← The AI Daily Brief: Artificial Intelligence News and Analysis
The AI Daily Brief: Artificial Intelligence News and Analysis
How to Build an AI Native Team with Mike Cannon-Brookes
00:29:39
Loading summary
Transcript22 lines
- [00:00]
A
Today on the AI Daily Brief, a conversation with Atlassian's Mike Cannon Brooks about why context matters, how AI moves outside of the chat window, and what separates the enterprise AI leaders. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, welcome back to the AI Daily Brief. Today we are doing a bonus operators conversation in partnership with Atlassian around their Atlassian Team 26 event. Now, at this point, most of you are probably familiar with Atlassian or at least their software tools like Jira, Confluence, Trello, Loom and Rovo. Atlassian's products are used by more than 300,000 organizations. And for today's conversation, I'm joined by Atlassian's co founder and CEO Mike Cannon Brooks. In this conversation, Mike speaks from both sides of the AI transformation. Atlassian is a company adopting AI internally and Atlassian as a platform provider, building the AI infrastructure and product experiences that other enterprises use. And what I wanted to do in this conversation is get Mike's perspective, both as the leader of a company who is trying to adopt AI internally as well as a builder of tools and platforms who are helping productize and bring AI to the rest of the world. We discuss why context is becoming a core layer of enterprise AI, how agents, MCPs, CLIs, and headless tool use are changing the relationship between human software and digital teammates, the real factors constraining enterprise AI adoption, and why Mike thinks 2026 is the year AI starts moving beyond chat and into more natural product experiences. Now, since one of the big themes is the difference between teams using AI tools, or on the other end of the spectrum, actually deeply collaborating with AI, I put together a fun little companion quiz that you can find linked in the show notes, where you can answer a handful of questions and find out what kind of AI team you are. Are you a demo floor team who tries a lot but hasn't actually changed how you work? A team full of closet power users or an actual super team that treats agents as participants, not utilities? Like I said, there's a link to go do that in the show notes. But with all that out of the way, let's dive into my conversation with Mike Cannon Brooks. All right, Mike, welcome to the AI Daily Brief. How are you doing?
- [02:18]
B
I'm doing good, thank you. How you doing?
- [02:20]
A
Very, very well. I'm excited to have this conversation. I think one of the things that I find so fascinating it's been this way for a while, but especially in 2026, is this is a year where AI capabilities have gone up where the recognition of those capabilities have gone, up where a whole bunch of things have been thrown into a frenzy because of that. And what's always fascinating to me about people who are building companies in that space is that you're kind of dealing with this on multiple levels. You're dealing with it on the level of what are the implications for our customers that are interacting with us. But you're also thinking about it from the standpoint of how do we run a company that does well and uses these tools and integrates these new processes. And I actually wanted to start on that second part because I think it's sort of a such rich territory and it's always fascinating to hear how companies are adapting to how they build, how they work, to these new tools. So what are some of the key changes that you at Atlassian have been trying to, you know, implement when it comes to using this new set of tools, using new agent to capabilities, Just thinking about changing workflows more broadly.
- [03:30]
B
That's a very difficult question for all businesses around the world. Right. We talked briefly before this, you know, model acceleration is happening. Are you getting that intelligence to accelerate your business? It seems like when you apply one to the other, you don't get it natively. Right. We have a lot of reasons why that is the case. I think one of them is the talent changes you require are quite high. The business process re engineering you require is difficult. Right. And that has to be done to get the maximum output from that. So we take that on a lot of different ways. Look, we're looking at a lot of processes internally and where we can make them significantly more efficient or high quality or both. So the quality of an output is a measure of efficiency, if you want to think about it that way. Secondly, we have a lot of sharing programs. So this stuff is moving so fast. We talk about AI joy at work. You want people to use AI tools and technologies. Firstly, you have to make them available to your team, which is a non trivial exercise given enterprises and customer data and all sorts of security issues. So availability of tools, we've seen that with Diya, right? When we bought Deer, the browser company in New York, four people at Atlassian were allowed to use it out of 13,000. And they're like, cool. So we bought it, can we use it? And the answer was absolutely no, no one can use it. And they were like, what? Because the security of the AI in the browser is too much. Right now we have 13,000 people using it. Right? Because you have to put in all the right Enterprise controls to enable them to get these technologies. Then when you've got them deployed and rolled out, which is a non trivial challenge, and security teams shout out to all security teams who are doing the right thing to try to defend their organization against these potential new tools, then you have to have people actually use them, enjoy them, share them, fail with them, succeed with them, and then share those learnings. So we try to promote that loop at Atlassian, right? The, the learning about. There's a new capability, you know, in deer. How can I use it? You have to give them time and space and joy to play to try it. Can I remix this page in a certain way? Can I turn this into a, a slide presentation? Can I make a browse? Does this browser know these things? Does this tool know what I. You know? Then you have to share those learnings. If only one person learns it, you can't accelerate. What you need is. And we have lots of internal places to do that and blogs and encourage people to show loom videos but also encourage them to show failure where they say I tried to do this, I thought it would work and it didn't. Here's what I thought, maybe someone else can go, oh, I tried it again, but I did it this way and it did work for me. And that, that loop of skill sharing, we're all learning. It's not like we have people with 10 years worth of AI deployment experience inside the organization. They don't exist in the world. So we all have to learn simultaneously. Which makes it feel like this accelerated foot race for everybody. But it's also pretty exciting. But it comes down to that culture of, of sharing an openness, I think, and also sharing the learnings and the values so that we can, we can keep accelerating and we have plenty of wins along the way. But there's a lot in that chain to really get that I think effective inside organizations. And that's pretty similar to what we see inside our customers. I would say trying really hard to deploy at scale, but enterprise requirements, which are totally valid requirements, security requirements lining up against speed of technology progress, a
- [06:45]
A
whole bunch that's interesting to dig into from there, but maybe picking up on that piece specifically, I would love to hear your sense sort of from the front lines around where enterprises are right now, especially as this sort of big new agentic set of capabilities comes online and maybe more specifically how you guys try to on the one hand meet them where they are, while also to the extent that it's important to drag them into the future, you know, how are you helping them reconcile the difference or the space between those, as you put it, completely legitimate concerns with also the fact that the world could pass them by if they don't figure out how to adapt.
- [07:26]
B
You have to do all the things. There's no easy answer to that question. Look, a ROVO platform as a part of the overall Atlassian platform. The Rovo platform is where all our AI technologies and tools and libraries and the teamwork graph and MCP and everything agents live, right? The Rovo platform itself has to meet them where they are from an enterprise requirement point of view, adding data residency and private model choice and customer managed keys and all of the compliance requirements and everything there has to be there for them to be able to trust this platform. Right. And they have to have a whole set of controls. We've seen that in the, in the Deer browser. Again, we now have a lot of enterprise controls to be able to say, look, this URL set cannot be chatted to, this cannot be involved in memory. This URL set can. There's a whole lot of controls in a browser because you want to let them, you know, read the New York Times. And if they want to run AI in the New York Times, you're like totally fine. If it's your internal customer list, you may say, that's not fine, that's up to you. But you have to provide those controls and they have to learn about them. So I don't think you can not have the enterprise requirements across your AI platform, which is for all the vibe coders out there. It's still pretty hard to get it. And then you need the customer to trust that platform and to interrogate it. And some of our customers take six months of queries to turn on our AI platform. Some turn it on in a day. Either is fine, right? It depends on their very different customers. Often secondly, the customer workflows that currently exist. There are a lot of things we can do with AI. So we shipped a whole lot of agentic capabilities inside of jira, for example. So you can take any JIRA work item and assign it to an agent that can be a coding agent, a cursor or a claude code, that can be a business agent from Agent Force or some other agentic platform. You can bring those agents into JIRA in your existing workflow. So that's important because you don't need to change how that team is working. You're just accelerating what they currently do. And that's really important because it doesn't require as much people change as much process change in business reengineering, that doesn't mean you don't have to also be thinking about how, what's the different way to do this? Right, so you can use Rovo to file service requests in various ways now, and you don't need to think about, oh, well, I've got to create a ticket for that. I got to go and fill in a form. I just chat and it goes, hey, I've got like 90% of what you need. Let me do the last 10%. I'll ask you one more question. So you need to think about the new ways of working. At team 26, we shipped our teamwork graph CLI to take the context graph, which is, you know, we have the best context graph in the world in enterprise data sense, in enterprise knowledge sense, shipping that as a CLI and an MCP to bring that graph to wherever you're using those tools, whatever agent framework or desktop tool or whatever it is that you want to do that is necessary for modern ways of working. But we still have to take all those organizations and bring them along. And I think you've got this mix of all these things right, which is the beauty of all these different flavors have to happen simultaneously for an organization to really move from that sort of AI novice state to sort of AI native state. And what is AI native is a bar that's continually going up as well. So, you know, this is a, this is what's joyful and exciting about these technological transitions. Does that help?
- [10:36]
A
Absolutely. I mean, it's super interesting. This isn't necessarily the way that you said it, but it almost feels like when you're discussing this as a, you know, as a product builder, you're. You're basically trying to solve some set of their adoption challenges with your product so that they cease to be challenges.
- [10:54]
B
Right.
- [10:54]
A
If you build into the products that you're giving them, or especially if they're tools that they already use, platforms they already use, answers to some of the frequent challenges that come up around governance issues, compliance issues, privacy issues, data access issues. It sort of short circuits what they are inevitably going to find on their side and allows them to perhaps move to the more advanced stages of figuring out adoption rather than being stuck on those early kind of hard sticking points.
- [11:20]
B
Yes. Yeah, certainly. I mean, I think this stuff is all being developed very fast and shipped very fast, but the customer concerns are as old as time, as the phrase goes. Right. They need their software to be secure, they need to be compliant, they need to trust it with their data to some extent. We have trust as Atlassian as a, as a strategic partner. That doesn't mean our customers should blindly trust anything we ship, right? So they will investigate new capabilities we have and how it meets their internal concerns. And they all have new risk matrices to do with model choice and PII and all sorts of other things which are totally valid concerns. So it is certainly our job to build an enterprise strategic AI platform, which is what we've done in Rovo and continue to do every single week, right? As new capabilities. We talked about 5.5, we've rolled it out already, right? And so people will have to be like, oh, can we use that? How does it work? What is new in it? Our job is to take that intelligence, those model acceleration, and deliver them to our customers in value, both at the experience level in the user interface of JIRA or Confluence or dir, and to deliver that at the infrastructure level, the terminal level, right? The requirements and the scale and the performance that they need. And to do that and deliver that intelligence into their workflows, that is a common set of enterprise concerns. Right? And we have to do that no matter what they do require changing their business processes to get some value out of it. But they can get a lot of value out of it without changing their processes. And both of these are really important, I think, because as an example I can give. Confluence is a massive application. Millions and millions of people every day write content in Confluence to share ideas and communicate and connect to lots of different systems, figments, Google sheets, everything else, right? And the writing of those ideas hasn't changed. Some people want to write text. We can use AI to help them write that text in lots of ways better. We can use them to do all sorts of different things faster in the way they were already doing it in writing a document and sharing it. We've all seen text creation and rewrite this paragraph for me and help me with this sort of stuff. We can also. We have a feature set called Create with Rovo to help them make slide presentations out of a document. Now, we're not trying to compete with PowerPoint. It's got basic slide capabilities, but the ability to take a document and present it, to transmogrify the content is something totally new that didn't exist beforehand. Both of these are important capabilities to deliver inside of a product like Confluence, to just help people working the existing way to be faster and higher quality and better in their output, and also show them new ways of working, new ways of remixing the content they have, the knowledge they have and bringing in outside knowledge into that. At the same time, I think both of these things are incredibly important to do simultaneously. And sometimes the industry only focuses on the new way of working in the new thing. Our job is to bring those customers, bring their teams along, not ignore the new way of working, totally valid and important and exciting, but also take the existing teams and existing ways of working and helping them be faster. And doing both of those simultaneously is what Atlassian's always done really well. And I think we're doing very well through the AI transition about keeping that balance correct. It's a balancing act.
- [14:35]
A
And it feels to me like that theme of both helping them with the way they work now, as well as, you know, kind of skating to where the puck is going in terms of how they're going to work in the future. Is that sort of spread over the set of announcements that you guys just made? Because there's a whole slew of things that are about kind of, again, more, more sophisticated features in the things that they're already doing. But then, you know, you announced some new features around the teamwork graft, including the cli, including tools and mcp, that almost speak to some of the trends that, you know, we've been tracking on this show recently of thinking about headless tool use for, from agents and the combination of how people and digital employees are going to work together and collaborate. You know, I would love to hear a little bit more about some of those updates around the teamwork graph, CLI and tools at mcp, just to hear about how, how you guys have been thinking about those evolutions and, and maybe context as an important part of the enterprise AI process in general.
- [15:31]
B
Sure. Look, context is, is incredibly critical. Simply put, the way, the way that we think about it is acceleration for your business is about sort of intelligence multiplied by context. Right. It's. It's a simple sort of equation. The intelligence acceleration is very obvious, right? The, the new models coming out, new capabilities, new. Whether it's reasoning and planning and on through the, through the ages of models, context is the thing that we're trying to help organizations accelerate. We've had the teamwork graph for seven or eight years now. It is the best enterprise context graph out there at the moment. It's amazing in its capability and we're investing very heavily in continuing to make it better. So part of the announcements at the moment at team 26 are a whole lot of new capabilities for the teamwork graph itself in its connectivity. We've got a whole lot of new Connectors to other SaaS apps and SaaS data and partners of ours. We've also added whole new categories of knowledge to the teamwork graph. So three quick examples there. One is code itself. So we've had repositories and pull requests in the business objects around code for a long time. We've now added a full semantic index of your entire code base. That's incredibly important because we can help coding agents of all forms understand more about your business data and your code data together. And we can answer business questions across your code base which makes your coding agents faster. It gets better quality answers and it also makes them significantly cheaper to run because they don't have to do as many hops and as much thinking because we pre do a lot of the thinking that's in my graph bringing full semantic index of your your source code. Because we don't have billion token context windows and probably won't for a fair while, we can have a huge impact there in terms of connecting your technical and business teams together, which is what we've always done. Code that's a non trivial challenge at massive scale for some of our customers. We've been using it internally for a while. It's amazing in terms of the results. It can multiply with better context. Any coding agent you choose to use and we partner with all of the majors. Second thing we've added is we've greatly increased the the people capabilities of the teamwork graph. So we now understand your entire org chart. We now understand your skills of people. So we calculate by looking across your entire organizational context people's skills and continue to keep that updated and where they are growing. And if you want to think about who's AI native and who's not, we can tell you a lot about that. For example, across your org we obviously have your teams, your people, your talent. These two are really important to combine because you might want to ask questions about business content that sits in in Google Docs or Confluence and about code of only people who report to me, only people who are on my team. So I need to understand my org chart, I need to understand all of this and put it together and I want that answer fast, right? And I don't want to use burn millions of tokens to get that answer. I want to be able to do it quickly. That's why the people content is incredibly important. Who do I collaborate with? How often do I spend time with different people? We have all in the context graph connected to all your business content, your knowledge, your goals, your projects and all of your technical content. Lastly We've added physical assets. So about 50% of our customers operate some sort of large scale physical asset world. That's what a lot of businesses do. Whether that's TVs or trucks or satellites, whether that's laptops in businesses meeting rooms. We need to remember it's not just a purely digital world. When you bring that in, you see some fantastic examples about how you can give context to any AI tool rovo inside the Atlassian world. You know, cursor or CLAUDE code in a technical world, Agent Force or the Gemini platform, any of our partners in the business agent world, we can provide them through customers with amazing amount of context that's continually updated at the 150 billion object and connection scale across many of our largest organizations in the world. Now then, we are opening up that teamwork graph. So the last announcement is around that teamwork graph. Now we have a TWG CLI which is amazing in its ability to get there. It's built for agents, so 60, 70 different command sets, each of which has multiple commands to operate, navigate, query. That context graph incredibly fast built for agents and all of the technical capabilities. We're adding the simograph to our MCP server so all of those coding agents get access to all of that content in terms of org charts and everything else from a people side. So it's about having that graph be rich and continuing to grow. It grows with general usage of Atlassian applications and your SaaS applications in terms of process and we're obviously bringing that teamwork graph. We talked briefly about DIA as a web browser and things like we're bringing that into the browser, right? So your, your browsing activity and your security context of the browser is all been added to Deer. And at the same time access to the Timo graph via that MCP server allows you to take that content as long as the tabs you're looking at today and start, you know, getting answers across all of it with your model of choice kind of thing. So it's incredibly important, I would say that you have organizational context that is rich and that context graph is a massive investment of ours and hugely impactful to customers. And those customers that are in all the early access programs are seeing amazing results.
- [20:45]
A
One of the things that I wanted to ask this for builds on that is, you know, you have this frontline seat to a huge, I would imagine, array of organizations specifically in terms of where they are on their AI journey, on their adoption journey. And one of the things that we've been kind of tracking this year is the in many cases increasing gap between the companies that are starting to break out and lead and the companies that are starting to lag behind a little bit. Are there common characteristics that you're seeing among those companies that you would put in the category of leaders.
- [21:18]
B
Right.
- [21:18]
A
Who are sort of using more of the capabilities that these tools offer, who are thinking in more advanced ways about what they can do, especially anything that isn't just they use the tools better but are kind of characteristics that other organizations might look into?
- [21:34]
B
That's a good question. I think you're certainly seeing organizations leaning in, I would say heavily and thoughtfully. If I think through the customers I know who are doing it the best, there is a heavy and thoughtful leaning in and heavy being like, we've got to push this stuff hard. We've got to try and find real world examples. They're not looking for the sort of 2% improvement, you know what I mean? They're looking for good 20, 30% improvements in certain areas and they're really pushing to get those, but they're doing it quite thoughtfully. So they're building platform constructs, they're really concerned. The best customers ask us how does Rovo connect to these 10 things? And I'm like, ah, you've thought about this problem. Because what they don't want is a thousand different AI frameworks and tools and technologies are all colliding with different security postures and content. And they've been there before. They don't want, you know, another explosion of tools, but they see the acceleration and they're trying to work out how all of this works. Right. Our partnership stance really helps us there. Right. We've long been a highly collaborative organization with many of your other tools. Coopetition, if you want to think about it that way. You know, we've been long term partners with both Microsoft and Google in the, in the sort of workspace or Office area teams and Slack Zoom in Teams, etc. Right. GitHub and GitLab and BitBucket all work together, so there's a lot of tools that you have to work with. I think they're asking for that collaboration to happen and we are trying to give to them because they're pushing hard on the results they need and they realize the results don't happen in a silo. I think you're seeing more focus on overall output and an increasing focus on cost understanding. Maybe not cost management yet, but cost understanding. And you know, DX is a good example. DX we acquired eight months ago now, something like that. Absolutely. Flying off the shelves like going gangbusters. Because the ability to understand your engineering organization got 5,000 people, 10,000 people in engineering and how AI native they're using. Who's consuming the most tokens. Which AI coding tools are working better and not better is a useful understanding. What's really important is how is that affecting my overall engineering productivity? Because some of these tools are very seductive. We've all sat in front of some sort of chat tool and it starts spitting out text and we're like, this is just the magical experience overcomes you. But is it useful? Did it help me do my job better? Is a totally different answer. Should an engineer be using more tokens or less tokens? That's kind of the wrong question. I think the best organizations are talking about throughput and flow and output and quality of output, not quantity of output. If someone's using 0 tokens, probably not doing a great job. The people using the most tokens, are they actually the most productive? These are the types of questions that I think they're asking and we're trying to help them answer some part of those questions, but they're non trivial questions. Right. We're all, I think trying to do that. But that's what the best organizations are really asking of themselves. What is the overall quality of output that I'm getting and is that accelerating rather than just raw tokens at a ROI basis?
- [24:51]
A
It almost feels like that. One of the things that I also noticed from your announcements was the general availability of the Robo Studio no code environment, which is, it feels to me like at least partially trying to deal with the exciting capability set of non engineers building things to solve their problems, while also kind of understanding that that creates a whole new set of challenges in terms of process, sprawl, product sprawl, all sorts of new challenge vectors. And it sounds like that this is sort of part and parcel of the. I don't know what I might characterize as the systems level thinking that that characterizes some of these leading companies.
- [25:28]
B
Yes, I certainly think so. I mean the. As I said, there'll be a lot more people creating technology doesn't necessarily mean they're all software developers. They're everything from Robo Studio to Deer are now writing code. Right. Rovo itself writes code in our new chat models. You know, uses tools. DIA can write code in your web browser and use tools. Rovo Studio can write code and use tools. All three of those probably don't commit the code to git. You probably don't get woken up at 2am in the morning when the code is down. So it's not sort of true engineering development. But it doesn't mean it can't be incredibly useful for you to solve a problem of financial modeling or just a little tool for yourself. All of that needs to be built securely. So between Robo Studio and the Forge extensibility platform, customers that already trust Forge and trust our teamwork graph, Rovo Studio can build small applications for you that leverage a teamwork graph all the way up to full fledged agents. Now, if you're using full fledged agents, proper large scale agentic collaboration, again, Rovo Studio, I think we've passed 5 million agents created at this point, something like that. Those agents are often written with code when they're large ones. Right. Proper development teams have to be versioned and be very thoughtful about how it's changing. You don't just like update the model all the time because it might break the usage of the agent. So you have to be very careful and thoughtful. That's written by, I would say true development teams. But that doesn't mean there's not a world for a lot of people using DIA or using Robo Studio to just make an application that solves their own problem. Right. And we have to help understand that spectrum, help customers understand that spectrum. And we're trying to do end to end there for the customer, but always with the teamwork graph and your Atlassian platform security levels and compliance levels underneath it. So you will not get access to things that you wouldn't in that application you built. That application, you know, in those personal application levels runs as yourself. So it gets access to only the content you already had. And we have all sorts of security controls that have to be built on that for enterprises. That's the nexus of what you talked about beforehand, which is trying to meet that customer where they are, but also knowing that those applications don't have to run 247 for a million people with scale and performance and all the other things. They're just meant to solve a problem. If code is cheap, that can be thrown away and you can write another one tomorrow. Which is kind of what you get to deer at the end. Which is exactly what it does is writes a lot of code you throw away in minutes.
- [27:53]
A
I could spend all day talking about disposable software and what's that going to mean. But as we wrap up here, I wanted to just close on a question about when you look across the rest of 2026, what is one thing that gets you excited about AI either for your team, for your customers or for yourself.
- [28:12]
B
One thing is very hard. If I had to pick one thing it was that we are going to start to see AI move beyond chat. Chat is incredibly important. We're going to start to see it. I believe we're already seeing it. But over 2026 more and more change in the experiences we have and the design experiences and bringing it to the every person that is very exciting. It does not require you to be a crazy smart prompt engineer with chat and a desktop terminal client open, bringing that into other areas. We see that across our platform from from Deer to Jira to to to Confluence to Loom. I think that's, that's the most exciting is is the the design era of AI bringing these technologies and powerful tools to everyday users in meeting them, where they work, how they work and changing the way they work at the same time. I think that's incredibly exciting.
- [29:12]
A
Mike, awesome to have you on the show. Really excited to get your take again from the front lines of both building a company using these tools as well as building the tools that companies use. And so thank you for spending some time today.
- [29:23]
B
No problem. Appreciate having me. Thanks Nickel.