The SaaSpocalypse that wasn’t, with Atlassian CEO Mike Cannon-Brookes 分享

Today, I’m talking with Mike Cannon-Brookes, who is cofounder and CEO of Atlassian.

Atlassian is one of those companies that every other company runs on — it makes important platform tools like Jira and Trello that allow people to organize and manage big teams, create shared databases of company information, and generally allow work to happen. As you’ll hear Mike say, all of Atlassian’s products are actually different expressions of a single core platform, which really shapes how Atlassian itself is structured and how those products are built.

All of this means Atlassian is also right in the middle of the way AI is changing how all these companies work — AI tools might be able to look at all the different tools and systems you have and just read them for you, making a big migration to Atlassian’s platform less enticing. Or, if you buy the idea of the so-called SaaSpocalypse, AI might just build all of these tools for you, destroying this entire category of businesses.

Verge subscribers, don’t forget you get exclusive access to ad-free Decoder wherever you get your podcasts. Head here. Not a subscriber? You can sign up here.

Obviously, Mike had a lot of thoughts pushing back on this. So we spent some real time talking about the value of design, human users, how AI is actually increasing the usage of Atlassian’s tools, and what this all looks like a few years from now.

Of course, AI has also changed Atlassian. Like so many other tech companies, Atlassian did a round of layoffs earlier this year, with Mike saying he felt the company needed a different “mix of skills.” So I asked Mike what he thinks that mix of skills is and how AI is changing what it means to run a software company.

But Mike was also refreshingly direct about what AI can’t do, and he was willing to push back on his own friends like Cloudflare CEO Matthew Prince, who was just on the show talking about AI eliminating what he called “measurement roles.” This was a fun one — and shoutout to Mike, who recorded this from Atlassian home base in Australia, which meant he had a deep conversation about org charts at 5AM.

Okay: Atlassian CEO Mike Cannon-Brookes. Here we go.

Mike Cannon-Brookes, you’re the co-founder and CEO of Atlassian. Welcome to Decoder.

Thanks for having me, man.

I am really excited to talk to you. I believe you’re in Australia?

Yes.

It’s tomorrow for you. What’s going to happen tomorrow?

It’s tonight. It’s a great day here. It looks beautiful. It looks like it’s going to be an amazing day.

At this point in the AI news cycle, I feel like I can just demand to know what’s going to happen tomorrow, and everyone understands the urgency behind that question. There’s a lot going on with Atlassian. There’s a lot going on with your products. There’s a lot going on in the very concept of SaaS businesses and business processes generally.

Let’s start at the very start. I think people know Atlassian. They know Jira. I started my career as a Trello person. Tell people what you think Atlassian is today.

Atlassian is a platform that enables businesses to collaborate and manage work across their teams. We connect their business teams and their technology teams to handle the most challenging work problems in a singular platform across any organization that is technologically driven.

For any organization that has software and technology as its core competitive advantage, we make a broad platform that allows them to collaborate on content, manage projects, and unleash the knowledge of their teams and organizations across their business, their strategy, operations, all the way through to their service teams in all areas of their business. It’s a very large platform now that we provide.

Just in terms of what work is today and what it might be in the future as we add more and more AI to these enterprises, maybe this is too reductive, but I have always thought of Atlassian as a company that makes tools to help teams figure out what they’re going to do tomorrow. Jira is a perfect example of this. You don’t do the software engineering in Jira, but Jira helps you organize large teams of software engineers and file tickets and prioritize problems and tasks.

I was a Trello person a long time ago. My goal in life now is to never use any enterprise software. I feel like that’s a sign of true success if you’re just an iPad person in first class. I’m working on it. I’m not saying I’m there yet, but that’s one of my goals.

But there was a time when I was the managing editor of The Verge and my job, straightforwardly, was to look at Trello every day and make sure everyone was doing the right thing at the right time. And I’ve always thought of Atlassian as that class of products. We’re going to organize the processes of the company.

There’s something about AI that’s changing that. Has that conception of Atlassian changed for you? Is it changing faster because of AI?

Understanding what work you have to do at an individual level like Trello, at a team level like Jira, or at an organizational level like our Strategy Collection for large-scale strategy and operations — that is definitely a big part of what we do.

Understanding where processes are at — It’s less about work to do than processes. If you think about a business as a system of processes that are put together, everything in a business is some sort of system. It’s a process in the collection of all these, and how well we execute them is what your business is doing. The fundamentals of that haven’t changed. The way those processes run, the number of them running, and how to figure out what’s going on hasn’t changed.

We think about Jira as a human reference to work. When we say that, it’s a really interesting term because, as you said, the work is not done in Jira. Developers don’t live in Jira. Marketers don’t live in Jira. Finance people don’t live in Jira. Jira is a workflow engine with highly collaborative parts that enables teams to understand what they, their colleagues, and their broader organization are doing. That’s why agents and AI and other things are going to increase the speed of those processes. In a lot of ways, they will increase the reliability, the quality, and the consistency of output of certain processes.

They also put more emphasis on where the humans, the judgment, the intuition, the human bits that are unique, and the initiation come from. It’s our job to make sure that that’s all still understandable so someone can just look at any level of their business and know what the fuck’s going on. What are people doing? Are we successfully doing whatever it is that we’re trying to do, whether that’s service function in a finance sales deal, exception service desk to work out how many exceptions did we give out in sales this quarter, or it’s an engineering team who’s trying to work out, “There’s stuff happening all over the place. What did we build this month? Or what are we going to build next month?” All of those things are incredibly hard to do at scale and volume in a business.

One of the more important features of the tools you’ve built is legibility. It’s user experience. You have a big database of exceptions given out by salespeople, and you need to look at a dashboard and see how the company’s doing. Or you have a whole bunch of tasks, and Jira will let you see at various different levels of abstraction how many tasks are being completed at any given time.

That was something that was very important for human managers. A lot of the capability unlock of businesses as they added software like the software Atlassian makes is to just be able to see and coordinate more things at a higher scale.

The turn that’s coming is that maybe those databases aren’t going to be looked at by human beings anymore, or they’re going to be fed into more synthesized databases, or AI is going to make some other kind of information at different levels of abstraction. I know you’ve got a product called Rovo which feels like the new face of Atlassian in some ways.

What’s the path here? What does the end state look like?

There’s no doubt AI is going to help with a lot of those processes. There’s no doubt it’s going to pick up some steps in a process. Most businesses at the moment have a process, and they’re using AI to automate 80% of this step. It may not be 80% of the distance. It can be the width. If you think about that sales deal exception process that you just mentioned again there, if AI can take 80% of the exception — say this customer wants 45 days, not 30 days — and we can write a relatively readable document, AI is pretty good at reading the document and saying, “Yep, this one’s fine.”

But there’s going to be a customer that needs a lot more complex exceptions. They say, “I want to pay in 90 days in this way, and I want this thing and that thing.” That needs to go to a human for intuition and judgment. So 80% of those may be easier done with AI now because it’s very capable at processing our rules in this particular part of our business, and that’s great. It’ll increase the speed of the process, the quality of it.

It doesn’t take away the need to know what’s going on. Someone in the business will still need to know, “How many exceptions did we give?” What happens in most of these areas is businesses find way more ways to make themselves scale and efficient, and they find way more ways to make higher-quality products and services with AI. I see this all the time where they are utilizing AI in ways that people don’t expect. It’s not all about mechanistic efficiency. There’s a huge amount of quality that is able to be measured and understood that was never before. The deal exception process is a good example because it’s actually a very human process. It’s not usually rule-driven. The rule-driven parts are easy. It’s the non-rule-driven parts that are more complicated and require judgment.

The business still needs to know, for that process, “Do we need to add more people to this process? Do we need to change it? How much is going through it?” You still have a lot of work to understand the flow of work around an enterprise, which gets more complicated as the flow’s volume goes up a lot.

I’m asking because I’m headed towards a series of questions about the supposed SaaSpocalypse, and I’m very curious where you think tool providers like Atlassian fit next to the frontier models are just going to eat more and more capability. Eventually, Claude will just do this for you. You’ll say to Claude, “Run my business for me.” In the backend, Claude will burn a bunch of tokens, and it’ll just do it all for you, which is a promise that many, many people seem to believe in.

You’re describing a somewhat different path where the tools get smarter or more capable because of AI, and you’re still adding a lot of human judgment, but at no point is a frontier model just coming and running your entire business for you.

If a frontier model is capable of coming and running the entire business for you, if it’s truly running the whole thing, even assuming four or five years worth of increase, you have a relatively simple business. Most businesses aren’t simple. They are global conglomerations of rules, compliance, laws, staff in different places, human inconsistency — which is the same as saying human creativity — that they’re putting it all together to try to deliver products and services for their customers in their industry: healthcare, university, finance, automotive, or whatever the industry is.

They’re trying to compete with some other business. They are still going to have a huge number of humans, I would argue more knowledge workers, more developers, in five to 10 years time than today because the ability to do tasks to compete will go up.

The bar for competition will go up in almost all these industries. You will still have a huge amount of things to go do. Most of the things that we think are simple will get done for you. That is the history of technology. I’m a big fan of saying AI is just technology. What we can do with spreadsheets, what we can do with the internet, what we can do with ordering goods online… You think, “Once Amazon arrives, we’ll all just never have to go to the shops again.” It doesn’t happen.

I heard someone talking about bank branches as an analogy, which I thought was a really good one. Around 2000 when mobile banking came along, everyone thought bank branches were dead. Bank branches started closing down. Everyone wasn’t going to go to their branch anymore. It was a pretty negative place. At least it was in Australia. I’m pretty sure it was in America as well.

If you look over the last 10 years, bank branches are increasing. People are like, “Huh, this is a narrative violation. Why is this?” And the answer is bank branches did close down. People did go to internet banking. They did use their mobile app. They don’t go into the branch anymore with a passport to say, “I’d like to send $100 to Nilay, please. Here are his details,” and write them all down.

But bank branches have adapted. The reason they’re growing is there is a huge customer service element to a bank branch. The services they provide today are totally different than what they provided 25 years ago due to the technology of the bank and the availability of things. They are much more about helping you with higher-level processes and services than they ever have been.

That turns out to be, presumably, profitable for the banks, which is why they’re opening new branches. Bank branches are going up. The things those branches are doing are totally different to what they were 25 years ago. I think you’ll see the same thing with knowledge workers and everything else, that it will still be incredibly important.

I don’t see that part changing.

I am working my way up to getting Jamie Dimon on the show to find out what happens in all the Chase Bank branches in New York, because I don’t actually know, but we’ll set that aside. Jamie will come on the show. We’ll ask, and we’ll figure it out.

One of the reasons I’m asking all these foundational existential questions is to understand how you feel about Atlassian’s relationship to businesses and how businesses might change as we add new technology. I agree with your general framing that automation technology arriving to a business is a pretty familiar phenomenon, but there’s something else going on with AI that is either making that faster or more dramatic. It’s obviously happening to your business as well. You’re the CEO of Atlassian, and you’ve made some changes around AI.

I want to ask you the Decoder questions now. How many people is Atlassian today? And how is it structured?

We are, I think publicly, 12,000-13,000-odd people around the world. We are structured globally, I would say. We are “Team Anywhere.” Employees have the choice to come into an office or not. About 60 percent of people come in three days a week or more. About 25 percent of people don’t come in a single day per week. We run as a globally distributed company as we have done since the start. When you start in Australia and San Francisco, you kind of get used to the Pacific Ocean being a little bit of a distribution. This was before Zoom. We used to have these giant Polycom systems in meeting rooms. We don’t have those anymore. Technological progress is nice.

We have customers all around the world. We have staff all around the world. We are structured functionally, I guess you would say, with a lot of matrices internally. I don’t know if that’s the answer you’re looking for.

No, that’s absolutely the answer I’m looking for. What do you mean by “functionally with a lot of matrices”?

We have a CRO. We have a CFO. We have two CTOs at the moment. We have two large product groups, each of which has a CTO and a chief product officer, one in the enterprise and emerging side and one in the future of teamwork side.

Our products are organized into collections. We have collections of apps that are all built on a single platform. Well over half of our R&D is on the platform. The apps are increasingly a smaller amount of the proportional total investment in building. More and more is on the singular Atlassian platform. Customers hire us as a platform across their business to get work done, not for a single application.

That requires a matrix internally. We don’t have salespeople per product, if you want to think about it that way. The sales, customer success, and the FDEs are all customer-related motions. They’re organized as per the customers and their geographies and their size and scale. Then we obviously have to have designers and product managers on particular products who care about the purpose of a product, so you end up instantly with the matrix between how the customers are organized and how the products are organized.

We have a giant platform. We have the technical platform that runs across all the products. It’s a very large platform team. We then have finance and talent and other things that are also broad processes. Naturally, you end up with collaborative areas that you have to get together, and we manage that through our operating processes.

Do you use the Atlassian platform to run Atlassian?

We do. We run entirely on the Atlassian platform. I used to say we’re the number one user — we’re not anymore — of our strategy and operations products to understand what our business is focusing on, what our large-scale goals are, how the people, the investments, and the technical systems all come together, and what is going right and wrong across all of those. We are a big believer in our Strategy Collection to operate a large-scale business.

One of my favorite questions to ask enterprise software CEOs is how much they personally use the enterprise software. How much do you personally use your products?

An awful lot.

What’s your number one feature request for your teams?

My number one feature request is usually continual UI congruence. I guess that’s the category that people would put it in. I’m a bit of a stickler for design and consistency. We want to achieve eventual consistency in design, in which we’ve come a huge distance in the last five years. In the last two years, we’ve been truly world-class at doing it. That’s usually my biggest area where I use a lot of our products on a regular basis. As such, I’m moving around the platform, and I’m looking for ways where this thing over here happens this way. They do it better over there. We’re trying to achieve that eventual consistency across a large product surface area.

I use Dia, our web browser. I think 96 to 97 percent of Atlassian now uses Dia on a regular basis. I mean, that’s a daily basis, so I literally spend many, many, many hours of the day on it. We are building a web browser for ourselves because we’ve reached the point that we feel, “This thing should just be better for knowledge workers.” We’ve already built, I would argue, the best browser for knowledge workers in the world, and it’s getting much, much, much better. Keep watching this space.

So, yes, I would say there’s no 24-hour period that goes by where I don’t use one of our products. I doubt there’s a six-hour period that goes by where I don’t use one of our products at the moment.

I want to talk about Dia in a second, because I’m very curious about that acquisition. Obviously, we covered The Browser Company very deeply at The Verge. That was a big acquisition for you guys.

Setting that aside for just one second, one of the things I’m particularly curious about here is the distinction between the platform and the products that run on the platform. Again, maybe this is just reductive, but it sounds like there’s a big layer of capability, and the products are expressions of those capabilities in different ways, but they’re all running on the same core platform.

Is that how you think about it?

Yes, absolutely. Yes. We have to build a singular R&D platform, whether that’s how we talk to AI gateways, whether that’s how chat works, whether that’s how automation and identity and logging and governance and compliance and content classification work, whether that’s how our home and our search engine work. There are many, many more technical capabilities that have to be shared.

Then you have all of the next layer up, and customers want that to work the same. It doesn’t matter if I’m in Confluence or Jira or Loom or the Service Collection, automation shouldn’t be different per place in the world. That’s very expensive. It’s very hard. It takes a lot of time, but it delivers a huge amount of customer value and loyalty. Most vendors don’t try to do that. They literally don’t try.

Then you have a layer up, which is important, which is the consistency of operation and how content can be mixed. Operations and processes can be mixed across applications. If I open a Confluence page inside of Jira, I want that to feel like a Confluence page, but I want it to be inside of Jira. I don’t want to leave Jira to open it. I just want to read it, close it, keep going. There’s a lot of UI layer, a sort of recomposability, if you want to think about it that way.

Lastly, each application has to do its job. If you blend enterprise software together, you end up in a world of hell. People have tried this many times. You want a tool to do a job. You want to pick up a screwdriver or a hammer or saw and feel like that thing’s going to do the thing you hired it to do. The interface is designed to help you do that task. That’s where there is a difference in the interface layer and the interaction layer, between those tools.

The combination of those is where a lot of our design challenges live. For example, our chat layer, Rovo, is amazing. I would argue it’s one of the best, if not very close to the best enterprise chat products in the world. Our customers continually say that. “How come you’re better than insert-famous-brand-here?” There are a lot of reasons why we perform better in evaluations and testing with Rovo. It’s not because we make a foundational LLM, I think, which is the flaw of thinking.

However, using Rovo inside a product, if you’re going to Rovo for the sake of Rovo, is awesome. You have a question to ask, you want to search across everything, fine, you go to a particular interface. If you’re inside of Loom or Jira or the Service Collection and you just want to open up chat, ask a question, you want it to understand the context you’re in, but you want that operation to be consistent across the platform. That’s what we’re striving to deliver to customers.

One of my constant tropes on the show is that if you describe to me your org chart, I can tell you 80% of your problems. The structure of the company leads to some kind of natural politics in some kind of way.

The tension between “we have products that are designed to do a job” and “we have a platform that contains the core capabilities” is pretty well known.

Mm-hmm.

Maybe your product teams are going to desire platform capabilities that don’t exist yet or they will desire different platform capabilities from one another. How do you break those ties? How do you make those calls?

We have a number of ways that we are different and we try to do that. First, I would say most importantly, we have a platform-centric CEO and co-founder who thinks in terms of platforms. Thinking in Systems is my favorite book, man. I have a copy on every desk.

You have to put the platform first collectively, and then you go solve a bunch of problems. That does not mean genericism wins, but it does mean when you make trade-offs, you have to spend the extra effort and time to make it platform centric, to make it work. And that is a forever task. It helps to have a CEO, if you think about the referee across all the functions, that’s platform centric. They know where I’m going to land and so we’re going to end up there.

Secondly, we have design as a very, very senior role. We have a massive investment in design. Our chief design officer reports to me, sits on our executive team. We now have three, four, five former chief design officers in the business from other very large businesses, some larger than us. We have a huge investment in design and experience. Especially in the AI era, that’s getting ever more important actually, much as Claude Design or something would tell you the opposite. I don’t think so. I actually think [when it comes to] design and the experience of software, if the cost of building goes down, the quality and differentiation is on the experience and the design, and that’s not something that’s easy to do. That requires a huge amount of taste and judgment.

Lastly, I think we aren’t afraid to put things together in unusual ways in the organization where we need to solve problems. The latest example of that probably is our internal IT and engineering function now reports to our chief people officer and I’ve combined those two roles, which is unusual. “Why is HR running IT? That doesn’t make any sense.”

It actually does, I would argue, at the moment because of the things we need to do to transform our business internally continually, which is that if I look at AI and becoming more AI native as a business, I end up with two piles of projects. A pile of projects that relate to talent, how I need to change and grow the people we have, get them using these technologies, get them thinking in certain ways. I have a huge talent problem. Hiring, growing, training, changing people is very difficult to be competitive.

I have a huge system problem. I need to change our internal systems to have MCP servers. I need to make sure that everything is operable. I need to make sure that we have all of these system changes so that AI rolls out. How are we managing our token spending? There’s a huge amount of system change.

I ended up refereeing a bunch of, “Is this a talent problem or a system problem?” As I say, that doesn’t make any sense. Usually the answer is this one’s 60 percent talent, this one’s 40 percent system, and this one’s the opposite way around. I’ve put them all together inside the business so that our system changes to apply AI internally and our talent changes to apply AI internally are in the same function right now, which is a little odd.

The people who deliver the laptops and the people who manage compensation work for the same group, but it actually makes logical sense based on what we need to change about our business at the moment. What is this three-year epic that needs to change? We organize around the problems we need to solve and the goals we have as an organization when we need to.

This is a brilliant segue, because I wanted to ask about this executive you have. I believe the formal title of your HR person is now “chief people and AI enablement officer.” That’s quite a title. And the point you’re making about the systems of the business needing to change leading to some change in your talent mix is coming true. In March, you laid off about 10% of the company. I watched the video you made in Loom about that.

Your point there was you’re not necessarily replacing people with AI, but you had identified a change in the mix of skills that the company needed and you were going to make these changes proactively.

Walk me through this. What made you say, “AI is here sufficiently that the mix of skills is different”? And what changes did you make to your systems to make losing 10% of your people effective or worthwhile?

We have to start with the environment. The environment we’ve lived in in technology for a while is continual hiring and growth. The markets have changed in their view on that. Because one of the ways that we’ve managed changing businesses over time has been hiring people in different areas. If you have red dots and blue dots and you say, “We need more green dots,” you just say, “We’re going to hire all green dots for a while until we have a blend.” You manage it that way, that is no longer a feasible path, firstly.

Secondly, we have certain areas of the business, our AI roots and area-

Wait, actually, can I just ask one question about that? You’re kind of describing the COVID era over-hiring that every tech company was doing, and you’re saying that’s kind of over?

Yeah. People hire at the rate of the best decisions they know at the given time and what they’re encouraged to do by others. Hopefully they don’t look at everybody around them, but I know that a lot of CEOs do look at the people around them and say, “I should do that.” I always think those are poor choices.

Over-hiring, I don’t know. Pre-COVID, we were a few thousand people. You know what I mean? So we are four times, I don’t know, three times the size we were pre-COVID probably. Again, we’re accelerating as a business. We grew around 30% last quarter on a $7 billion run rate, which is the fastest we’ve grown in about two years. Customers are opting for more acceleration.

I’m just trying to understand the “you can’t just hire a bunch of green dots” comment a little more clearly because I’ve heard from many of your peers, “During that period we just hired everybody that we could.” It sounds like that’s not specifically what you were experiencing, but it rhymes in some way.

It does rhyme, but there are periods where that is the way to go. That’s not the current three- or four-year period of time. That’s the first thing I say because then you need to work at how we change our skill mix.

We have a huge amount of internal programs to try to do that. We also have certain areas, our AI products and services specifically, and our enterprise sales areas are growing very, very fast. They’re doing well and we want to invest further in those areas. When you have certain areas you need to invest further in, you run into very difficult choices where you say, “I need to invest further in these two areas. I can’t grow the overall in a significant way. I have to manage how to transition the organization between that.” Which leads to a lot of very difficult choices.

We are still hiring really hard in those areas in order to change the skill mix, but you also have a question of how fast we need to do that. It doesn’t, I always say, take away from any “internal training programs.” I say that in air quotes in terms of how AI is adopted and used. It’s more around learn, play, share, and mutual learning. It’s less like an L&D, in terms of “I’m going to sit down and watch a class on how to do this,” but that doesn’t take away from that. It’s a combination of all these factors as we’ve massively changed the business to be truly world leading at AI product development in the last two years. There’s no doubt that’s been a rapid journey and we’re trying to move as fast as we can to get there and to stay ahead.

One of the things I’ve heard from a lot of CEOs in your position is, “The value of our senior people has increased.” There are more senior software engineer job listings out in the market right now because as you’ve said, that judgment and that taste is very important and maybe the tools can just build that stuff for you.

I’m curious for your view on design. The idea that all of our designers will not be enabled to just ship code is very tempting to a lot of people. There’s a third view that says product managers roles do everything and all these roles are going to come crashing into each other.

What mix of skills specifically did you identify and how are you prioritizing in this sort of classic triumvirate as more and more people get the ability to do all the other jobs?

It’s very easy for people to have very binary takes on these things. I find it amusing that if you put an engineer, a designer, and a product manager in a room, they’d all sit there and they’d all say, “Oh man, it sucks to be you. Your job’s going to go away.” I say, “Ah, that probably is your answer right there.”

We have a lot more product managers and designers who are writing code from prototypes to actually shipping code than we ever have before, that is for sure. We have a lot more marketers and finance people and HR people who are doing the same thing, by the way. The code that they’re writing and shipping builds on two things. One, a platform that has the engineering robustness to be able to scale and deliver the customers a European-compliant data resident solution that is necessary, the legal constraints of country X, et cetera.

You have to build a technical environment that is able to get out to a customer with the required degree of customer safety. That is a non-trivial problem. That’s not something you’re just going to vibe code. If I’m vibe coding my own solution to something for my household, fine, it can just work well enough. I’ll go fix it if it doesn’t. If I’m doing that for a large organization, I need some other controls and rules around it.

Secondly, the specialization of those areas all doesn’t go away. The human judgment required in different areas is like the edges of the roles blending. They’re able to do a little bit more of someone else’s job. I have a lot of designers who are able to ship prototypes that are working code prototypes instead of a picture or a fake prototype in terms of a mock-up. That’s great. I still see them sitting in Figma a lot and doing that.

I have a lot of engineers who are sitting in Cursor or Claude Code or whatever on a daily basis, writing features and building things, and they’re able to do more design than they were before. That doesn’t replace the need for the other job. It increases the quality of the overall software. It increases the speed of collaboration between people if you want to think about it. If they each understand each other’s role, they can have a much more fulsome conversation more quickly. The specialized nature of what they do goes up. It becomes more spiky, which is really, really important.

The ability of building platforms — that’s going to get overused here — becomes more important. Just this week, for example, one of our designers wrote an interesting blog post about how we now have an MCP (Model Context Protocol) and a CLI (command line interface) for our design system internally. We have an award-winning design system, atlassian.design. It’s very public. We publish all of our design system.

I’ve always said in the role of design, 50% of their job is to make our design system and to keep it scalable, flexible, and buildable. 50% of their job is to solve the hardest 5% of problems in design because the half where they’re building the platform, what used to be templates and design documents that are now CLIs and MCPs to do design internally in the way that we want design to work so it’s consistent and coherent, that is a really hard problem to build that.

However, they’re unleashing thousands, if not almost tens of thousands of engineers to build better-designed things when they do that. Their leverage is a lot higher in building a design system than it is in solving, “Help me with this screen.” It’s the same in security. It’s the same in customer support and customer service actually. If you think of bug fixing as the platform on which you reduce customer service, more and more of that role is required on building those scalable templates, platforms, ways that the other functions can build on top of so they can do that thing.

If a product manager is vibe coding a feature, as in they’re just stabbing around some sort of code generating area and it’s an area where it’s not infrastructural, it doesn’t need to scale, it’s fine. It should leverage our entire design system. Turns out, out of the box, that’s not very workable. We’ve just shipped MCPs and CLIs internally, which are on V4 that are amazing at understanding our design system so the designers can evolve that system and the features that are being built will continue to match what customers expect in that congruence and consistency. That gets us huge leverage overall, but it’s a non-trivial thing. It doesn’t mean the designers are going to go away, the engineers are going to go away, the product managers are going to go away. I don’t think any of those roles disappear.

I’m always joking that Decoder is fundamentally a show about org charts, which means I think I know what the next five years of the show is going to be about because we’re on the cusp of some of the weirdest org charts in history in AI.

Matthew Prince from Cloudflare was just on the show. We talked a lot about his Wall Street Journal op-ed where he said, “I’m going to fire all the measurers at the company because I can replace them with AI. They can look at all of my systems and do all the auditing for me and we’re going to have fewer middle managers because what they’re really doing is measuring and auditing things and AI can do that.” There are reports that Mark Zuckerberg wants managers to have teams of 50 empowered by AI somewhere inside of Meta. This strikes me as not a great idea, but we’re going to see how it goes.

Where are you? Are you in classic org chart? Are you at wild ideas? How is this coming together for you?

I would say I know both of those people very well. I’m certainly more measured than that and I think the reality of how the world will be. I think it’s easy to point to extremities. Is someone somewhere a manager with 50 people? Sure. Are they effective? Maybe. People always talk about Jensen Huang. He’s obviously very effective. He’s a good dude. He runs things in a very weird way. Is that scalable and repeatable? I’m not sure it’s scalable and repeatable. I’m not sure if you took an entire set of large companies and made them all work that way, they wouldn’t have worse results as a collective.

Part of business is scalability and consistency, and being able to be understood and managed that doesn’t require uniquely talented individuals to understand a very odd structure as they move around a business. I suspect you are going to get more role blending and people able to handle multiple types of skills. I suspect they’re going to be more generalist. The edges of the role will blur into the roles around it. A product manager is able to do a bit more marketing and a bit more engineering.

That applies almost all across the way. Finance people are probably able to do a bit more legal than they were before. They’re able to understand the legal frameworks we have internally with agents and that’s great. I don’t think that changes the roles themselves. It may make teams a bit broader. I do agree with Zuck’s general direction that organizations will become a broader triangle, a fatter, wider triangle. The companies will get wider and hence less deep.

Broadly that’s a true direction. Do you go to 50 person teams? Then the span of control is like three people. The organizational depth is like three or something. If you multiply them by 50 at each level, if there are two and a half thousand people with two levels of reports, I would question that. That’s probably not an average. I don’t know about this term, “measurers.” I don’t know that anybody could sit around in a knowledge work job and say, “My role, my purpose is to measure things. I just walk around with a ruler and I’m like, ‘That’s a foot.'”

I know there are a lot of jobs where — I know what Matthew Prince is saying — that becomes the effective output of a role. We’re measuring it for a purpose, I would believe, and we are going to work out how to make that more scalable. My probable way of looking at it is a little bit different. There are two types of jobs. If you want to simplify an organization, there are supply-constraint and there are demand-constraint jobs inside a business. Input bound and output bound.

But what is the constraint on a job? Look at a marketing or a technology function, a creative function versus a legal or a customer service function. Legal and customer service are input bound. If I deploy a lot of AI, I don’t suddenly get a lot more contracts. The number of contracts coming in, if I’m looking at NDAs or leases or trying to understand the company’s legal position in certain areas, I can potentially process that work faster with AI. Through systems, I can be more efficient in how I do that. That is one set of roles that you can look at almost everything in a business process and say that’s input bound.

I’m going to manage that differently than things that are output bound. If I think about marketing or engineering, unless I run out of roadmap items — I’ve never heard any engineering team in my history of 25 years of doing this where they say, “Actually, we’ve reached the end of the roadmap and we’d like some more ideas of what to do next, please.” Said no one building software ever that has a large customer base. Their output, their creativity is bound by their own initiation and their ability to organize groups of people and processes and everything else, which is what the Atlassian platforms do. That is their limiting factor. It is not the human ability to create ideas. It may be that you have less input bound functions and areas as a proportion of your total employee base and you have more creative output bound functions.

How much is this human ingenuity and creativity and imagination? A possibility. That’s likely the way things shift and trend over time. That’s going to take much longer than people think. We’re big in technology. “Next year everything will be like this!” and it takes 10 years to happen, but it will happen. I think I’m very fond of this term. I keep meaning to put up a rant online about it. I’m stealing another friend’s quote here, but we used to talk about information asymmetry and how much certain people know things and that is their source of power or advantage.

I do think we’re moving to — he said it and it really hit me like a bolt — imagination asymmetry. We’re moving from information asymmetry to imagination asymmetry where your ability to create and think is going to be far more important as a competitive advantage for your team, your job, and your business than your access to information, your control of information, your understanding of information. That part is going to be cheap. That’s what LLMs are really good at. What they’re not good at is imagining things. They didn’t decide to start Atlassian. What business has been started by an LLM? Zero in the world ever.

Anthropic has run a failed vending machine for several years now. Come on.

But it requires a human to say, “Here’s what we’re going to do. Let’s fail at running a vending machine today.”

They were probably sitting in front of a vending machine cursing the thing when it didn’t drop their candy bar and they said, “I’m going to do this better with AI,” which is probably the way that most of those companies think about every single problem as they’re staring at a vending machine. It required a human being to say, “Let’s go do this.” It required them to write a bunch of stuff to go create the vending machine operation code and try it. We all learned a lot as a result of this scientific experimentation.

Let me ask you the other Decoder question, then I want to apply all of this to what is broadly happening to enterprise software and SaaS. The other Decoder question I ask everybody: How do you make decisions? What’s your framework for doing that?

Someone told me you were going to ask this question. I said, “What a broad, open-ended question.”

I would say I make decisions by talking a lot. That’s what people internally would truly say. I find the need to discuss things from different angles very important. I like to let other people talk. I like to get all the information together. I do believe there’s obviously a time where we need to stop and just move in a certain direction. The more important part for me, rather than how I make decisions, is I really like to explain decisions to people. At Atlassian we have a value: We’re an open company, no bullshit. I tell people all the time that does not mean transparency. We don’t owe you as an employee all of the details behind every decision. All the sausage making is not useful. It’s overwhelming to people. It doesn’t help us move fast.

We do owe employees clarity on why we are making a decision and an explanation of said decision. That’s one of the most important things managers can do for training, for helping their talent grow is not, “We’re going left,” but, “We thought about it. We looked at the different options. We considered straight, right, and left. Here are the pros and cons. We’re going left. This is the reason why we’re going left.” Often people don’t do that because people don’t like the decision. I say that all the time. When you talked about our talent mix changes in March, we had to make a lot of hard decisions to run this business. I want to make sure that we are always explaining those decisions to people in the clearest and simplest way we can, why we’ve made a decision and what the decision is without any dressing.

I really don’t like these corporate emails where I read the whole thing and think, “So what are you saying exactly? You’ve got a lot of words here.” I say, “Here’s what we’re doing. We need to change our skill mix. Here’s why. You should have the right amount of empathy to explain decisions, especially hard ones, but you also should not expect people to necessarily like it. We make a lot of hard decisions where I’m sure for 60% of the company, 7 percent of the company think it’s a great decision and 40 percent think it’s a terrible decision. We build trust as leaders by explaining our decisions and being clear about why we’ve made it and how we’ve made it. Hopefully we improve decision making. I always say decisions get to my desk when they’re presidential decisions. I think it was Obama that called it that, where it’s 51 to 49, that’s why it made it to my desk.

If it was 90% certain what we should do, it should not have made it to my desk in the first place. I only get hard decisions. That’s a well-operating company. Any leader should be in that situation. If it’s an obvious decision, why did it make it to you? That then requires more thought and more explanation about those decisions. That’s probably the most important part than any framework we use to make them.

Let me put all of this under some pressure. You’ve described Atlassian as a platform, and a lot of an organization’s decision-making and data and capabilities are in that platform. “What are we going to do next? How are we going to do it? Is it working?”

The entire point of the SaaSpocalypse was that we don’t need your beautiful user interface. We don’t need your design surfaces or congruence of design across products. All we need is your data. If I unleash Codex upon your data store, it won’t even matter that it’s the same platform. You can have 15 different vendors and the AI will create itself some middleware and read and synthesize all the data and that’ll be great. Actually your business can leap ahead in meaningful ways.

My favorite example of this from a previous era was I always had Daniel Dines from UiPath on the show. Instead of modernizing your healthcare billing system, he would just build robots that would use Windows 3.1 for you. This is a very early version of what I think every AI company is selling right now, which is, “Instead of modernizing your backend and spending all that money on a migration, our robots will just use your horrible old databases and present to you a new user interface.” That’s one version of SaaSpocalypse.

It is.

I see it coming for a lot of businesses. How do you expect to change all of Atlassian against that? Because a lot of what Atlassian sells is upgrade, upgrade the whole platform, use one product and use all the products, use the whole platform, modernize your entire business. The other answer is that it’s like $200 a month and you can just tokenmaxx your way into the future.

There are a lot of different parts of that answer. You’re talking about the headless world, I guess. I say that headless is brainless. It is literally the same thing. It is submission in slow motion to do such a thing, if you’re a software vendor. I say that very carefully because again, it’s one of these answers that requires nuance. We have one of the most-used MCP servers in the world. We’ve just shipped a giant upgrade to it, rewritten it from scratch. It is massively used on a daily basis by millions of people and it’s amazing. [Along with] MCP servers, we’ve shipped three or four CLIs that are also used incredibly heavily.

The ability to have inter-operation of tasks is very, very important for customers to get jobs done. I totally agree with that. However, people who use our MCP server and our CLI actually grow faster. They create more issues in Jira, they grow their seats faster. We’ve said this on public earnings calls, I think it’s like at least 5% faster. They grow their ARR at twice the rate of customers who don’t use our CLI and MCP servers. The customers using these things are growing their Atlassian spend faster than they have before. That world I totally believe in.

When I hear the word “headless,” though, I think that is the only world that we’re going to exist in. We just sell an MCP server. That is the brainless part about it because then your software really has no value. If you truly just have a database, fine. If your interface sucks, and I think a lot of enterprise software has horrible, horrible, horrible interfaces—

This is why I ask all the enterprise CEOs if they use their own products. It’s really asking, “How badly does your interface suck?”

I’m a big fan of recording the Loom and saying, “Guys, this doesn’t make any sense.” If you do that, that yak-shaving over and over again, you end up with amazing interfaces and that’s why tens of millions of people wake up to our software.

The investment in design and

(内容过长，已截断)