Atlassian Founder Keynote: Your business knows better | Team '26 Europe

0:05 Please welcome Atlassian CEO and co-founder Mike Cannon-Brookes. Good day everybody. Welcome to Team Europe. Whether you're in the room here or online, thank you so much for being here. You are the builders and the operators who make modern business run. And recently, your jobs haven't gotten any easier. We are living through one of the most creative, dynamic eras in the history of technology, which is thrilling and yes, a little bit scarier at times, too. One thing hasn't changed, and that is that the world runs on teams. Your teams. Now, in the last few months, I've met with hundreds of customers across tens of countries. And talking with leaders just like you, the sentiment is equal parts optimism and uncertainty. The questions that I hear the most fall into three big groups. On models, which one's the best? Which ones do Atlassian use? On people and agents, are agents going to replace my teams? How do I get them to all work together? And in terms of progress, the acceleration at the moment is so fast. How do I actually know what on earth is going on at my organization? Now, I want to help answer these questions with Atlassian's philosophy and our vision for the future of teamwork. We believe it's less about which model you choose than it is about knowing the right context. We believe you don't need to replace people with agents. You need to know the best way to get them all to collaborate efficiently. And we believe that the new ways to provide confidence are there so that you know best how to move at speed. I want to unpack these starting with context. Now we love model progress at Atlassian. The curve of improvement is hugely exciting and the unlocks are very real. But we don't subscribe to the idea of a single winning model. For us it's going to be about all those continuous leaps forward week after week and harnessing the best models that are available.

2:12 Think of it like this, that models are the engine, but context is the fuel. No matter how capable a model gets, if it doesn't reliably know why you made a decision back in 2022, it can't help you win in 2026. The good news is, if you're an Atlassian customer, you already have the context for that decision and so much more running in the Teamwork Graph. It is the institutional memory of your company, a living web that understands all of your work and how it connects. Not just what happened, but who was involved, why they made the call, and what impact it had. The Teamwork Graph brings together six distinct types of context. First, the knowledge context. What your company knows in its document, spreadsheets, and presentations connected to the work context.

3:00 what your company is doing across projects and epics, initiatives and goals. The people context, who your company is, the communications context, everything that your company is saying. The code context, what your company is building, and the asset context, the physical things that your company owns. Now, why all six contexts? Because AI operating in a silo only delivers a fraction of its potential value. Real leverage happens when every type of context is interconnected. And it doesn't just capture what has happened. Importantly, the Teamwork Graph works ahead of you. It continuously indexes and makes connections, running inference before your agents even ask questions. This upfront heavy lifting makes it lightning fast and cheap to query. And it learns through self-reinforcing loops, building a deep understanding of how your company works.

3:55 So bring your cursor, your chat, your cord, or any other agent harness. We love and use them all. They all plug into the same context and they will all run faster, smarter, and a hell of a lot cheaper with the Teamwork Graph. So don't worry so much about model choice. Bet on compounding context. Next, we have to find a way to get your people and agents to collaborate. Helping teams work better together has been our mission for decades. We've been investing at scale to enable open multi- vendor multiplayer teamwork. In addition to the Teamwork Graph, we've built a world-class enterprise platform that's deeply connected to the wider technology ecosystem. And across the board, we're seeing that agents aren't replacing your people. They're joining your teams.

4:43 Agents are triaging service incidents, tracking company goals, documenting big decisions, and yes, shipping a ton of code. Great outcomes are coming as teams shift to new ways of working. But as things accelerate, we believe that experience, the experience of collaboration is going to be so important. This is why we're bringing together all the capabilities of our platform to help teams make that shift and continue to interact in ways that feel natural. That's why I'm excited to share that today Atlassian is introducing AMP, our agentic multiplayer protocol.

5:19 AMP is a collection of human- centric experiences, technologies, and patterns that positively shape how people and agents can collaborate on and off the Atlassian platform. Now, AMP is comprised of three interconnected layers. The real-time and asynchronous collaboration that introduces new UI patterns showing how agents can operate in real time, tracking the work that they've done, or highlighting when they need input. This means that people can work alongside them exactly how they want and when they want to. In identity and shared presence, no matter what combination of agents and people are in a team, we make sure that you know who is who. We do this by surfacing agent identity right across our experiences with clear attribution and easy to understand accountability.

6:09 and context and governance. Our agent governance extends on and off the Atlassian platform with the right controls and scoped permissions. This means people are always able to initiate work, deal with exceptions, and exercise their judgment and intuition where it matters. So you don't replace people with agents. You create a new and powerful system of collaboration for your organization because AMP is deeply integrated across the entire Atlassian platform. You'll see these experiences live in action throughout Team 26 here in Europe. And a third group of questions, how do you manage the pace of progress? Well, when every person and every agent starts moving faster and faster, new challenges arise because it's not just one person moving faster. It's a multiplication of teams increasing their outputs, resulting in dozens of parallel learning loops.

7:02 Now, without a way to connect it all, this velocity can turn from something incredibly exciting to something quite disorienting. But how do you calmly harness all of that momentum? Well, we think that confidence comes from seeing more than what's happening. It's having the insight to understand it all and the ability to act on it. It's knowing what loops happened when agents did what was asked and understanding how that change ripples outward throughout your organization.

7:34 Most of all, it's knowing how to handle the inevitable speed bumps. Confidence means catching those exceptions immediately and having ways to manage the unexpected. That's what creates a sense of calm as you move at these new speeds into the future. Now, for individuals, this means starting the day knowing what's ahead and being able to seamlessly transition between different tools to get work done. For teams, it means seeing a wider view of work, where all things are flowing well across related projects, as well as where bottlenecks are arising, so we can quickly take action to get back on track. And for organization, it means having the broadest understanding of how strategy and execution connect, giving leaders the information needed to provide clarity and manage risk. Of course, all of this requires a platform of enterprise trust. We're bringing governance, permissions, and capabilities like inference residency to ensure that AI in your business stays compliant, private, and as secure as the data it runs on. Tying it all together, the Atlassian platform can help you with all of these groups to handle whatever comes around the corner. By providing context, collaboration, and confidence together, they form your organization's central nervous system, rewiring how it is that your company operates. Now, this is my 17th year of coming to Amsterdam since we opened our office here. And if there's one thing around here you learn pretty fast, it's that when a bike hurdles at you at 30ks an hour, it's not really time to schedule a meeting or send off a database query. You just have to trust your instincts and react in milliseconds. Usually for me, jumping out of the road. In the AI era, your organization needs its own nervous system to react just as fast to all that is hurtling towards it. And the Atlassian platform is just that. It connects your teams, your agents, and the map of your business so you can move forward with a sense of confidence and calm. Okay, let's get into it. Some demos to show you what that nervous system looks like in practice.

9:46 Now, a core belief of AMP is that this new technology only matters if it feels completely natural and keeps you in your flow. So to show you what I mean, I want to start with Rovo as a tangible example. We've reimagined our Rovo desktop experience from the ground up with craft and speed front and center. So this is the new Rovo desktop app, a native app that brings together Rovo Search, chat, and the Teamwork Graph directly on alongside every other app that you use. We finesse the experience to feel fast and intuitive and smooth.

10:21 You can see it in all of the small details and micro interactions that help keep you in your flow. So, let me show you how I use Rovo nearly every day. We're going to start with search. Getting ready for a customer meeting. I'm going to start by checking on the status of our product road map so I'm ready for questions that may come my way. Now, Rovo's reasoning and searching across information from the entire Teamwork Graph.

10:49 I want to call out how fast that was. Those search results are now 40% faster than they were a year ago, despite the Teamwork Graph being much, much larger. And of course, I can see links from all of my apps here. Got Slack and Google Drive live side by side alongside Confluence Loom, so you can find exactly what it is you're looking for. Now, right at the top of the page, the first thing I see as we continue to improve our smart answers to help keep you in the flow.

11:14 And of course, I can easily scan all of these icons at the top and quickly filter to get things that I need. So, for example, I can use our new Zoom connector to filter to find meeting transcripts very, very quickly. Now, one of the customers that I'm meeting with here is Delivery Hero. They're a quick commerce company, super cool business, great European customer, and I'm going to need to send them a deck for our meeting. So, let me start by opening mail.

11:43 Email be this one. Great. So, I'm going to start by opening mail. The first thing I want to point out is some of the micro interactions we've baked in. So, let's use them. First, I'm going to press a very quick keyboard shortcut, which will bring up Rovo's compact mode. So, the Rovo desktop compact mode, which I can see right here. one keyboard shortcut instantly brings up my always on partner ready when I need it, anywhere on my desktop in any app.

12:08 Now, all of my recent files are here, so I can quickly scroll through them because the Teamwork Graph knows what I've been working on lately across all my apps. My deck is right here at the top. So, I can either just hit command C to quickly copy it, or if I want, I can just quickly drag and drop it straight into my email. No app switching, no link searching. That is your context traveling with you as you move through your work.

12:33 Now, another example of a different micro interaction here. Someone on the team has sent me an email that covers some of the details I need to know about our upcoming meeting with Delivery Hero, their expectations, the features we've delivered on, and some of the key promises that we're still working on together. Now, I want to use this to help me prep for the meeting. So, now I can just simply click and hold my mouse on any window and you will see that I can add that as context for my query. It's the lowest friction way to start a chat with your entire team at graph from the windows on your desktop. So I can ask something simple like help me prepare oops I look down prepare for this meeting.

13:15 Incredibly simple to build that muscle memory that streamlines how you work with the Teamwork Graph and rovo alongside all of your AI tools. So let's open this in the full screen mode. Now while that's running I want to point out a few things. We've worked on making it easier to get back to what matters most, making it an incredibly personal experience. So again, this chat here, I can now quickly pin it to have frequent searches and chats easy to get back to. I can see that I've got the agent status here of agents that need input, what I've chatted to recently, and all the contacts that they started with. And another new feature of Rovo today, I can now schedule chats. You can see here I've delegated some repetitive work, maybe daily morning to-dos or my weekly summary. Another feature of Rovo that we are launching. Now, I want to point out I asked a really generic prompt here. I just said help me prep for the meeting. I haven't given it any other context except for that one single click on the app. So, it'll go into Teamwork Graph and come back any second now. We have a lot of live demos for you today. So, we might need a little bit of a few seconds. Of course, you have all of the apps that you can filter by while it's coming there. any of the apps in your graph. There it is.

14:26 Excellent. so I can see that it's helped me prepare for the meeting. Again, I've got a visual summarize of the feedback themes here. It knew that I was meeting with delivery hero. So, it as it told me what they were expecting because it knew what was in my email and it picked up that I was meeting with delivery hero when I didn't even ask it that what it was we've delivered, what we're still working on, and key promises. And again across many many sources here from Google Drive, Confluence and Salesforce in this case everything that I needed immediately out of my email and that's the summary. So it's all there. Oh and lastly one thing I forgot I have other teammates who are joining me in the meeting. So another new feature of Rovo very quickly is I can now share that chat. So here I can say I want it only specific to to specific people. Say I want to share it with Shri for example. And you'll notice when I copy the link, it automatically updates the access to that chat, permissioned and sharable just like you used to with confluence pages and everything else to make sure everybody is on the same context. So that's the all new Rovo desktop. It's fast, it's beautiful, and it puts your organization's context right in the flow of the work you're doing on your desktop. But it is just one example of where we've extended that experience to all of our Rovo clients. Whether you're using Rovo on the web, Rovo in embedded mode, Rovo in the CLI, or a new Rovo mobile application, which is also coming out right now, and truly one of the best design things that we've ever made. You'll see some examples of that in just a second. So, everything I just showed you in that super quick Rovo demo is built on your organization's context. It's what connects your teams and their agents to the information they need to make effective decisions. to share more about what's new with context and the entire Teamwork Graph. Please welcome our chief product and AI officer tomorrow.

16:19 Welcome. Thank you, Mike. How you doing? I'm doing great. Oh, it's wonderful to see all of you here in Amsterdam. so Mike just talked about why context is so important. So let's take an example and let's say you have an engineer who's worked at your company for 10 years. They're going to be much more productive now than they were on their first day. Well, the same is true for AI. AI is going to be much more effective if it understands your company. So this is why we've continued to invest so much in the Teamwork Graph. The Teamwork Graph that we have is the most robust enterprise context in the world. Six months ago, we said the Teamwork Graph had 150 billion objects and relationships. And today, it's nearly doubled and has over 250 billion. That's pretty fast, isn't it? Very big number. Anything with billions in it is a big number. but it's not just the size of the graph that matters. It's the breath of understanding that the Teamwork Graph has from across your entire business. Earlier Mike talked about the six different context types. They extend beyond glassian to all of the tools and applications you use every day. And at your company you probably use dozens and dozens of different tools. So we bring them all together in the Teamwork Graph through connectors. We've added over 20 new connectors in the last six months including Zoom and Gong. And we've made over 65 enhancements, new capabilities to these connectors so that the Teamwork Graph can understand your work better, including things like custom fields in Salesforce, issues in in GitHub and meeting transcripts in teams. And we've also made it easier for employees to add a connector right where they are. So you can add a connector now from chat or from your CLI, however you're using the team of graph. And for the admins in the room, we've made it easier to manage your connectors with deep enterprise controls in the admin app.

18:24 And we're introducing a new context type data. Data context brings live metadata from your structured business data and from over 20 sources. Things like data bricks, snowflake, tableau, bigquery, and all puts it all into the Teamwork Graph. So for example, you're writing a project brief in Confluence and you need customer usage data. Instead of going to a separate database, you can just at Rovo and Rovo will query the live data to create an interactive visual chart and embeds it into the page. This helps you and your agents work with data alongside the project's goals and decisions around it. Now if the data context tells you what's happening with your business as Tamar just showed the code context helps you understand all the technology that your teams are building.

19:20 Now previously the Teamwork Graph understood repositories branches and pull requests as part of that code context. But in huge step today the introduction of source code is now generally available. full semantic understanding of all of your functions and symbols and classes. Helping humans and agents move from diagnosing problems to finding the right code faster. Let's make it real with an example. Every engineer I know is pretty passionate about which is dark mode. So here I'm going to start with this jur work item and Rovo is always here where I can ask in natural language in the embedded mode for help finding screens which won't work in dark mode.

19:59 So I'm going to move here from the embedded mode to full screen so we can see the thinking traces a little more clearly and you'll see that the Teamwork Graph will kick into action immediately and go off and start querying all of my code from across multiple repositories. Here I can see GitHub and Bitbucket. Now let's talk for a second about why this is a really impressive piece of technology and very necessary. Like many of you, our codebase is now billions of lines of code, thousands of repositories, more than two decades of commit history, and it's growing really, really fast today. Not everything is called dark mode because we use different programming languages and different standards across the history of all of those apps. So, oh, I was going to say while it was searching, but it's come back very, very fast, which is excellent. code search has looked across all of my connected repositories. So I can see here I've got the sources from Bitbucket both cloud and data center GitHub both cloud and data center and next month coming up GitLab. It's also cross referenced it with other contexts from across a Teamwork Graph. Here I can see conference and Google drive but you can have support tickets, customer feedback, internal road maps, comments from sales calls. Multiple contexts coming together into a single query is what the Teamwork Graph is all about. So let me call out the first item here just as an example of the results. So here I can see something from Loom that it has found. If I look at this example, I can see that it's found some screens where the background colors are hardcoded to white. Now it knew this background color was relevant even though it's not called dark mode anywhere because of the semantic understanding of what it is that this code is actually doing. And better than that, it's now suggested bug fixes. Right, bringing together multiple context types. So here below I can see that it's taken the code understanding from Bitbucket and it's got in the knowledge context our Atlassian design system. So it now knows how to fix it by moving to design tokens.

21:54 Of course below this you've got user quotes. So here I can see some user feedback sessions stored in Google Drive and this customer call from that new GR connector. A great example of the compounding context that I'm trying to do. not just about source code. And did you know Mike that this is not just for engineers? In fact, last week I wanted to know if there was a feature we had talked about if it had been checked in yet. So instead of asking an engineer, I asked Rovo and it gave me an answer because of the new code context. So it's a simple but powerful example of how you can save time for this saved time for me and for the engineer. That's right. And as Tamaras said, it works with well, you work with a lot of engineers. and we've got a little something we've designed specifically for engineers, which is our new code search app. So, it's a dedicated place for searching code accessible from the app switcher. And I can now switch to the dedicated code search app. Same Teamwork Graph, same code context underneath. So, I can search here and do a different query, dark mode themes, for example, and it'll come back. Now you'll notice the interface of the code search app is different. It brings together results still for my entire engineering organizations repositories side by side and I get used to the filtering that is more of an engineering centric approach. I can see the files and I can obviously filter them by extension or path etc etc. Now, we've been using this internally for months and it's become pretty indispensable to our engineering teams when they're doing those deep spunking exercises across large code bases and it's coming to all of you very soon included as a part of the Elassian platform. Now, since that source code is part of the Teamwork Graph, it's not just Rovo chat as we saw in the first example or here in the dedicated code search app that you get these results. In fact, all of these results are available everywhere. So that code context you'll see is available if you're in Jira or Confluence anywhere that you do a search you will get code results and if you're using the MCP server or our CLI or your favorite agentic harness again all of the same results in 6 months of our early access program on code context the customers using it have been seeing amazing results for example Atlassian Williams Racing greatly reduce token costs and improve time to query if you have your repositories already connected to Jira. It's now a single click in the admin hub to turn on code context for all of those repositories.

24:21 Now, one thing we haven't talked about because they're doing a lot with code is how agents are going to be represented in the Teamwork Graph. You're right, we have not yet. So, let's talk about agents. agents are joining your teams and they're going to be just like your human team teammates. so, we need the same understanding of them that we have for the people we work with. This includes ways to identify them, manage their permissions, and have visibility into their work.

24:48 We already have people in the Teamwork Graph, and now we've added agents so you can connect and attribute work to them. But today, that work happens with in communications. You're having many conversations with your agents all day long, just as you are with your other teammates. And when you close that chat though, the information is lost. when you close your CLI, the information is lost. So, we're adding all of those agent sessions to the Teamwork Graph as a part of the communications context. Let me explain why that's quite a step forward. Now, your teams, as I said, are spending their days steering agents and building a shared understanding together. Those agent sessions, though, are an incredibly rich source of connected information. They capture a lot of key decisions, a lot of linked documents, and a lot of mutual learnings from the back and forth.

25:38 We're now connecting them into the Teamwork Graph as well as indexing them lexically and semantically the same way you just saw with all of that source code. Then we take what was learned, we distill the important essence of that session and store it as a reusable session memory. So when our next teammate or agent picks up a piece of work in a similar area of the Teamwork Graph codebase, they build on what was already learned, the definition of compounding context. Now, of course, this saves you a heap of tokens and it works across all agent harnesses. We'll continue to unpack agents and agent sessions throughout the keynote in a number of different examples.

26:22 But we're not only adding to the Teamwork Graph, we're also delivering more powerful ways for you to work with your context. So we're introducing rovo work. We've added a whole computer for every Rovo session that arrives in about 60 milliseconds. And you might ask, why would I want that? Well, it lets you solve a whole new class of problems with Rovo. ones that require multiple steps that may take longer or may even require writing code or creating new skills. So, here you have an Atlassian expense policy. It's a rather long document, maybe a little bit dry, but it's very important to Mike that all of the employees at Atlassian understand it. So, we're going to make it a little bit more fun to learn. So, we're going to open up Rovo right here and say, create a game, a interactive game with three questions from the content on this page. make it fun and have a surprise at the end. So, I am going to pick the all new Rovo work mode and, send that off. And you'll notice that Tamar used the voice features of Rovo.

27:34 So speaking directly into the embedded mode at that case to quickly get her message across in the way that is most natural to her. And you'll notice that I can close my laptop and I'm able to do this because Rovo work runs in a secure sandbox in the cloud which means that it can run as long as it needs and that's important because some of these tasks can take tens of minutes to tens of hours to run. Now for the administrators out there, that secure sandbox runs in the cloud in the Atlassian platform with your ability to control secure network access and granular controls. Now because work is part of the Atlassian platform, it has access to the exact same team graph and inherits all of your permissions here. It's going to run as Tamar. Any enterprise features that you've turned on like data residency for example are fully supported in those sandboxes. So everything you expect from the Atlassian platform exists for these amazing computers that we are giving to Rovo. So Mike, should we check on it? We should. So now I can check on it on my phone, not just on my laptop. So I'm going to use that new Rovo mobile app that Mike talked about.

28:41 Going to open up Rovo and I see the game that it's working on the quiz. And you'll see a dot buy it, which means that it needs input. So it's going to ask me about the guidelines. So I'm going to say use Atlassian guidelines. and we had said we want it to be funny, make it fun. And let's see, a quiz that title. So, we're going to send that off. It's going to create a plan and it's going to work on it since this could take a couple minutes. We ran it earlier and to show you the results. So, this is where we ran it and it will come back and it starts with the plan that it shows you the plan that I had agreed to and accepted earlier, but I could have made changes to that as well.

29:19 And here is the quiz which I can click on in Rovo and then I can also full screen it. so Mike, shall we test how well you know our expense policy? Yes, I read it frequently but this seems a much more entertaining way to to learn about our policy. so where must you submit all business expenses? I'm going to go with B through Atlassian's designated corporate expense portal. Okay, that was a pretty easy one. yeah, you have that right. Beautiful design. Love that.

29:50 At the team dinner with several Atlassian, who pays and submits the expense report? I'm guessing me. So a the most senior Atlassian in attendance by job level. And I think you are always the most senior person in the room. yep, got that one right. Okay, last one. When are you allowed to expense hotel laundry or dry cleaning? well, look, let's make this fun. I do spill a lot of coffee on my shirt. So, let's let's let's have some fun. Let's go with a Okay, I can attest to the spilling coffee. no, you got to pay for that one yourself. so, how did you do two out of three? Remember, I asked for a surprise at the end. So, isn't this cool? I can do a scratch pad and I can see our new Atlassian team emojis. Isn't that cute? And you also get get a code that can pay for your laundry if you need it.

30:46 so that is rovo work. You can make a more complex task all grounded in the context of your Teamwork Graph alongside any Atlassian app that you use every day. So we just made an interactive quiz here that was fun, but there's a lot more things that you can do with Rovo work. You can do things like building calculators, building prototypes, doing deep research, but you're not just using Rovo. You're using lots of different AI agents. It's amazing to see what people are creating with the AI these days. There's there's so much happening. There's so many cool HTML pages that people are creating around the internet, but then what happens to them? You might have shared it in a chat in an email but they're also not accessible to your team and sensitive work may be getting shared without the right permissions which is a security nightmare. So this is why we're introducing the artifacts app. Artifacts is a new object in the Teamwork Graph. And through the artifacts app, you and your teams can find, share, secure, and reuse all of your artifacts. Again, this is the artifacts app. and you'll see the game that we just created here, but as I said, you're not just creating from rovo. So, we're going to go to chatt. And here we have a beautiful 3D render of the Atlassian Central, which is a new building. Yeah. Amazing building. Does it have dark dock mode? It does. Look at that. Of course it does. Dark mode in a building.

32:23 Mike's favorite part of this 3D render. This was built with Codeex. And okay, I have to show you a couple other cool things. A street view, the rooftop, and the entrance. Amazing type of visual 3D object created in those Rovo computers. In this case, created in chat GBT and showed up as well. the type of content we've never had in the Atlassian platform beforehand, but really shows how we're going to consume things in different ways, 100%. And you notice I said upload to the Atlassian artifacts. This is using our new MCP, which will support uploading to artifacts.

32:58 But that is not the only way that you can upload to artifacts. You can actually drag and drop. So I can go to the Finder and I can have this. You I think you all have some kind of MD files sitting around on your desktop. I'm sure Mike does. lots of mark files. so I can just drag and drop it, set the permissions and it will come up and I can see it render a markdown file. Pretty rendering. That's right. In our platform way to read your markdown files. And now I go back to artifacts and I see the MD file and I already see the Atlassian central with a nice preview of it that came just now live from ChatGPT into the Atlassian platform. so you'll notice as I scroll that there are artifacts from all different AI tools and you can see the icon and the icon represents which tool it was uploaded for. So most of I when I talk to our customers, most of our customers are not using one AI tool. They're using lots of them like we do at Atlassian. So it's really important you have one place where you can store all your artifacts from across the entire platform. I want to show you a couple more things about artifacts. So here is an artifact we created earlier for project savings for for the Atlassian central and I can share it with the same dialogue box that you see in in Confluence. So it's familiar to you. So, I'm going to share it with Mike, but I can also make it open. When I make it open here, this means that anybody will be able to search for it in my organization and find the artifact. So, I can go to this search box, but I could really go to any search box in the entire platform. I can search for Alaskan Central, and it's going to find any artifacts that were created. You see the the artifact icon that were created. And I can also just filter and just search for artifacts. And I see the the savings calculator and I'm going to copy the link so I can show you how I can embed it to a confluence page. So this is a confluence page again about Atlassian central and I'm going to just paste it right in here and the artifact will render embedded in the confluence page and it's live.

35:12 It's a living artifact which also means if I make changes it will automatically get updated into the confluence page but I can embed it anywhere in the Alassium platform. So here I'm in Jira and I am in a ticket and I can just embed an art a different artifact there. I can go to a whiteboard in conflict in Confluence and here you see that cool 3D render again. Have you ever seen a 3D render in a whiteboard in Confluence? Not until today. and you can also embed a an artifact in a project. so this is a power of artifacts. They can be accessed and embedded anywhere across your platform and you're getting that compounding context because they're all in the Teamwork Graph continuing to go forward. Yes. So because they're built on the Atlassian platform, this makes everything that Tamar just showed made in Revo work. It was uploaded from ChatGPT. you use the the harness or any other agent harness or markdown files makes them all searchable, sharable and governable with all the permissions you're used to. And of course, all those artifacts are going back into the Teamwork Graph. So any of the links they have are creating further connections and more context for any agent, for example, accessing the graph through MCP. Now, do we have any developers in the audience? Anybody? So usually some vendors and developers here.

36:34 they're very quiet, I guess, today. little Easter egg. Everything you just saw from Tamar demoing in the Atlassian artifacts app. Everything you just saw was built on Forge. The UI, the app, the Teamwork Graph access search, the smart link, and the MCP tools is all built as a single native Forge app that we are going to be shipping to all of you as a part of the Atlassian platform. Really showing just how far our extensibility story has come. That's built natively in Forge by us.

37:04 Very cool. And that also was very quick to develop because we use Forge. That's right. So we've shown you a lot of ways that we're making the Teamwork Graph more capable. We have more connectors. We added data context. We added source code. We added agents, your conversations with them and the things that you make with agents. So all of this together, you have a shared foundation that your team and your agents can better collaborate. Now those 250 billion connected dots in the Teamwork Graph are creating that compounding mode for your business and I think you can see it's going to continue to grow at a very very rapid rate. Massive thank you to Tamar and all the team who've been working on the Teamwork Graph. Now that's a lot of talk about context and context is amazing but on its own it doesn't get work done right.

37:57 So people and people working with AI increasingly is how we get together to achieve great things. That's why we're investing so much in our agentic multiplayer protocol or AMP. Enabling how people and agents collaborate both on and off the Atlassian platform. So to get into a few examples, what better way to do that than collaborating with another of my colleagues, Share. Come on out. Thanks, Mike. It is so good to be back here at Team Europe. I am really excited to show folks what AMP looks like in practice. So, I thought we'd start by looking at the Atlassian MCP server. Now, Mike, I know you you love that server a lot. I do. I do.

38:42 We're very excited to announce that today we have an all new MCP server rebuilt from the ground up so all your agents can be part of these collaborative loops. So let me tell you what's in it. So firstly, we've given agents a lot more tools. So in the first version, we had dozens of tools and now we have a 200 and 220 tools that are available for your agents to use. Now secondly, we've given agents a lot more context. So not only do they have access to the context from the team graph that Mike and Tamar were talking about earlier, we've also given them access to almost all the Atlassian apps across all of our collections.

39:22 Thirdly, the license MCP is extensible. So, it supports the customizations that you all build, whether it be a custom forge app that you've built for your business or using one of the thousands of apps that we have on the Atlassian marketplace. And soon, if you built a custom Revo agent, that'll also be available through the MCP. And all of this can be governed through access controls thanks to the power of Atlassian Guard. And we've also seen up to 25% reduction in tokens when you're using our MCP server with SAS apps. It is a lot. So more efficient, many more tools.

39:59 But Atlassian's MCP server was already super successful. In fact, millions of you out there use it every single month. So we've been building this new version in close collaboration with our partners like Anthropic. In fact, Atlassian is one of Anthropic's top used MCP servers across the entire globe. Should we show off some of those capabilities in a bit of a demo? Yeah, live demo. Nothing can go wrong. Nothing can go wrong with live demos. Let's jump into it. So, let me show you how the MCMP server underpins a lot of the AMP experiences both on Atlassian as well as off Atlassian. And I'll show you this by giving you an example of how agents can understand more context than ever before. So over here we have a team that's whiteboarding together and brainstorming ideas for a Dutch houseboat rental site and they want to add a new hero listing here to their homepage. Now this whiteboard is very unstructured, something that wasn't previously supported in our MCP and the team is building their site with Figma make. So let's see how the Figma make agent can use all that unstructured content in that whiteboard to understand and make that change. So I've asked it here to go and look for that whiteboard just to make the live demo more risky. Mike, I haven't told it where the whiteboard is. I've just said go look for it graph. let's see how it goes. Now previously the MCP didn't understand the unstructured content. So the whiteboard is the layout with different options and images there. And you can see here it's picked up that there is an immersive hero image in the whiteboard. it's also picked up what else has it picked up here? That there's going to be a fact banner or whatever. So it's automatically reading the whiteboard. Now this takes a few minutes to run. So here's one I ran this morning. And to contrast it with the site that we have the new site on the left is the old site and on the right is the new site. So, it's understood. We want that hero image at the top and the footer and the the rental listing there. It's done an incredible job. So, that's pretty awesome. We pulled all that context into Figma's agents through our MCP server. Again, Sheriff didn't ask for that specific link. It found the right link, processed the whiteboard, and all that unstructured data. Ensuring that AMP works well off at Lassian's platform required all that context to be securely available to other agents in this example from our partners at Figma.

42:24 What about though when we need to collaborate back and forth multiple loops with that agent? How do we do that? You know, Mike, they say a picture is worth a thousand words. Then I think a video is probably worth a million words. If I could, it would be awesome if I could just tell the agents what I would like visually. This is where the new Loom can help. Today, we're announcing the new Loom desktop app for Mac. You see, Loom powers video in AMP.

42:50 It is the most natural way for you to collaborate with your agents. So, not only can Loom record your screen, can record a meeting, can record yourself, Mike sends me a lot of himself, but now you can also record a Loom for an agent. And this is awesome. It just doesn't capture my screen. It understands everything I'm saying. I can point to something, select a specific element, and it understands me just like a teammate. Loom takes all of that context and turns it into a prompt that the agent can take action on. So, let's take a look. I'm going to jump here into our premium listings page to make some changes. Let's load up the new Loom desktop. Oh, is that what I look like? Sorry folks, it's live. you'll see all the desktop recording options there and the beautiful new design. And I can even see my microphone's working. Gez, I look a bit loud, but it's really handy. this I don't have to go is my microphone working back and forth and see the indicator there. Beautiful. Sweating the details. It's very nice. Quickly change my video permissions in line and make myself look good with one of these gorgeous backgrounds. And when I'm ready, I can record for an agent. So, check this out. I'm going to select the browser. Three, two.

44:09 on this website, I just want to reduce the height of this hero image. It's a bit tall. So, can we reduce it by about 25%. This button here on the right, let's make it more prominent. Let's go with a red color and just change the text to make a booking. And at the bottom here, I like this footer here, but it'll be nice to add some icons. Oh, the icons are done by our design team, and they have the design system documented in Confluence. So, let me switch over here. So, if you can grab these icons from Confluence, that would be amazing. Thanks. All right, I'm going to kick that off. And while that's going, let's jump into the Figma page, the Figma make site, and ask it to go find that Loom that I just created. All right, Mike, that loom I showed you. while that's thinking, let me open it up. And, what Loom is doing here now, it's generating an agent prompt. So, it's going to understand everything I just said, pointed, selected, and bring it up.

45:15 Don't act too surprised, Mike. Now, I can send this to my agent of choice up the top right. So, a really good example of AMP being able to share like you would with a human teammate. You can share it with an agent. And Mike, do you want to geek out with us on this prompt? Sure. So, what Sharief has done is not just create a video with Loom. He's created a detailed, rich prompt for an agent. Shri used a very natural human way to record what it was that he wanted to change. You saw him speak, he pointed, he drew. That is the essence of Ant in terms of how a human collaborates with an agent, doing it in a very humanentric way. And you can see in the content over here, it's got the screenshots and acceptance criteria. You can see it's highlighted the red background and the button. It's even got the link to that confluence page which can all be provided in a prompt that's going to combine multiple content sources together that that prompt is then going to send off to Figma. Did it work? It is awesome. Did it work?

46:10 Let's Let's find out. Let's hope it did. Mike, over here you'll see. Wow, that was quick. It's gone too fast. That happens in real time. The image height reduced button red make a booking. Perfect. Check icons from the conference page. Isn't that incredible? This is such a good example of AMP's powerful combination of experiences, technology, and patterns in action. It's done everything that we needed. So truly multimodal asynchronous collaboration in AMP. Now let's have a look at what we can do for real time collaboration.

46:53 Yeah. to look an example here. Let me show you real time collaboration with a new feature that we have launched in the teamwork collection and it's called planner. So planner is an experience where humans and agents can collaborate in real time and turn ideas into plans to solve really complex technical problems. So here's what I'll show you. We'll collaborate with a planner agent to create a sophisticated plan from all our context in the team of graph. Then the agent will give us a readiness assessment. How ready are we to action this plan and what do we need to collaborate on? And then I'll pull in multiple of my teammates and specialist agents to help us move that work forward into our team's backlog in Jira. Let's jump in. First demo went swimmingly, man. It's only downhill from here. Is that what I'm saying? This one's only a bit harder. Over here, we've got a team that's working on a new payment option in a banking app. So, think of this as like a really critical big change. It involves entire stack.

47:50 Lots of teams need to collaborate together to get this done. So I'm in my team's Jurospace here and I've got my Confluence docs connected and I'm going to go ahead and create a technical plan. Now you can use one of our out of the box blueprints that we ship or if you're a company or team that has your own standards for a plan that you want to implement, you can go ahead and and use them as well. So I'm going to ask it to create a plan for our new payment options heading into Europe.

48:20 So, while that's loading, I want to point out a few things going on here. We know that every team works differently. AMP on the Atlassian platform supports all the way that teams work. So, you can see that for example here, large complex changes, we have those blueprints. We support standards like OpenSpec all the way through to your own company specific technical plan and specification items. However, if you've got quick changes, you may have already seen that we launched earlier this year the JER agent for Microsoft Teams and Slack. Great for going very quickly from your chat tool into action for environments and teams that are working like that. One other thing I want to call out, we are using the planner here and we've kicked it off from Jira, but planner works from Confluence and indeed across all of the teamwork collection. So, you can call it from multiple places and you can see it's running live here. It's gone off to the teamwork collection. We're getting the code context in here to build a really comprehensive plan from a lot of sources. And this is an example of real-time collaboration that we're going to show you across lots of places. Yeah, it's picking up code context from GitHub. It's picked up Google Drive context. I haven't told it about this stuff. So, let's jump into what it creates because this takes a while to run. So, you'll see here that the planner agent has helped me created up the top left three specifications. A design spec, feature requirements, as well as a technical spec. And you can see up the top left, that's me and the agent collaborating together. Now if you have a look at the page, it's picked up all this extra context from my team graph. It's identified the goal that this is related to. I haven't told it which goal what the goal is and it's found the relevant Figma design and even a video design walkthrough that my designer has made. Now if you cast your eyes at the bottom of the plan here, it's given me a readiness assessment. So, it's saying, "Hey, this spec needs a little bit more work." And it's given me some ideas of what I need to collaborate with with my team to try to close those gaps out and give me ideas of what I need to do next. So, now let's contrast what I just did to what you might be doing, some of you might already be doing.

50:21 You're probably working with a local agent harness to create a plan. Now, the downside of that is it's not sharable. I can't collaborate with my team. I can't bring it bring the rest of my team in and the other agents to work on it together. So to show that in action, I've asked some of my mates backstage to jump in and show off some multiplayer collaboration. So Dave, you want to jump in? There's Dave. You can see up the very top as well. And you can see a presence between now myself, Dave, and the planner agent collaborating together to move work forward. And because this is a native conference page, even though I'm in Jira, I can just share it with my team. Or if I just jump over here to Confluence, you'll see that all that stuff is synced in real time. Now, this is a really big shift. What we're trying to do is show off the real-time capabilities of AMP. It's not just planning that we're making collaborative. We built an entire platform for a multiplayer way of working across the Atlassian platform and across the team collection. And a really good example of AMP, the alternative of local markdown files stored in repository checked in and checked out doesn't have anywhere near the same collaborative or real-time capabilities.

51:27 People and agents working together in real time with the exact context that they need. Yeah, exactly. And AMP works with all the agents that you bring. So for instance, over here I've got some details about the specification and I'm going to ask the Figma agent to come in and give us a hand. So I can simply I mention Figma just like I would another teammate and give it some specific instructions. Now you'll see here that it starts to call the Figma agent. And I can immediately see that it's thinking in real time just like I would be chatting to another teammate. And if you cast your eyes at the top of the page, you'll see that the presence indicator of that agent also comes in. Again, transparency is baked in into AMP. And this agent here will go in and make some changes.

52:13 So I've asked it here to embed the main design and give us some more information about what's needed. Nice. Worked. Does he sound surprised? so the agents embedded an image there in real time and written co-authored the page with me. You saw its cursor appear. Oh, and Dave backstage is also working with his agents of choice. Rover wants some feedback. Survey. This goes to my team, so I should probably give him some feedback. but up top here, you'll see that Dave is working with ChatGPT. I can also see him and chat working together. And up the top right, you can see all of us collaborating. So truly a multiplayer collaborative environment. Now, let's zoom out. We can see everything coming together. I can see all the people and agents that have contributed to this specification. I can even see I love this feature which agents have accessed this context and who called it. So I can see that the Figma agent is using this page as context and to get the full history all of the attribution of people and agents is also available. So I can jump into sections and I can see that yep this section was edited by ChatGPT and Dave. This is why AMP really matters at an enterprise scale. You get full identity, real-time presence, and transparency. So, it's awesome that you've got the Figma and ChatGPT agents. Figma running in there from called from Confluence, ChatGPT running in a CLI backstage, all collaborating with multiple people in real time. In this case, called as I said from Confluence and from the console. Once your team and agents have weighed in though, we've got a refined spec, those three specs that make up our plan. What are we going to do? What are we going to next? What do we take the spec specs and turn them into? I think the audience knows this very well. so when I open up the spec over here, I can see now by the way the agents updated its confidence high readiness. Got to impress my boss. And on the right here, I can anytime start to break this down into my team's backlog. So I go ahead and do that and it'll create work items for all the specifications that me, my teammates, and all the agents they've brought in have collaborated on together. I can see that it's some are already assigned to an agent. some of this team has agents in agentic automation so they've picked up the work immediately and all of it is tracked in one place. Now it's amazing to see AMP as we said enabling Figma ChatGPT in the planner agent to work alongside your human teammates and we want teams to be able to choose their own agents their own harnesses. Every integration gets the benefit of AMP, which is why we're really excited to announce a new closer partnership with OpenAI to help you do even more with AMP. For example, you'll soon be able to assign those work items directly from Jira to ChatGPT's codeex cloud agents. And we'll keep working with partners like OpenAI, so teams can use the agents they want in the place they want, locally or in the cloud, in the way that they want to work. That's a lot of ground covered in that one demo. Different ways of working, different places where work happens and increasingly people and agents working side by side to get one outcome. How we collaborate whether it's asynchronously as we saw in the Loom desktop app with record for agent or whether it's in real time co-authoring with the planner agent, we built shared experiences, patterns and technology for people and agents to collaborate. Secondly, it's really important you have transparent identities. AMP is built with that transparency and accountability in mind as you just saw. Everything from live agent cursors to avatar stacks and the history showing you who's contributed across the system of work.

55:58 And underpinning all of it, open context, the new Atlassian MCP connecting hundreds of tools across your entire stack into the Teamwork Graph and all of those partners coming in to work with us to help you get more done. It's more capable, it's more secure, and it works with whichever agent harness you choose. That's why we think about AMP as fueling multiplayer collaboration. That is a lot, Mike. A lot of content so far. And you know how much I like video. I'm a visual thinker. And sometimes the best way to digest a lot of content is is a video. Do you want to show them something earlier? So a little bit of fun here. Early this morning, Sharief recorded me in a loom recapping my thoughts on this topic. So let me start if we can bring up the screen on mine. Thank you. I want to start by showing you the original video. So, let's watch this video recorded very early this morning. I apologize. This was before many call. Right, it's recording. no one's here yet. Let me get you a bit of background once. All right, bit of background. Everything you just saw isn't a collection of isolated features. It's a result of fundamentally thinking how humans and AI work together. Every platform leap we've made has been about unlocking the next layer of connection. In 2011, cloud brought our teams together online.

57:13 In 2020, enterprise data residency made Jeez, Mike, you look like collaboration there. You look like you just woke up. I I I did. It was very early. Real jetlagged. Okay, so that the point is that's the original video. You're probably pretty familiar with that from Loom. One person talking to a camera to get thoughts across. Now, as part of the new Loom, we're launching a new feature called AI overlays. Now, think of AI overlays as an agent creating alongside you to help you get your message across in the most effective and compelling way for my audience. We go back to the demo. We're going to watch that same video again, but we're going to this time toggle on a little option. So, again, AI overlays are very easy to create. So, I can do things like, you know, change the color of the overlays that I want to get. I can have a few very simple options here, but I'm just going to, for the purposes of speed, turn on the AI overlays, and we're going to watch the exact same video, but this time with our AI overlays turned on. Tell me if you think it's better communication. Right, it's recording. no one's here. Let me get you a bit of background.

58:19 One sec. All right. Everything you just saw isn't a collection of isolated features. It's a result of fundamentally thinking how humans and AI work together. Every platform leap we've made has been about unlocking the next layer of connection. In 2011, cloud brought our teams together online. In 2020, enterprise data residency made collaboration trusted at scale. And today, you saw the Agentic Multiplayer Protocol or AMP in action. And the foundation of AMP is context.

58:51 In the last 6 months, we've accelerated to 250 billion live connections across the Teamwork Graph. But that graph isn't just sitting there. It's actively powering an explosion of work across our entire ecosystem with over 11 million monthly AMP interactions between teams and agents and more than 15 million daily tool calls via MCP. When your tools, your agents, and your people share the exact same context, AI stops being a solo assistant and becomes a true teammate. Working with AI isn't single player anymore. It's multiplayer. Wow.

59:32 That that made you look a lot better, Mike M. That's right. Pretty pretty awesome feature that is doing it all for us. Right. So those motion graphics that you saw multiple places were entirely generated by Loomai using Rovo using the Teamwork Graph and using that same exact source video. So using that transcript in advance and of course using the Teamwork Graph and that is what the power of what we can do to help improve your communications. No scripting no designing no coding was involved there just me and my loom agent creating alongside me to hopefully help my message land even better. And with that I want to say thanks to Sharief. Amazing set of demos of AMP on and off the Atlassian platform where human judgment happens. See that?

60:21 All right. Touch a bit of wood. No demo fails. That one worked. That was good. now agents and AI are increasing the amount of work that is happening across all of our organizations. And we're all feeling that acceleration. We're all trying to work out how to understand what on earth is going on sometimes. So, how do we gain confidence and calm in all the things that our agents are doing? To help me explain how we're tackling this challenge for your organizations, please welcome our co-CTO, Tun.

60:58 All right, Tun, first time up here. Feeling calm and confident. If I'm being honest, it's very hard to feel calm with like thousands of people on this big stage. However, in all seriousness, I do feel very confident because of what our teams have built. So at Atlassian, agents are already an integral part of how our 13,000 employees including 6,000 engineers get work done every day. In fact, we have over 11 million agent invocations running on Atlassian platform every single month.

61:28 And I get to use those exact capabilities that we are showing off today to all of you to keep my own teams running with calm and confidence. That's right. As the acceleration increases, we're seeing that confidence show up in three big places. Firstly, we're working at the individual level. When agents are producing a lot of output, how do I trust it? How do I know what it is I should be doing with it? At that team level, how do I know what everyone on my team is doing, what they're working on, and how do I get that understanding? And at the organizational level, I need to know what's happening across the entire company, where we're getting value, and where we might need to make further investments. Why don't we start at the individual level? All right, let's do it. So you know for software teams most of the code today is AI generated not a surprise to anyone even in Atlassian and last month alone 77% of all our pull requests were completely agent generated now with that speed it's creating bottlenecks in other parts of the life cycle for every engineer I mean I'm sure there are some in the audience our pull requests are growing longer and longer and typically approving a pull request is a very tedious process you need to get the context of what changed and why You also need to build confidence that this change is correctly implemented. It meets our team standard and it's safe to ship to production. To make this much easier, we are introducing interactive PR reviews with Loom. Now, it brings the Atlasium best of Atlassian platform together. Rovo Teamwork Graph and Loom to turn every pull request into a short consumable video that can explain that key change, how that change is implemented and what was done to verify that change. How about we see it in action? So this is my pull request queue.

63:10 Don't judge me. I will get through this work. But let's go over to the first one on the top. Oh, all right. So this is about adding object opacity controls on conference whiteboards. 54 different files. Okay. So this is going to take me a while. That's going to be our rest of keynote. But no, we will go back here and that's where that Loom interactive PR review is there. So how about we give this a watch. All right. This change adds two separate opacity controls to confluence whiteboards. Previously, anything you put on a board was solid and hid what was underneath. Now you can fade the fill or the whole object. Let's see it on a whiteboard running this branch with both new controls on the toolbar. Here's a regular object with text sitting on top of some notes. Now the fill slider drops its background to 50% so the notes behind show through while its own text stays solid. And this second one fades the whole object except the handles so you can still grab it. So why two sliders instead of one? The design doc answers that with only one. You can't fade the background without fading the text along with it. All right, let's just pause and recap what we just saw. Rovo showed me visually what was built. Then Rovo also automatically pulled the right context from Teamwork Graph explaining that decision to use two sliders instead of one. And best of all, it actually linked that design document where that decision lives. So just to be clear, nobody sat down, nobody checked out the code, nobody recorded that walkthrough or ran any code. Absolut absolutely not. This was all done automatically. No humans involved. Ro used the context from the Teamwork Graph to understand the change. And if you noticed, it ran the old code and the new code. and it only recorded parts of the experience that changed with that PR, making it super easy to understand this. So, how about we watch the rest of the loom and understand how this change is implemented. Both opacities now arrive as uniforms. Previously, this shader wrote the fill color straight out, fully opaque. Now, this line is the rule. Background alpha is fill time object. The DOM path computes the same product. Existing boards go through a migration, which is where I'd expect a regression. So here's the suite on this branch. The new tests pin the two values undo and redo and that migration. All right, that was our review. So what it did, it called out all the important changes including the need to upgrade the whiteboard version with this change. Now version upgrades can be tricky. So in the end, it also talks about how it tested that version upgrade including running the entire test suite. So now with a very simple video I was able to get full context around the change, how it was implemented and how it was tested. Now at this point if I feel confident I'm going to go ahead and hit that approve and if not I can always dig deeper. So this is a great example of what we're trying to do for individuals.

66:02 We've taken a lengthy technical audit and turned it into a 60-second visual decision, allowing Taroom to decide, should he just merge or does he need to look deeper as more and more content is coming in. And what I love about this is not just the comprehensiveness and the amazing technology that actually made that for you. It's that it can help bring non-technical people into the loop. So your product managers and your designers who wouldn't check out the code and wouldn't be reading the code in the pull request, they can now watch the video and drop any comments they might have right onto Loom. And of course using our MCP, other agents or other people can pick up and collaborate right there in that Loom video. Exactly. And for those folks in the audience, in case you are wondering, this works across all source code management systems, not just Bitbucket. And it also works across every coding agent and it's coming very soon to teamwork collection. Now, it's important here. Software teams. Oh, thank you.

67:01 We use software teams as an example here because they're really at the forefront of AI and agent adoption. But the challenge of reviewing AI output at scale is coming to every individual regardless of their discipline or craft. We're leveraging AMP so that every knowledge worker gets the full understanding of what it is their agents are doing. Again, do they want to merge or create an exception? But that's at the individual level where that person is dealing with more coming in. There are now hundreds of equivalent agent sessions running across any Atlassian team in any given week. How do we understand the work that's happening and how do you get confidence that all of your teams are understanding the work that they're doing? So it's a big problem today as you called out. Too much of that agent work is getting buried inside people's individual developer environments. Now you need full visibility of the work being done by humans and agents so that you can build confidence that your team is making progress and you can step in when help is needed. So to help with that, we are bringing agent sessions right into the heart of Jira. I'm going to show you three ways it can keep the team and agent work connected. First, how an agent session becomes visible in Jira. Second, how to get that full visibility across all agent activity on your team. And then finally, once we have that visibility, how easy it is to move that work forward. So, you ready for another live demo, Mike? This is one of my favorite live demos. All right, let's do it. So, this is my team's Jira board. Now, it tracks all the work being done. Let's go back and pick up the work that is assigned to me. So, what I typically do is I will head over to my dev environment. So in my case that happens to be a terminal with claude running in it. We're going to ask claude to start helping out on this issue. So let's go back here. so what you can see my claude is connected to Atlassian team bug graph. What that means it's going to pull all the right contexts needed to work on this issue including PRs. Oh yeah that's right. Thanks working. Yeah you see it working. It'll pull in all the related PRs. Oh actually before Yep it did show up. What you see here is that session that was running in my terminal, that session is now part of Jira. So what that means if that agent in my terminal produces a PR or it gets stuck somewhere or it finishes that work, all of that state is showing right here in Jira. Pretty incredible.

69:28 Again, the idea is to work the way that you want to work here. I can keep working with my agents locally if that's my preferred method and the Atlassian platform makes that word visible not for me but to my teammates everybody else who's also working in Jira on that issue or of course on the broader board absolutely so talking about board let's go back to the board and see what's going on there so here as I said all the work of my team but I can now go see every agent that is running on my behalf or on my teammates's behalfs let's zoom in right here so that was the agent that we had started in my terminal and if you can notice it's actually running locally. You can see it from this indicator. Let's go back and see what others are doing. Mike, looks like I can see you are also working on our team's board. Our team really appreciates sometimes stresses out as well. Not sure the team generally appreciates when pull requests come in from me. It usually causes stress. Yes, a little bit I would say. But here is the thing. So your agent shows up here and it looks like you are running your agent in the cloud environment. So here you can see that regardless of where the agent is running, be it in the local environment or in the cloud, all your agents running across all your teammates, all of those agents show up right here. That's right. We've seen a cursor agent, a claude agent. We also obviously support codeex, Rovo, etc. Yes, all of them shows here. And now that you have the full visibility, you can see what's moving and what needs attention. So talking about attention, one of the things I usually do on this board is I can click here to see what agents need my help because sometimes they get stuck. They ask for clarification. Now here is an agent which is running on my behalf. Now this one is running in the cloud and it's asking for some clarification. Now typically I'll have to head over to my cloud environment and then unblock that agent. But I can do that right here without even switching.

71:17 So let me click here. Oh, and what you see here on the right hand side, I can chat directly with my agent running in cloud, which in this case happens to be cloud and it's asking me to give some additional information about types of tests being written. So I just simply click here, send that answer to that agent and unblock or move that work forward from right here. That is amazing. So we are unblocking your cloud agents and your local agents and giving you that visibility. So not only local agents immediately real time as you said showing up visibility of all agents and the ability to unblock whatever is running directly from within Jira if that's the most sensible place and quickest place to do it. Now like many in the crowd I'm often using agents to kick around ideas create some stress for the engineering team. and I'm usually building prototypes and doing things h maybe before we create any work items. I'm usually just starting in the agent. Yeah.

72:10 Yeah. So actually Mike, I'm the same way as well and I would kick off the work in terminal without filing a work item. But now we've got you covered there as well because all the agents that your team is running regardless of whether there's a work item or not, they all also show up right here. So what you see here is list of all agents running against your team and I can see what that agent is doing, what type of agent it is and who started that agent. So I see you have been busy.

72:40 I see another agent. I see another agent run from you. This one is cursor. I made a pull request for this one though. Yes, you did. You did and thank you Mike. But I do see that this is actually a pretty big change. You are changing or you're creating a new new toolbar layout for whiteboards. Now, Jira has always been a place where we bring all functions together. This is the type of work we will benefit from making it a drag work item so that our design team can give some input. So, we're going to take this and we can very easily take this agent session and just drag it right here. And what you see happen is a work item was automatically created. No more filling out any fields. It pulled up everything from that session. And not only that, it actually attached your session back to that work item. Isn't that awesome? That's right.

73:31 Again, you're seeing lots of examples of the Teamwork Graph coming in here. So all the untracked sessions it knew based on the board the team the repositories how to create the right things. There might be thousands of sessions running we've only showed you the ones that are relevant to this team. And similarly when tun dragged and it created there flawlessly the title the summary the description all beautifully filled out from what was happening already in the history of that session and again attaching the session so we can dig in and keep looking. Absolutely.

73:55 And let me show you one more thing on this board with full visibility. One of my f favorite things to do is to figure out if people are working on similar work items or similar work but not talking to each other. So now in this case I can see Carlos who's one of the engineers on our team. He's doing some work which improves the headings or the section headers for confluence whiteboards. I also know we have this big work item going where we're going to redesign navigation. Won't it be nice if we can link these pieces of work together so that they have full visibility? Now it's super easy to do that. I can simply take this session and I can easily drag this over over to this one was the one which says reimagine the navigation of whiteboard. So, I'll just going to drop it here. And in one click, I managed to just link that work from Carlos back to the rest of the team. And you can see some of the real sweating of details there, right? If you see as Thin's dragging around, you can see the red that's putting on the edge of the of the work items. So, you can see where you're going to drop it, such as a drop zone. The red obviously relates to Claude because we take the colors from the agent session. If you see a mouse over the codeex one below that, yeah, you'll see you get the blue hint, etc. So, a lot of sweating of the details to make those interactions and experiences at the micro level just really beautiful and and and it's also very relaxing. I keep doing that in my free time. So, truly amazing. I want to pause for just a second here because you've seen an awful lot. Live agent sessions now connected from your CLI real time into Jurro work items. all of the agent sessions that you have running shown on the board with the ability to see and filter quickly to those that need input with all the filtration that Jira allows you as well as on top of those agent filters to see what needs input and then the ability to talk back to those harnesses to get them to kick them off unblocking them from any challenges they might have or questions they might need to ask. And again, all of those untracked sessions, the ability to connect them either to seamlessly creating new work items or linking to existing work items. And of course we did a lot of dragging and dropping. You've got automation in Jura. So you can do this in lots of different ways. Now if we zoom out, it is one of the biggest shifts that we've ever made in Jura. For 20 years, Jura has been a picture of the work that the team had planned. Now it becomes a system of record for all the work that's actually happening in real time. your work items, your agent sessions, the connections between them and those that are unconnected all for one team with all the contacts directly connected in the Atlassian platform. Now Tan, that was a single team. That's right. What are we doing for confidence? You have hundreds of engineering teams that report to you tens of thousands of agents running. How do you get confidence when you tie all of that together across Atlassian's engineering organization? Yeah, so for that Mike, I use DX. For folks in audience, DX is our app that helps you measure engineering experience and productivity. There's a feature in DX called agent effectiveness. What it helps me measure and figure out if my entire fleet of agent is working effectively. What kind of work these agents are doing and are they using the right models which is top of mind for many leaders today all at our company scale of 6,000 engineers. So let me show you how that works. So now DX pulls each and every session across the entire company and it analyzes them on three dimensions. First is requirements but the instructions to those agents clear and did that agent have enough context.

77:16 Second is steering. This is where agent gets blocked and you need to go unblock the agent. So did we do that effectively at the company level? And then finally scope. Very important. Did those agents stay on the task and not wander off? So now it pulls all of this information and gives me an all of view. But I can also slice and dice a 100 different ways and I get very actionable insights and then I make improvements. I've been doing this for last nine months and things that I got ideas about is where can I improve agent skills for my repos where I can close the context gaps and most importantly where are the hotspots in my team where I need to do more training and enablement for my engineering team. The second part which is top of mind again for everybody is am I managing my budget responsibly? Are these agents using the right models for the right kind of job? Now this is where a model fit score shows me opportunities where I could have used a less expensive model without sacrificing quality and it gives me very concrete recommendations. For instance, last month in Atlassian alone, we had 3,200 agent runs where we use the agents to produce documentation. And for 59% of them, I could have used a much cheaper model. So insight like these help me keep the budget under control. A great example of where DX gives Terrun superpowers to manage large numbers of engineering teams and keep us on track. But as a CEO, I have to manage confidence across every department from IT ops to marketing to finance and HR. Let's have a look at how the service collection brings that same confidence to critical processes running across your organization. So we have a whole lot of new features launching at team26. For example, now when an incident occurs, the Teamwork Graph brings together all of your incident code and people contexts together into the incident command center. Rovo surfaces potential root causes, cross references them directly to code, and recommends concrete next steps and hypothesis.

79:15 What about preventing those incidents though in the first place? Well, the new change risk assessment features in service collection. Rovo analyzes the complex releases on your behalf. It surfaces potential risks, looks for mitigation steps so you can have confidence before your next big deployment. That's the kind of things your ops teams will really thank you for. And engineering for sure. I mean, our engineers won't be woken up in the night. So, that's always good. On the people side for employee on boarding, for example, we're able to track every new hire task in your HR and people teams, whether that's been assigned to a human or assigned to an agent, just like in all those features that share earlier and tun before this showed you. So HR and IT teams can now see all their agent sessions running, what's been done, and what still needs attention.

80:02 Overall, helping to make these processes more efficient. But we also want to make them more human. Rovo can now generate tailored onboarding plans in Confluence, keeping a connection between Confluence and the service collection, bringing together helpful information about a new hire's team, their goals, and all the resources that they might need. Using assets, it can even tell you who is sitting on the desks either side of them, so they can feel confident as they start their first day at work. But how do you know whether all of this is actually moving the needle for your overall business? Well, with the strategy collection, I gain that confidence already, seeing all of my goals, priorities, investments, and progress in one place. Now, at Team 26, we're announcing a large set of AI capital management features directly within the strategy collection.

80:52 In one view, I can get insights on how my AI spend is tracking against my budget and priorities. Now, of course, I can see that spend broken down by different model providers and I can look at it across departments and teams and focus areas. So, we can see how AI spends are trending by team. I can look at human and token capital next to each other and look at it break it down obviously by initiative and everything else. Now, as a leader, that gives me confidence that I can understand what's happening and where to lean in. And today, these features are available in open beta for all strategy collection customers. But confidence isn't just about understanding ROI. As a leader, my confidence hinges on my organization's ability to govern and secure our AI footprint in today's world. So, Tun, take us through what we're doing there. Yeah. So, Mike, in the last few months alone, we have shipped over 50 different enterprise security features. Let me call out the ones I'm most excited about. So, first is nonhuman identities for agents. What that means is Rovo or any other third party agent on our platform, they get their own dedicated account. Now you can fully control what these agents can see and do. Second is Atlassian Guard. It actively protects against data leaks using real-time policies. It also automatically redacts sensitive information across all AI chats and connected tools. And finally, an announcement I'm sure folks in the audience are going to love. So historically your data stayed in Europe but the LLM processing needed us to make calls across the Atlantic boundary because the frontier models were hosted in the US. But today we are officially starting to roll out EU AI inference. All right.

82:39 Yes. So you know we already had data residency in the EU but now your prompts your model inference and the results they all remain 100% within European borders. That means you get access to the exact same frontier models from OpenAI Anthropic and Google running natively inside Europe. So, I hope that gives you a sense of how we're trying to build confidence in your organization, helping individuals make sense of what their agents are producing, keeping teams and agents in sync and understanding all the work going on, and knowing that you're getting value across your organization while keeping you protected and secure. No matter how you work, the Atlassian platform will help you understand what's going on so you can have that calm and confidence. Tun, thank you. Thank you to all those hundreds of engineering teams building these capabilities. Thank you.

83:35 So, we have covered context, collaboration, and confidence. And hopefully you've seen how they all tie together, right, with the Atlassian platform in terms of the better the context you have, it makes for better collaboration. And that understanding is what gives you the confidence to embrace the future, right? That's what the Atlassian platform is built for fundamentally to be that living nervous system as we refer to it that connects your teams, your agents and your tools so you can move forward with certainty. Now, as I've said, the world is changing very quickly.

84:03 We see that. But 25 years in, our mission hasn't changed. I think that's really important. Our mission as a company has always been to unleash the potential of every team. Doesn't matter if they're design team, marketing team, finance team, HR team, engineering team, whatever the team, whichever way they make it, the Atlassian platform is here to help you to do more than you ever thought possible. So, we wanted to recap that with a fun demo and show you one last little touch that we have in terms of the experience, which here you can see what looks very familiar as a Jira board, bringing your company's context to right where you're working. Now, as you do in your companies, when you're going to ship things, and we're all about shipping things, you take that and quite simply drag it over here to done. Right? So, boom, Rovo Desktop shipped. And we showed you, we showed you an entirely large series, our biggest ever amount of development, a delivery of incredible features across a series of things in context, starting with data context all the way down. We showed a whole series of collaborative features as sharif talked about our MCP server and all the other things we did and Turan came out to talk about how we're trying to give you confidence and you didn't just see slides as we select all of these you saw live working software live demos as you're seeing right here people and agents working together in real time on real problems with real code delivered now as such we are going to select all of these and drag them all to done and call them all shipped in Team 26. So all done.

85:44 It feels amazing to ship so much quality software and massive thank you to all of the teams. It has been fantastic. So for me, just a huge thank you to everyone from being here to our customers. Thank you for taking the time out of your busy days. As I always say, you're away from your friends, your families, your partners, your colleagues. So, we really value the time you take to come join us here. And to every Atlassian who's poured their heart and soul and code and their agents and everything into making this an amazing event, from the product teams all the way through to the event staff. thank you very much for everything you've done at Lassian Staff.