 5 years' time we'll have watermarked content for human-created content because it will be 90 percent AI-created content. That sounds very strange to a lot of people today.

What do you believe today that is quite spooky today that will be incredibly common in 5 years' time? If you think of parallels, finding your wife online would have been crazy 20 years ago. Now 60 percent of people do it. Putting your credit card online—I remember when I started buying on eBay, my parents wouldn't let me put my credit card in.

When we started selling software, you wouldn't believe it. We had a fax machine, and you would download a PDF off our website. It was a PDF. I think we had a Word version as well. You could download the Word document or the PDF, write in your credit card number, and fax it to us. We would come in in the morning with these rolls of paper on the ground, and we would punch it into a physical machine.

That is how we originally took payments for internet software.

Scott, Scott, get off your fucking honeymoon, dude. We got another one.

Yeah. We never shipped a disk. We were in the internet era, but still, payments back then were faxed numbers. That's how companies wanted to pay you, and you would bang it into a physical machine, typing the number in.

So, dude, what's your kooky thought that you think will actually be incredibly prescient for 5 years' time?

4. Synthetic Content Becomes Normal

I think in 5 years' time, maybe the watermarked AI. I think there's a good chance that the whole watermarked AI—the reverse of what you said—is probably true. We don't know what is untouched by AI, and we also don't think that is good or bad. We don't attribute quality or subjectivity to AI-affected content, AI-improved content, or AI-written content of any form.

Just like today, when you go to the movies, we don't say, "I want to see films that aren't color-graded." It's not a thing. You just go and watch a film. You want a great film. There's CGI in all sorts of films you don't expect. They film in Sydney at nighttime, and they make it daytime with CGI and stuff. Does it matter to me? I don't think so, really. I care about the quality of the story.

I think whether text is AI-generated—how many people are like, "I don't want to read your content if Word helped me with spell-checking and grammar-checking"? There's some version of that that's similar. I understand there's some version where it's entirely AI-generated slop. So how will we think about this spectrum?

I suspect we all just get used to dealing with the fact that it may be entirely synthetic content, and hence artisanal, crafted things will stand out more. I still think we'll probably have human judgment that makes things stand out.

You work with some of the biggest companies in the world. I was chatting to Jason Lampkin the other day, and he was like, "Dude, I tell you, I tell you, none of them want to work with the open models that are Chinese. They just don't want it."

I'm a big investor in Fireworks. I believe in open models. I think it's a great thing for the ecosystem. But I am fearful that enterprises don't want to work with largely Chinese open models. Do you see that, and how do you think about that statement?

5. Enterprise Trust Shapes Model Choices

Jason's increasingly trying to run one of the smallest companies in the world, I think. So, for good or bad, I always find his takes amusing.

I think that is something that will probably evolve. We want to be very customer-led in our decision-making, and there is certainly some customer pushback in some places to, quote-unquote, Chinese models. There is no customer pushback to open models. We use a lot of Nemotron in our platform, for example. We have a deep partnership with NVIDIA. That is fine, right?

We use some Mistral. We use a bunch of other open models in different parts. Our job, again, is to apply AI on behalf of our customers. It's not to pick a model. It's to use the best model for the best job, blending all the models for the best price for the customer and to get the outcome they want, right? Applying the AI is our job.

But if there is customer pushback to something like a Chinese model, quote-unquote, that is something that we have to be very careful of. We don't use any Chinese models in production. We're very clear about that.

I think what people don't understand is the distillation and the back-and-forth. Knowing it's a Chinese model is really interesting, but a lot of the US models are increasingly using Chinese models to train themselves, and they don't say that. You end up with all these weird loops where I'm like, I don't know if that disappears over time. I suspect it does, but I think we'll have to see and watch.

I do talk to customers sometimes and say, "Do you use Linux?" And they say, "Yes." And I say, "Is that European?" And they say, "Yes." And I say, "Well, I think something like 40% of contributions, or something like this—it's quite high nowadays—come from China." There are some amazing software engineers in China doing all sorts of things.

But because it's open source, we think, "Well, someone's checking all of that." And actually, weirdly, with AI, we can check that even better than we ever have been able to before. They're happy to use Linux, so—

So you think the skepticism is overwrought?

I think there's a geopolitical issue around who has this sort of strategic control of the world. That is a different issue from whether we should use a model that was trained in China or by Chinese nationals or by someone—these are very different questions, if you know what I mean.

In terms of who's running the world, America versus China, that's a big question. That's out of my league and my pay grade.

Well, I mean, I—I'm sorry. This is where, dude, years ago, you remember when you first met me, I was like, "Harry Potter." But they're not. They're kind of one and the same because the people building the models are not able to leave, and their families are not able to leave without letting the government know where they are going and when they are going. And so they are actually merging.

But I'm not commenting on any of the social issues or other issues in that area. What I'm saying is, from a technological perspective, for someone to say, "I'm not going to use a piece of technology that is open"—if I'm saying, "I'm not going to use this piece of open-source code," I'm like, "But I have the code."

If you choose not to use it, that's totally fine. That's your choice. But for an enterprise to say, "I'm not going to use a cheaper alternative that'll give me better results when I know what's in it," I think is going to be a very hard long-term choice.

I suspect people are going to take those open-weight models, which they are already doing, and fine-tune and train them and call them their own. Most of the people who are training their own model—large company X claims they've built an amazing model for purpose Y, and it's got some wacky name. We've seen this many, many times. Almost every single time, if it's not Nemotron, which it's usually not, it is a Chinese model that they've trained or added their own fine-tuning specialization on top of, and then they call it their own model.

I'm like, hang on, that's bullshit. That is the same thing, right? You've added to it and you've tweaked it. You haven't trained it from scratch.

That was actually going to be my question. We've seen Harvey, and I think Ramp, talk about the benefits that they've seen from building their own models. You very clearly said last year that that wasn't a strategic focus for you and that you weren't going to go down that route. Now, with the benefit of hindsight, has your thinking changed when you see all of these companies come out and talk about the benefits of building their own models? Have you been validated? How has that thought process changed?

It's a great question. The jury is really out on the benefits of building specialized models at the moment. There are definitely cases where specialized models are doing really well. Most of the people building specialized models, I would argue, are not in those cases, though.

If you have a lot of proprietary data or very specialized applications, there are definitely approaches to that, right? If you're doing certain robotics, for example, and world models, there are a lot of reasons why people are actually doing it that make sense to me.

There are a lot of others that I don't think will stand the test of time. They may have a price-performance benefit for a specific application at the time of shipping, true. Will that price-performance hold up over time? I'm not sure it will as generalized models get better and better and better for a lot of applications, firstly.

Secondly, a lot of people who say they've trained their own model, I think they're being a little bit fast and loose with words, as far as I'm concerned. We could say we've trained a bunch of models at Atlassian. I say that's bullshit. We have a bunch of very specialized open-weight models that we fine-tune for particular purposes, which give great price performance for our customers. I don't claim they are our own models, I would say.

That's where I think the claim is often vastly overstated. Most of these people are taking a Chinese model, generally, applying some domain-specific intelligence. You see they train them for, like, a million bucks, and I'm like, there's no way you trained this for a million bucks. You took someone's billion-dollar model and tailored it, if you want to think about it generically, for a particular application.

That's totally fine, but that's not really training your own model, I would say, right? If you're not starting with raw data—if that's their definition—then we do that in a bunch of different places, but it's not my definition of building a model from scratch.

What, if it played out, would prove that thesis and your conviction to be wrong?

They could continue to show exponential gains over and above using the latest foundational model on a sort of price-per-watt basis. We are seeing many more applications, and it's very hard to answer these questions generically because it depends on the target application, the time to answer or solve a problem or a question, the cost, and the period of time you're in.

If you look at an agentic harness, it is a way of applying a set of prompts and configurations, if you want to think about them that way, to a certain type of model and reapplying it many, many times—if you want to think about it as looping—so that you can get much more value out of a given class of model.

And yes, if you apply better models, you get better results from a lot of those harnesses. Harness design is really, really important nowadays, but it's not model progress. These specialized models have to prove that they can stay ahead of the generalized model for a given price-performance ratio.

I don't think I see many of them doing that, certainly not the vast majority of them, because you would need to have a continued specialized dataset and a problem that doesn't get generalized away.

Many people who had startups, hopefully none of whom you've invested in, that were building their own models 2 or 3 years ago—I think now the foundation models have just basically smashed what they do, right? The quality of the generic model has improved far faster than their ability to stay ahead with whatever specialization they were doing. So I think I would need to see the reverse of that.

Luckily, none of my companies have, so phew, great relief. The question that I have on cost is actually a joke we have with founders that I interview because I interview a lot of very fast-growing company CEOs, and they're all private. I go, "How do you think about the cost trade-off frontier versus open, where maybe there's a little bit of a detriment in performance?" And they go, "Pfft, you kidding me, dude? You've seen how much money I've raised." And they're always like, "Can you take that bit out?"

But I give a shit about cost. No, you do. I mean, you're public—

Sure.

How do you think about that without the luxury of saying, "I don't give a shit; I'm private, and I've raised a bunch of money"?

Amanda, I love all these private companies that have raised a bunch of money and don't have to care about cost because they're not building the discipline to care about cost. Some of them do. But over time, discipline's really important. Anything you do in life, it's about discipline at some level, right?

But what if your product is 2% to 3% off theirs because they have the ill discipline to always do frontier?

I guess that depends on whether you continue to fund them and believe they'll be funded forever. Seems unlikely, given the history of technology, but maybe.

I think cost is really important, but it's not just because of that. I think if you look at a lot of examples out there, it's very easy to use a very expensive model to get pretty good results pretty fast.

I think using a cheaper model or an older model and getting good at applying that to problem sets is a real advantage for most technologies, prices, and costs. I think the customers we spoke to—large enterprises around the world—are incredibly sensitive to this type of stuff nowadays. They really care about price performance, and they are increasingly doing so.

We have a whole raft of products now that will show you your token spend across the board, in terms of your talent, in DX, and in your engineering organization. You can understand where you use which model, where you're getting value, which features you're using it for, and where you should change to optimize your spending, because this is a bigger and bigger budget item. They see it's only going to be a bigger budget item, so they have to care about the spending.

I don't think that's any different for them buying a generic model or a foundational model from any of the big 3, 4, or 5 as it is buying a product where their costs are going to be baked into whatever product they're buying is. So I think that cost discipline is going to arrive continually more and more, and I think it's a strength of ours, for example, in how we're able to use our FinOps strength to continue to reduce the price of features that are used heavily. I think that is a core strength and attribute of a great technology company: its ability to manage technological cost.

And that would be rooted in intelligent model routing on a per-task basis to optimize cost?

Model routing, model blending, applying the right model at the right time, using the right harness that again picks and adjusts itself—it's a really verdant area of exploration right now. The more you can get good at it, the more you can apply AI to smaller functional areas.

If you look at the so-called decision models, Jev and the 15 clones that have emerged over the last 3 weeks—which is insane, amazing, and scary all at the same time—they are often used for tasks that previously would have been too expensive to use even a GPT-3 or GPT-4-level model for. That's great, because we're now applying probabilistic thinking to more and more feature sets. I think we're all just going to see better products as a result.

How do you think about the model-routing layer? Is this an internal capability that every company will just build out and commoditize? Is there particular prowess in it? We obviously saw OpenRouter, as the dear Americans call it, bought for a lot of money. Maybe it's brilliantly strategic; maybe it's not. I don't actually know. How do you think about it if it is a core FinOps strength, as you mentioned?

I think for companies of our scale, you have to be very good at doing that yourself. There is a constant challenge with scale. I think any large technology company is probably going to do that internally, and that may involve using OpenRouter plus many other things, plus your internal router and everything else. These things can be continually layered upwards.

For smaller businesses, or businesses where technology isn't their expertise, those are just very practical solutions. There are many, many model routers and gateways with lots of claims about what they'll do. I think generically it's very hard to just route automatically, which is what they're showing. You need to give enough steer or intent, and I think harnesses will also be a part of this.

You're sort of talking about generic LLM calls. The more you're using harnesses, the more they're doing their own routing. The more you're using SaaS, or you come to the Atlassian platform and use Rovo or any of our AI features, you will get model routing happening for you automatically. It's our job to blend those costs and give them back to the customer for the best-quality answer, at the cheapest possible price, and at the best speed. This triangle trade-off is what we're always looking at in any given feature. That's part of our USP to the customer.

And you won't use third parties; that's internal routing?

That is all internal routing, yes.

Mm.

Yeah, we have an entire AI gateway team. It's not a small team that's doing that, but we have to do it, firstly, at massive scale, and secondly, we want to continue to reduce that cost where it makes sense.

I say to product teams all the time, "Don't over-optimize for cost in the short term. Build amazing, magical features. If they're taken up by customers, and what we thought was magical the customer thinks is magical, and it scales, now let's start worrying about the cost." Think a little bit in advance: can we possibly get this cost down? But worry about that later. If 5 people are using a feature, I don't care how much it costs. If 5 million people use a feature, I'm going to care a lot about cost. In the middle, I usually have a fair bit of time to keep optimizing.

What costs a lot today that you don't think will cost a lot in 5 years' time?

Model inference. Maybe a slightly spicy take is that the amount of energy, startups, capital, time, smarts, and brains—people who would otherwise have gone to work for McKinsey or NASA or something else—working on some version of the price-per-watt performance of inference is crazy. I suspect a lot of those things take years to come through, from new chips to better data-center design to ways of distilling and getting things out of foundational models.

There's a lot of it. The overall progress means I can't believe we won't bring the price-performance per unit of an LLM—we don't have a good term for this—down at a precipitous rate, just because the volume of capital and the volume of brainpower and talent being applied is so large.

Is anyone arguing that that's the case, though? I think they're just arguing that Jevons paradox means that we will use so much more inference. I have Instinct running about 50 different searches continuously in real time, every second of the day. We just use more.

We will, but it'll still mean the cost comes down. Where the cost is near-free, yes, it will explode. If something's free, you use an unlimited amount of it, right? You don't care; you throw it away. If there's not a waste component to it, which is ostensibly a cost, then yes, that will be the case.

I think there's always going to be an economic curve to that, though. Right now, the production of technology is faster than the consumption of technology. It is ahead. We've reached a tipping point: we can build features faster than customers can consume them in a realistic way that changes their business.

There has to be a point where that just doesn't stack up anymore. There isn't going to be infinite usage of all of this stuff. Your phone has amazing superpowers. It's a supercomputer in your pocket. Do you burn the CPU at 100% all the time? No, you don't—not even close.

So what happens to this inference layer of value and talent cohort, where you mentioned the verticalization of chips to different providers? What happens to it, Mike, in the next few years?

Look, I think it's just incredibly positive. I suspect for those who want to use those models to achieve things—whether you're coding, doing architecture, improving business processes, creating art, or whatever it is that you are doing—we should see that as a very positive thing. We are going to get higher-value technology. We are going to get better products and services for every consumer. It is going to be a creativity-driven economy.

One of the phrases I've liked the most recently—and I give Didier, the founder of Culture Amp, credit, because it's his phrase, I believe, or at least he's the first person who told me—is, "We're moving from a world of information asymmetry to imagination asymmetry." That's such a great way of thinking about it.

It's no longer that I know something someone else doesn't, or that I have particular business knowledge. That knowledge is becoming less about asymmetric value than about whether I can think about how to apply AI and these technologies to problems in unique ways. That's the creativity that we're all having to see in business now. If you're a first-principles thinker about business, it's awesome. It's the most dynamic and exciting period that we've ever lived through in technology, and maybe ever will.

I get why it's scary. I also think it's imagination. You're applying new things in fundamentally new ways.

If we live in this imagination asymmetry and you were to review the last 12 months, where do you think you have not shown enough imagination, where you needed to show more?

That's a good question. I think the biggest problem I have is time, which is maybe what all of us have as our problem. I don't mean that in some deep philosophical, life-and-death way.

I'm a venture capitalist, my friend. I sit on the sidelines and throw shade in when someone calls me. Time is not the problem.

And you seem to work fairly hard and fairly long days on a lot of different things, right?

Dude, that's because I'm a podcaster. Terrible decision as well.

You have 50 Instinct tasks running simultaneously. I haven't even had a chance to install Manus yet. I've heard about it. I know what it is conceptually.

So I think trying to carve out enough time to think is what I'm really trying to do at the moment. That deep, long thinking. There's something about the time-slicing of the world. People talk about the move from movies to TV shows to YouTube to TikTok. It's like books to magazine articles, to newspapers, to tweets or whatever.

Everything's getting sharded so much more, and in the agentic world you can get into this pattern: push the agent, do the thing, push the agent, do the thing, push the agent. Sitting back and thinking for a long time while still absorbing these changes and trying to look at them is difficult. We're living through a period of history where there's a lot of change on a day-to-day basis, but trying to see the year-to-year changes is much harder, and it requires deep thinking that I'm trying to create the space to do.

It's been a hectically busy year. I haven't had a lot of time to do that, but over the last 2 or 3 months, I've been actively trying to do that, even if it's only a couple of hours a week. It's very easy to be busy, and I think in AI, even easier, right? All of your agents and things running—your claws and instincts and muses, and whatever—are going to keep attacking your attention. How do you create time and space for real thinking? I think that's something I need to get better at.

Mike, just to give you context, my instinct is an automation to buy Laura Piana shoes for my girlfriend. This isn't some incredibly intelligent, sophisticated workflow, just to give you some boundaries. Maybe in Silicon Valley they're doing some things. Mine is just buying Brunello Cucinelli.

I think you'd be surprised. I think a lot of the agentic consumer use cases are basically shopping and travel.

Dude, if you're an aggregator or an advertising engine, good luck. A question for you: We see a lot of data marketplaces, whether it's Macaws or Surge's, paying for a huge amount of data from small and medium-sized companies. You talked about the context graph earlier. Data feeds that context graph. How do you think about the cost of data, the cost of acquiring data, and what this market looks like as we evolve?

6. The Context Graph Must Stay Fresh

Look, I think it's a little bit different. The corporate decisions and knowledge that underlie the teamwork graph are about us trying to help a company that has made decisions, projects, documents, and data, often. For example, how many times are PowerPoint slides the place where people put actual facts and figures? A lot. We're like, "Data is stored in databases." I'm like, "No, there's a lot of data stored in PowerPoints," and it never actually made it to a database.

How do we take all of that organizational knowledge, history, and decision-making, wrap it into a social graph of the people in the organization, and add code—the technological services created by a company—and mix all that together so we can give you great answers, learnings, and decisions? There's a lot of inference we can apply on top on behalf of the company, as well as making it continually self-learning, right? The teamwork graph's job is to infer ahead of your agent, before your agent has asked a question, the best possible answer to that.

Think about it like Google's index of the web. It doesn't go and crawl all of those websites to answer your question. It does that in advance. It has a lot of really smart indexing and inference technology to create an index so that it can answer questions very cheaply and at scale. That's our job, but tailored for every single company that has a wholly different set of information. That's very unique to each individual company.

I don't think they're going to sell that information. I'm not sure it's particularly useful. I know we saw Macaw or Google—I forget who actually ended up winning—buying the—

Spirit.

Spirit Airlines data. $30 million for 1 million emails and a bunch of other corporate data. I don't know if that will become a trend, but I think the value of that data inside a company is very much the same.

The important part is that it's alive. The social graph of a company is valid today, and only today. Tomorrow there will be different people in a company than there were today. Some people left, some people joined, and the social graph is going to continue to shift and change in terms of who works with whom on which projects and in which areas.

Our job is actually to be very fresh with that graph. It's to take history and infer learnings from it, but that has to be fresh and constantly built on top of. I say this all the time when it comes to our memory and our dreaming and how we understand the history of an organization. Knowing that I wrote Java code in 2003 and what code I wrote is totally useless to Atlassian in 2026. Knowing the fact that I used to be a programmer—a bad one—20-plus years ago is kind of useful because you may explain a concept to me in a way that a technology-savvy person would understand.

That's about as useful as it is, right? Today, I'm a full-time manager of people. I am a full-time storyteller. I am a part-time dabbler in design and experience in products, right? That's my job, and I love my job. But the history of it is very different.

If I looked at your history for the last 10 years, I'm sure stuff you did 10 years ago—some of the echoes and memories of that—are useful for you as learnings today, but the actual facts and figures of what you did aren't very useful. So an organizational graph has to be very fresh and current with facts, and very historical in its learnings and memories, to help the organization as best as possible.

Macaw is a little different. Those types of companies are looking for large-scale, generated, generic data sets to sell at scale to many other providers, right? To make those broad intelligence platforms valuable. I'm trying to think of a good analogy. Maybe if you think about self-driving cars, a self-driving car needs to understand all of the bits of a road, and it's a very specialized data set about roads: corners, signs, traffic, babies in prams, cyclists, all this sort of stuff. It has to be able to identify them and understand their dynamics and physics and these types of things. It needs to know that a cyclist isn't going to jump into the road, right? It has to understand all this. It doesn't have to know when Albert Einstein was born. It's totally useless to a self-driving car to know when Albert Einstein was born.

Mike, if you could acquire 1 company and its data would have the most transformational impact on your context graph, what company would that be?

I mean, acquiring companies for data isn't particularly important to us.

So there's no hole in your context graph, so to speak?

It's not so much a hole in the context graph; it's the customer's context. It's not our context. We are helping them put it together. We are helping them learn from the knowledge and work they've accrued in their communications. We're helping them connect meaning, connect the dots across that. We're helping their agents connect those dots. We're putting it into a giant graph. We're building a big lexical and semantic search engine on top of the graph, and then we have a query language that makes sense to agents through MCP and CLI, and we're helping them add their own custom data to that, right?

An investment firm cares about stocks. A manufacturing firm cares about supplies of materials. A healthcare firm may care about hospitals and beds and things like that. There is specific data, and then they all have meetings and documents and knowledge and code and these things. We're trying to put it all together in a way that they can retrieve it and the learnings from it, right? We're building an index for them. It's not the data that makes a difference to us per se.

Can I ask you a weird question that I don't know the answer to? As we see OpenAI intrude more and more into Docs and Microsoft Office, you're a student of the internet, Mike, in a brilliant way. How do you think about Anthropic and OpenAI's intrusion into traditional applications that you and Microsoft provide? And what does the voice in your head say when you think about that at night?

7. AI Enters The Application Layer

Look, that's an interesting question. Both of them are deep Atlassian partners, and we really value our partnership with them. I think they're probably partners with a whole lot of firms, right? Their business is, in some ways, selling computing services, if you want to think about it, right? They're also in the business of applications to some level. Where those lines emerge, I think, is going to be a highly fought-over variable in the future. I don't think anybody knows the answer to that question.

I will say, if it comes to things like Docs, we've faced this question with Confluence for a long time. Confluence is a business with north of $1.5 billion in ARR. That was our last disclosed number a couple of quarters ago. It would be a public company in and of itself. It is still growing well north of 25% per year and at a very large scale. So it would be a very good public company on its own, and ostensibly it's just content that you could put 15 other ways into your computer.

You probably have 20 ways on your computer right now to write down a piece of text, from ones that come with your operating system to ones that come with your workspace provider. In and of itself, writing content is not a definitive advantage. It comes back to design, experience, purpose, and integration into other workflows. So there are a lot of other things that go into that, and that's helped us build a very successful business there. I think that will continue to be the case.

Can I ask a really hard one? And you can tell me, "Harry, please—"

Sure.

"... I don't want to," and that's just rude. I love your focus on design, and I love your craft, and I feel it in your voice. Carrie in Linear always talks about the craft of design too, and I do feel it in theirs.

People seem to feel the craft of Linear’s design more than they do in Jira. How do you think about that, and is that unfair? And say if it’s unfair.

8. Atlassian Plays The Long Game

Yeah, I think it’s unfair. I think there are always going to be changes in those types of things. What’s important is never to rest. What has made Atlassian successful for 2 decades, and hopefully will for many decades to come, is to continue to invest in upleveling every part of our software experience.

The standard of what was good design 10 years ago is totally different from what is good design today, and it will be totally different in 10 years’ time. The ability to sustain that year-on-year for a long time is quite different. It’s a totally different way of thinking about continual progress and delivery.

Jira and the Teamwork Collection have had an amazing year of growth, accelerating and growing faster than they ever had at massive scale. That is in the number of users and attention, and a large part of that is because of the design experience. We’re showing off really foundational design changes, speed, performance, and efficacy in what we can achieve for customers.

In those areas, we just run our own race. I think that’s the best thing. Ultimately, it’s about customers’ experience of the quality of the software as a whole, and it’s a long race. I would say that to anybody. We’ve been here for a couple of decades, and we intend to be here for a couple of decades still.

I’m in my mid-40s. I don’t consider myself to be middle-aged, but I love what I do.

Do you just think you’re built differently, Mike? Honestly, my biggest problem right now—and the show’s successful because I’m very candid—is that I’m worried. I’m seeing a load of 20-year-old founders take the bid: “I don’t want to do this. I’m selling fast.” I’m 12 years in. I’ve got more energy than ever. You’re a little bit longer in the tooth than me.

24. 24.

Yeah. You just love the journey?

Yes. I think you have to enjoy what you do. I think it’s uncommon for everybody to manage their own energy supply.

Do you get lonely?

Do I get lonely? No.

No, but without Scott? I have a co-founder in Kieran. He’s not technically, but he kind of is. Do you miss it?

No. Yes. A bit of both. If you mean, am I lonely in terms of, “I have an awesome executive team and I work with 10,000 of the smartest people,” no. I always say that I get to work with an incredible number of very smart people on a daily basis, solving hard problems. If you designed a job for me, I’d be like, “That is what I want to do for the rest of my life.”

If you mean in terms of Scott, I see him pretty frequently. It’s not like he doesn’t live nearby. I’m in Amsterdam at the moment, and he travels a bit as well, but we hang out at least monthly, and we drink wine or beer and talk about Atlassian, software, and AI within the first 3 minutes of what’s going on.

It’s not like he’s uninterested in what we’re doing. It’s not like I don’t get near-daily text messages saying, “Why aren’t we doing this?” He’s very active for a retired person, which is great. You’ve got years of experience, but everyone has their own journey to run, right? I think that’s okay.

What I was trying to say beforehand is that you have to understand where your energy comes from. I think it’s one of the things I want to make sure I get across to my kids: figure out what gives you energy, try to do more of that, and you’ll be okay in life, just in a generalized world.

I think a lot of those people you talked about—and I get it in the AI era—the mental health challenges of being a founder have long been written about and have never been written about enough, because it’s a very stressful job. You feel like you have other people’s livelihoods in your hands, and less so at my time, but if you’re at that early stage—“I’ve raised money, I’ve got to get to the next raise, I’ve got to build this business, I’ve got to get to profits and revenue and customers”—your reputation and name are wrapped up in that. I see why this is incredibly taxing.

I think we should start by saying all founders are amazing people, that they’re taking on some insane challenge. They win, they lose, they fail, they succeed. They’re still taking on that same challenge.

I think the people you’re talking about who get into it for the wrong reasons find it very hard to sustain the energy. If you don’t enjoy creating products or services in whichever industry you’re choosing, I don’t think you’re going to be able to do it for very long. I think there are very few people for whom money or being in the right circles is purely motivating, and that’s going to sustain you for a 5- or 10-year period.

You’re probably doing this because you really enjoy it. Give me a microphone and a smart person to talk to, and I’m like, “Ah, this is great.”

Dude, I get to interview people like you every day, and it’s the most incredible learning experience.

Right.

I don’t care that I get paid.

And that’s awesome. So you’ve figured out what your source of energy is. It doesn’t have to be the only thing, but it’s one of the things that gives you that source of energy.

I saw you ranting about this on another one of these walks down the street the other day, and I think it’s totally true. I’ve been here long enough. It sounds like I’m a very wise old man, which is kind of scary. You see this with each of these iterations of the technology industry. There is a latest thing that attracts a lot of capital and a lot of people, and when that latest thing turns out to be like yesterday’s thing—and it still can be incredibly meaningful—the people who arrived to build mobile apps and mobile companies didn’t sustain. The ones that won and sustained are often directed in the same area.

No, my worry is more that it should be hard, and I completely agree with you on the challenges of founders. But we’ve made it so normal now for someone to come out, raise $15 million as a pre-seed, hire some people, and then, honestly, play the founder game and tap out the minute you can.

I think we’ve made it too easy in some respects, and that’s venture capitalists’ fault.

It’s a capitalist society, right? We can’t—there’s the market price for those types of things. I hope that people at some point realize that working with smart people on difficult problems for long periods of time is valuable.

The people I work with are a large part of the reason that I’m still doing this. The people that we get to create with are the source of energy, and I think that’s—It doesn’t sound like it’s in those people’s DNA to do that, and that’s going to be hard to sustain.

Mike, we’re going to do a quick fire because I could talk to you all day.

Sure.

What have you changed your mind on most in the last 12 months?

The likely future price of inference.

It goes way down.

I think so. Further than I thought.

What does no one say about having money that you think should be said?

That it is a wonderful thing and a big burden at the same time.

Why is it a big burden, honestly?

I think it’s a huge responsibility. I think it’s just something that’s temporarily held as a legacy, and you have a responsibility to do societally profitable things with it—not personally profitable things. I think you feel that burden. Lots of interesting things come with that.

How do I bring my kids up not to be assholes if they’re brought up rich?

Man, ask me in 20 years’ time. I hope I get that right. I think you probably try to be yourself.

I read a quote from, I think, a former Goldman Sachs CEO. He said you can tell a lot about people when they get promoted, as to whether they grow into the job or swell into it. I feel like I’ve been desperately growing every year I’ve been in this job. I definitively do not feel like I deserve this job, and I have to work crazy hard to deserve it next year, or to grow into it, or whatever term you want to use.

I think there’s probably some similar version of this with your kids. If you are curious, if you are humble, if you have gratitude, if you display all those properties, you probably give them the best chance to understand where that comes from. But I’m sure it’s a huge challenge. Tell me if you figure it out.

Actually, I got given some brilliant advice. It’s just like the National Geographic channel: watch the elephants. The baby elephants watch the adult elephants and copy them.

If you want your kids to be hardworking and kind, maybe be hardworking. Show them that you miss bath time because you were working hard on that Amsterdam trip to make money for the family, and take them to the coffee shop and be really kind to that barista and ask them how their day was. Let them see that.

Or, if you don’t want them to retire at 45, cheat on your wife and be horrible to the barista, and you’ll probably have assholes too.

Yes, I think so. If you’re asking whether they copy the way that you act and treat people, I think that’s definitely true.

That was the advice that I took. What’s the best advice that you’ve been given, Mike? You sit around some pretty interesting tables.

I think the phrase that I would say—certainly for the last handful of years—is, “Wage peace on yourself.” I think lots of things are really hard, and having a good attitude and knowing that you tried your best at something, whether it succeeds or doesn’t succeed, et cetera, try not to be too hard on yourself if you’ve done all the right things. I think for a lot of people, especially in the tech industry, the bar they set for themselves is incredibly high, and that can, in its own way, be a new sort of stress. So, touch grass and wage peace on yourself.

Final one. What do you not know that you will be capable of achieving that you do need to personally scale?

I’d probably say the same thing I said before: how to create moments of slow thinking, moments of slowness, in a world that is moving increasingly fast. I think that’s going to be a skill that we all need to develop. If we think the notifications on our phones, the pings from social media, and all these sorts of things are distractions, if you’re not able to handle that today, geez, I think the accelerated technological pace is going to be really hard. It’s a real world out there. Sometimes in the tech industry, everyone just lives in a screen like this. Ironically, I’m staring at a camera and a screen, right? All these bits and digits and things. There’s a real world out there of stuff that we’re creating, and we need to slow down to go fast. We’ve got to learn how to slow down and how to really do that deep thinking at a time when everything appears to be more fractionalized and speedy. It’s a real challenge.

Mike, do you know how I define a good person?

I do not.

Someone who’s willing to help with no immediate or clear benefit to themselves from that help being given. You’ve always been incredibly kind to me. Thank you. Seriously, I really appreciate it. You haven’t needed to, but I really appreciate the relationship.

Hey, no worries, man. Anytime. Let me know when I can help.