Tiffany To on Atlassian’s Cloud Transition & AI Integration | Atlassian Team 25 EU Barcelona

Hey folks. We're at Atlassian Europe and we're talking with Tiffany To, who's executive vice president for Platforms and Enterprise, and we're having a little chat about cloud and ai. Tiffany, welcome to show.

Thanks so much for having me, Mike. My pleasure. Um, you guys have announced that you're gonna end the Life Atlassian data center, which probably doesn't come as much as a surprise to folks because you've been pushing folks towards the cloud for a while.

But, um, why now? What's the transition that you're trying to achieve and well, how far are people making that migration anyway? 'cause you've been at it for a few years.

Great question, Mike. And you're exactly right. It did not come as a surprise to any of our customers.

We've been investing so much in our cloud platform the last seven years. Um, and as folks hopefully saw in the keynote so many new things in terms of AI and all the collections and the system of work deliver all this new business value for customers. Um, but why now?

A big part of it is, as we talk to our customers, many of them have been migrating to cloud. Uh, 99% of our customers have some bit of footprint already in cloud, but what we hear from some of them is that sometimes there can be inertia in their organizations, right? Many of them have had data center for 10, 15 years.

And so being able to make a clear timeline for them that says, this is when March, 2029, it's time to be off of data center and into the cloud platform, enables these teams to start planning right backward from that timeline within their companies and start to build a business case because quite frankly, they're not migrating from, uh, JIRA data center to Jira Cloud. They're moving from kind of traditional behind the firewall, siloed, uh, products that work very well and have scaled, but into a cloud platform. And that's what it's all about when it comes to ai, right?

Every customer that I talk to is so excited about bringing AI to all of their workflows, but to do that, it's gonna be leveraging cloud technologies. It's gonna be leveling how these systems are built out on the cloud. And so for many of these customers having that balance of new value by really rebuilding their workflows with AI in the cloud, and then also having new deployment models in the cloud to support even the most regulated industry customers.

Um, we've announced that we have support not just for, uh, high levels of security with Atlassian Guard on top of our commercial cloud, but we now have our gov cloud for our US government customers that has FedRAMP certifications soon to be IL five as well. Uh, but we announced isolated cloud earlier this year, and so isolated cloud gives you, um, a dedicated tenant. You have isolated, um, uh, compute storage and networking.

And so that's a great environment for some of our regulated industry customers and we'll support that, not just in the US but all of our data residency regions. So mm-hmm. So in effect, I get a private data center is just on somebody else's infrastructure.

Correct. Um, what's been the reception to that concept among folks who do have those requirements for compliance stuff? Are they willing to do that?

Because a lot of them seem to, sometimes they wanna hug their servers and they're like, I own my infrastructure, but, you know, psychologically speaking, that may not as be as, uh, compelling as an issue if I have a private cloud, right? Mm-hmm. No, you're right.

And, and I think that's also another reason why it's been great to actually put this announcement for a send out because we can start to have these conversations with these customers. Um, what we're hearing has been really positive. I think initially we had, uh, a very small set of customers in the US that we thought were gonna be going on to isolated cloud, but once the announcement came out, we've been hearing a lot more demand for it, and not just here in the us but you know, obviously we're here in Barcelona at the Team AMEA conference.

Um, we've had a lot of impromptu meetings these this week from customers here that are excited about looking at an isolated cloud environment for them. So yes, there's gonna be a, you know, change management is, is always hard as, as you alluded to. Um, but I think the signal we're getting from customers is that the excitement to rebuild all this with AI in the cloud and a clear timeline enables them to start putting plans together.

So that point, does AI not force the issue? Because ultimately I have all these models, I need to expose them to data, and the more data I expose to them, the smarter the models get, that all lends itself to the cloud. Because if all my data's sitting in some isolated data center somewhere, I'm really not gonna be able to do that as efficiently.

You're exactly right, Mike. Um, for, uh, customers, a lot of what we discussed with them is, look, everyone is using, uh, AI on the consumer side, and that's all one uniform data set, right? It's everything on the web is the dataset that feeds chat, GPT.

But when you think about harnessing the power of AI for your company, how are you gonna expose all of that data? And oftentimes that data doesn't come from one vendor. You know, they have products from Atlassian, they have products from Microsoft, maybe they have products from ServiceNow, Salesforce.

And so being able to provide that rich context to customers is gonna require a lot of interoperability between these platforms, and that's gonna be done best in the cloud, right? And you, you see that right now with, um, a lot of the conversations around MCP model context protocol, how the agents are gonna access this data, how the agents are gonna work together. So you're spot on.

Any customer that wants to bring AI agents into their workforce is going to have to be able to build some sort of platform strategy that connects this data together. Um, and that's why we've built the teamwork graph. Um, hopefully you've heard a little bit about that.

The teamwork graph, as I understand it, is the, the thing that maintains the context and discovery and the relationships between all the various components that are in the cloud or in the SaaS applications, but not just your applications, third party applications as well. So, um, a lot of folks probably don't know what a graph is, but explain, yeah. Um, a graph is really important because it's about the relationships between the data, right?

You can have all of the data sitting there in buckets, but if you don't have, uh, an understanding of how that data is related to each other, what kind of insights can you draw? What kind of insights can an agent draw? What kind of context can you give it?

And so the teamwork graph was born from, um, kind of pre all this AI stuff, but at Atlassian, customers were asking us to help them get more insights from the data that was coming into the system. And like you said, not just from, you know, Atlassian's products, JIRA, confluence, JIRA service management, but they often are using a pretty rich set of developer tools, right? And other functions, Salesforce, et cetera, bringing in data.

Mm-hmm. And so as we started to look at the relationships between that data, we saw that we could be quite opinionated about it because the way teams work, uh, is usually modeled in a specific way. Development teams have certain objects that they're using, whether it's, you know, repos, right code, et cetera.

Uh, teams work around goals that they're setting for themselves. They're managing projects in Jira. So all these data objects have relationships together and context.

And so we started to invest in making that bigger and bigger. And you, you saw in the keynote, we've got now billions of relationships now between all those objects. And so what that means is think of it as a very unique fingerprint of your company.

And so what that means is when you log into our systems and you make a search query, or you initiate some task, it knows Mike, it knows what projects you've been working on, it knows what team you're on, it knows what your team is working on. And so it doesn't just look at the structured data to answer these questions, it's actually using the teamwork graph to provide context for the query. Um, and now it has memory, as you probably saw in the keynote, right?

So it makes your agents smarter and smarter about the way you work, but also the way you work with others in the company. And that's the really hard bit, right? It's like there's personal productivity and there's team productivity, and those are at very different levels.

And I think the teamwork graph, um, is really exciting because we're starting to see not just, obviously our own teams build experiences on top of it, but we announced, um, today that we're opening it up. And so that means people can extend and add their own custom objects into it. They can pull from the graph and build their own applications off of it.

So we really see it being kind of an, an exciting piece of, uh, making AI really valuable, um, for customers. And That goes back to what you were talking about with context and my personal preferences, and everybody has a slightly different way of working. Does the AI agent in that context become, um, my primary engagement partner?

Mm-hmm. Or am I managing a hundred agents that all have different kinds of memory and different kinds of experience? How do I kind of marshal this small army of AI agents?

What's that gonna look like? That's a great question, and I feel like, uh, the whole agent strategy and what that workflow is gonna look like has been evolving so fast the last six months, right? I think initially it was, Ooh, everyone's gonna have an agent.

You know, Mike's gonna have his personal agent. And then it became, well, no, he's gonna have hundreds of agents and they're all gonna do lots of stuff. And then it became, okay, well every company's gonna have thousands of agents.

How are we gonna manage this? Right? What we've been hearing from customers is they want simplification.

You don't, you want to not have to think about managing a whole set of agents. So we're trying to provide a good amount of flexibility right now because we recognize customers are gonna try lots of different things. And across the wide range of use cases we're seeing, there's probably not a one size fits all.

Um, but we believe what's really important is the skills that you're endowing those agents with. So part of what we announced today is, um, a really large library of several hundred skills that an agent can have. So you could choose to build one Uber agent for Mike that has all the skills.

It can do a bunch of admin tasks for you. It could do your core work, it could do personal work, it could do all sorts of things, or maybe you don't like that and you'd prefer to have very separate siloed agents. So we wanna give you that flexibility to be able to do, um, the model that you'd like.

So I think there's gonna be a lot that shapes o up over the next, um, six months year. But what's exciting for us is we're seeing customers customize, uh, tens of thousands of agents and give those agents the ability to, uh, go and actually initiate workflows in our system. We're seeing millions of workflow automations being triggered by agents already.

So, uh, we believe that probably Atlassian right now is probably one of the largest AI workflow, um, generated platforms. Mm-hmm. To bring this full circle, a lot of folks who have their own data centers will be concerned that their data doesn't wind up training an AI model.

So how does Atlassian kind of protect everybody's data or isolate that data from all the other AI models and agents that customers might be using in the cloud? That's a really important question. I get a lot from our customers.

So the models that we use right now today, we work with OpenAI, we also work with Meta and have their models. We do not share any customer data with those model providers. So the teamwork craft data that we're using to provide that structured context, it's only used when you're querying.

And so it's providing that grounding so that you can get the right response back. But we're not sharing any of that data directly, um, with the model providers. And it's also isolated, uh, per customer, right?

So each teamwork graph instance, right, that we're training is for that particular customer. So, um, it's a really important question that I know customers and if they wanna see the architecture diagram, we've got a lot on the website as well that kind of goes through exactly what, um, the life cycle of that data goes through. So, uh, All right.

Hey folks, you heard it here. You're going to the cloud and there's gonna be a lot more functionality and a lot more features and things you never imagined you could do. So better do it sooner than later.

Tiffany, thanks for being on the show. Thanks so much, Mike. All Right, we'll be back in a minute.