The Future of AI Integration with Sherif Mansour at Atlassian Team ’24 Europe

This is Textron tv. Hi, I'm, Mitch Ashley here in Barcelona at Atlassian, team 24 in Europe. We're, uh, great pleasure being here, you know, extra, extra special pleasure because I get to connect with people.

Like I talked with Andrew previously, so Sharif Matsu, and I know each other. Going back to 2010 or 11, we met, well, first we emailed online. Yeah.

Product manager met at a developer conference at a hotel in half Moton Bay, California. Wow. How things have changed.

It was a lot of fun back then. Yeah. We were, we were living some memories earlier.

It was definitely, um, you've seen the Atlassian ecosystem grow from, from day one almost. It's pretty Amus. Yeah, yeah, yeah.

And implemented it at multiple customers. Yeah. So it's been a good part of my career.

So you're, you're head of ai? Yeah. Oh my God, that's what a great, great progression for you.

It's been a blast. Yeah, yeah. Yeah.

I've been at Atlassian for 15 years and I've, um, worked on almost all the products and, uh, keep jumping around in different things every few years. And, uh, it's been a, it, it's been a, a fun rollercoaster journey that, uh, gives us surprises and new gifts every, every few months. You must Like it, you're still here.

Yeah, It is fun. That's amazing. Gives me some new energy.

So, um, you could tell me if I get it. You know, one of the things, looking at Atlassian's approach, and folks have heard me say this on the video, it's not just ai, AI is on, is part of a platform of data across all your products. Teamwork graph that kinda gives the context, you know, the, the actual meaning to that data so you can do intelligent things with intelligent ai.

Yeah. Am I on on track here? Absolutely.

Spot on. Yeah. Yeah.

The, the, uh, I was talking to someone earlier about like, differentiation in AI and, and I, I generally believe a lot of vendors will have similar AI features. Um, and what matters is how people use that in the context of their work and the data that they use it with. Like, that's really gonna be the, the key difference.

Um, uh, yeah. And it'll, it'll, we're super fortunate as a vendor. We're such a broad portfolio.

Everything from IT ops teams taking incidents, their software teams planning and tracking their work to marketing and sales teams loving loom and like recording their videos for their clients. So the breadth of the portfolio really, um, helps us solve meaningful problems for customers that can enterprise customer of it's end-to-end, um, using the data and the teamwork graph. So that's, that's a pretty exciting thing time.

Yeah. Well, you kind of, it's like if you're gonna hire a consultant that didn't know what work you do versus hiring one that does understand the work you do. Yeah.

That's kind of the difference with the AI agent is I know what work you do because it's all connected through the teamwork graph. Yeah. And, okay, I know what you touch, what work will you do or who, who works on things with you, and you might get information from or hand off a process Too.

Yeah. The old saying AI is only as good as the data that you have, um, is, is, uh, I spoke to a customer who was running a trial, um, with a, with, uh, an AI product, and they're running it on an instance without any data like te uh, no test data. So of course the AI doesn't appear to be smart because it, it just doesn't have access to the information that it does.

Um, and we're quite fortunate with J and Confluence in 20 plus years of, of data. Um, it makes it very easy to extract that organizational insights that companies, uh, have tracked. Um, there, and AI has really helped, I think finally, like the whole big data movement, AI has really helped us leverage those insights from the big data.

Before it was like a lot of big data capture it, but it was really hard to consume it. Right. And I think to the general of ai, when the transformer model came out, like that really closed that gap for, for, for a lot of us.

So that was exciting. Yeah. Well, one of the thing unique things you've done, how, how did you come up with the idea of AI as a teammate?

Yeah. As a user in the system, not a process that runs or a bot that does something Yeah. For You.

Yeah. Yeah. Um, really good question.

It, it was certainly an organic, uh, aha moment. Um, I think maybe don't mean this in any sort of cliche way, but like our mission's to unleash the potential of every team. And then when you look at with, with technology trends like ai, mobile internet, you, it's really important to see the trajectories.

And once you can sort of see the trajectory, you can then make informed decision the way the trajectory is going. Uh, and it's very clear for us that, um, the future isn't this one God mode like agent. It's not gonna, it's not gonna happen for anytime soon.

It's a good way to describe it. Uh, there'll be lots of little, and maybe over time there'll be fewer of them because they'll be able to do more, but we're still a while away off that. Um, and then the second thing is, we believe in that all our products are highly collaborative outta the box.

You know? Um, and when you interact with someone, a colleague, you can mention them and ask them questions. You can pull 'em into a conversation, you can assign 'em a piece of work.

And with Jira, it's all about workflows and Confluence all about collaborating around content. Uh, it became to us very obvious that, I think the unique perspective we bring in that experience patterns is what if you can assign an issue to an agent, and what if you can mention an agent and pull in a conversation? It is actually quite the most natural way because I think one of the challenges we have with AI is behavioral change.

So, uh, you know, it takes a while for someone to adopt a new behavior. Um, when mobile phones first came out and we had our phones and, um, you know, you're like, oh, I'm not gonna use it for online shopping, but all of a sudden you just, you shop while you're watching a TV show, right? Yeah.

Um, I bought a water bottle for my son this morning, who's in Sydney? 'cause my wife was like, can you get another water bottle from Amazon? I'm like, oh yeah, sure, let's do it.

And Marshall a random, random water bottle topic, but, um, but the beha Describe that use case 10 years Ago. That's right. But the behavioral changes is now normal.

And so one way we can help customers that behavioral change and bring knowledge in the context of where they work, is to put 'em in the, put agents in the collaboration flows. So as a simple example, I can write a requirements document if I'm a product manager and a software team, and then maybe, uh, get an agent to pull in and help me break down that work into Jira tasks, I would, what would I do? I'd normally ask a product owner to come in and do, do that.

But now the agent can help with that and the product owner can focus on actually making decisions on which will, what we're gonna do next and how, and, and free them up to the more, more valuable work really than the the tedious work that they would need to do. Yeah. Yeah.

They ask you a strategic question. Um, let's, let's assume competitors say, oh, that's a brilliant idea. I'm gonna do that.

We'll go create an agent that's a user in our Yeah. Application. What do, what do they lack?

What, not naming anybody individually, but what do, what do they lack or what would others lack that to be able to effectively do what you're doing? No, no. Uh, I firmly believe everyone will be doing something similar.

So that's a, a strong opinion. Uh, I think they will be doing it over time. Um, and, uh, and I think that they'll, they'll, that's how we will wanna interact with them.

They're, they're, they're collaborative and, you know, you'll wanna interact with 'em that way. Um, and again, so I think what will be differentiated is the experiences that you do that on. So take an example.

Um, the Jira backlog, it's a fairly sophisticated experience. You see large, dense amount of work to do. You can group them into different themes or epics or whatever you might call them, and then you might prioritize them.

And that's a fairly complicated experience. Now the human brain understands that and sees that, but AI can help while you're working and help reprioritize content video. So it's really about applying that AI in the context of that experience and then using the data that you have that's underlining it.

So the reason we, I believe we do such a good job at breaking down and is issue into smaller tasks, uh, project manager trying to break down a project into smaller tasks, is we have the specs, we have the plans, and we also have the, the previous, the prior arc, the previous, the previous tasks. And we also know the people that worked on those specs. And, and so when AI's really good, when you're feed a bunch of data and ask it a bunch of questions, you get an answer.

And so it definitely helps produce a much better result that way as well. So, no, I do think everyone will have generally more ve experience. I hope they do as a con a user, It's validation too.

Yeah. And as a user of many AppSec, I hope it's just kind of natural to interact with an ai and it's not like, oh, I need to go to a separate thing to do an AI thing and then come back to the app that I'm using, but I gotta copy the results and do something. Yeah, Exactly.

You, I think hopefully as an industry, I hope we, we we head that way and we're trying to set an example. I think it's a, I think it's the, the right way of how people will do it, especially in the age of where there'll be lots of agents that will work with us. Yeah.

Yeah. What do you, what do you think about the model of I'm one of my own agents that know me better Yeah. And understand how I work, what I do, what I need, et cetera.

Yeah. Um, you think that those agents will become more personalized to individuals in a team setting as well as an individual setting? Yeah.

Yeah. Did you have, um, sounds like you might have an example in your head when you were thinking that. No, I wouldn't need, oh, and I, I I'm no confessions on video.

I'm not anything. No, I mean, I was just thinking about the trust that we put in Ai. Oh, totally.

Yeah. Oh, the trust. Yeah.

Yeah. If I, if I'm working with AI in whatever form you build trust with it. Yes.

Because you know how it works, what it'll do, how far it'll go, how to keep it in, in check if you Yeah. Yeah. The more it knows about me, but I can trust it's not going to let that go anywhere.

The more I'll use it. That's right. The more effectiveness.

Now put that in a team setting where yeah, there's, there's agents all work within share. This particular agent I created to help me. Yeah.

But not in isolation me as part of a team member as well. Yeah. Yeah.

You think that kind of thing will evolve maybe to something like that? Yeah, Absolutely. Yeah.

Yeah. And, and, and we actually spend most of our time on the team aspect. I think there's a lot, there'll be a lot of personal assistant ais and, um, yeah.

For even, you know, in Confluence, in Jira, you can create agents just for you. But, uh, if you, if you utilize, uh, our customers that use robo, um, we haven't created any personal agents. They're very much around team oriented, uh, work.

Um, but yeah, as they understand us better, um, they'll be able to do a much better job. You'll also be able to trust it. I think a key element of quite there also is we also, agents never do anything without human intervention.

So, um, I can, I can ask an agent to create a page for me or complete my sprint or whatever it is. Um, it'll always ask me, is this what you wanna do next? I'm about to do this.

Are you sure you want to do that? That kind of thing. So I think humans in the loop has to be a permanent pattern.

The industry maintains and we're strong advocates of this. Um, it's super important at the stage of trust, trusting in ai, super important for transparency, super important for reliability. 'cause agents won't get everything right.

They'll get stuff wrong all the time. And so that's very important at that stage. And, and then the auditability of that is a huge part of that trust.

So, as an example, a lot of our agents have profiles, like humans have profiles, and you can see an audit log, like, this agent created this bug last week and this agent commented on this page. And you can see exactly that. You can see who created the agent.

And the biggest transparency is its instruction. So all our agents, uh, uh, context and instruction is open source, if you lack for a better word, that that gives us a couple of benefits. It gives the human trust 'cause they can understand how it got to that conclusion, but also it gives them ideas on how they might shape it to mold their needs so they can tweak it.

Just like in the open source world, you'll fork someone else's code. Uh, I would, yeah. The many customers are copy example agents and tweak them and remix them and make them work for their use case.

Um, so I think it's Like having manifesto for an agent of like, I know what it's about, what it's totally do, what will work. Uh, and one of the things we do with all our agents is we, we do encourage customers, and it's in our examples. And, uh, it's part of the creation experience to think about the agent's character.

And some people go, well, what, what Is it? Sassy bot, sarcastic bot. They're like, why do you need that in the business space?

And you're like, 'cause you need the sarcastic bot, otherwise, how are you gonna get through your day job? Exactly. Um, no, but like there are so many agents that help with fostering team culture, but also your company's culture might be fairly informal.

And who wants to chat to an agent that is formal all the time? And there might be some agents where you say, I want you to be very deterministic and try your, try your best not to respond if you don't have an answer, because the information you gimme is gonna be critical. And then some agents are gonna be like, I want you to improvise as much as possible to give me ideas.

I'm, I'm in a exploratory mode. I've got a confluence whiteboard, suggest to me 10 other ideas that fit in this theme. I, you know, help me brainstorm.

So I think character is an important part that doesn't get discussed as much. We think they're very transactional. Um, but I, I think it's an important part as we design these agents or users design the agents 'cause uh, you know, we're really building an agent's platform here.

Uh, they think about that in, in that process, which will help with the trust because in work as you expect it to, to behave. Yeah. Interesting.

Yeah. So with Atlassian continue to move in the enterprise and up the enterprise with focus and, and align, how does that inform your AI strategy? Yeah, it is, um, it is absolutely a blessing that we have such a, the tools that the teams use.

Uh, I, I like to split our portfolio into tools for teams. And that's, just think of that as a smallest atomic unit. Five people in a marketing team capturing requests in Jira and tracking their work in Confluence or something like that.

Um, and then we have tools that are for teams of teams. Um, and so you can think of that more as the, uh, Atlassian home communication. So when I create a goal, uh, that's to align my team with your team.

So now, uh, and I share my weekly updates in, in Atlassian home, I'm broadcasting so that you are a dependent team and you need to follow my weekly updates. And now we've got the lever up, which is tools for lead, uh, tools for leadership teams. So you've got teams, teams of teams and leadership teams.

Um, and what that gives us is end-to-end visibility of from the company's focus area to what, how does that break down to actual things that your teams do? And then how does that impact your actual customer outcome that you're trying to do? Uh, and so AI gives us this, um, that teamwork graph of that knowledge, ai, uh, unleashing ai, and then the agents of that does help us to surface insights.

Uh, one of the demos we gave this week is in focus, you can ask a, why is this goal off track? Now if you do that in a, just a goal setting tool by itself, it'll tell you because of what the person wrote. And that might be useful.

We know many times people watermelon their goal updates or sugarcoat 'em, you know what I mean? Yeah. Green on the outside, but red on the inside.

Yeah. Mm-Hmm. Uh, but the reality is there's so much information that we can do to enrich that.

Well, it's red when, uh, the person said it's red because of this reason, but also in this particular focus area, you've had this team down at this leaf node over here. Um, say they're being blocked by this other team for four weeks. And so we can surface that all the way up to an executive because we have the full end-to-end stack, like, uh, like no other vendor can.

So, um, that's exciting. As someone who had worked on Atlas back in the day projects and goals, and seeing it evolve and seeing where it's headed and how AI's unlocked at it's, it's created a big opportunity. Yeah.

It's, it's, it would be a pretty amazing feat not only to connect the dots Yeah. From strategy to execution, but also like, okay, but I need to know maybe where don't we have enough execution on Yeah. Particular, yeah.

Where's the strategy that Yeah. Has the least underpinning of work for us to be successful? You know, gimme some assessment, or at least tell me where I don't, you don't see enough activity.

It's interesting, I was talking with a friend about an idea of also using kind of security techniques, but apply it to like software. Oh yeah. Say we seeing unusual behavior in check-ins on code in this part of the system.

Is it because Shreves never moved back and started working on that? Yeah. Yeah.

Or no, there's somebody else, nobody don't know who doesn't know who that is. It's an open source developer that showed up and poof. Yeah.

Why are they here? So it's kinda some interesting behavioral things that we can also, I like the, uh, security parallel. It is sort of introspecting, uh, like observing behavioral change and then calling that out.

Yeah. Because I need context to understand is that Yeah. Good behavior.

Yeah. Yeah, absolutely. Why Is it different?

Yeah. Yeah. That's a, that's a really good analogy.

Yeah. Great. Well, tell me, so in, in 30 seconds, the future of ai, the Future of AI At a, at atlass.

Yeah. I mean, what do you, what do you think what's gonna be happening over the next six to 1218 months? Where, where things evolving into?

Yeah. Look, uh, um, 30 seconds. I think our products con we continue to resol the problems we solve for customers today in AI first mindset.

And that's why we have AI in our addition. So in Jira and Confluence, you just have AI in there, more and more features. JIRA's all about planning projects and breaking down work.

That gets a lot easier with ai. And then our teams are re uh, in inventing and resolving those problems for customers in far better ways with this new tool in their tool belt that continues to remain. I'd imagine at some point, you know, the whole product's AI first, like, and we, we, we need to head towards that direction.

That seems like an inevitable, uh, to me in my mind. So that will continue to evolve, uh, at the, um, accelerating teams and helping them work faster and augmenting them. That's absolutely will be investing a lot more in agents.

I think two, two main things. Oh, let's, three main things to call out with agents. I think you'll certainly see them get a little more richer in terms of interactivity and interactive modes from today that are largely text tomorrow.

They need to be audio and they need to be video and they need to help customers with multi multimodal kind of problems, uh, to help, help move work forward. Uh, because we know not all work happens in text. Um, so I think that's the first one.

The second thing is you'll see agents permutate across the whole portfolio, uh, and be integrated with more and more products. Uh, for example, Bitbuckets Next in our, uh, integration pipeline. Good bucket.

And you'll do that. So agents is very much a horizontal across the, uh, Atlassian portfolio. And the third thing is, I think you'll see agents do a lot more highly sophisticated tasks that, and, um, to use a better, uh, there's probably a better term of this, that they think slower or sorry, they think for longer.

Hmm. So today, uh, a lot of the generative, uh, AI world, um, the transformer model is really about typing some text and the model just tries to give you the next possible prediction as quickly as possible. The cat sat on the, it knows Mac 'cause it's seen that a million times, right?

But there are some scenarios where you want it to go and think deeper about the problem. Um, so a simple example, uh, let's not pick a code example. Let's pick a, uh, an example of like trying to find out what what might be causing a project dependency.

I could ask an hour model and it could try to return something really quick because it, it, it's designed to return something really quick. I think we'll see agents that will say, I dunno, Mitch, I'll get back to you. Like, gimme a day maybe, hopefully not a day, but like, it'll come back to you.

We'll go away for a while and send you a notification and go, all right, here's what I think. And I think those results that you get for those use cases will be far better than the ones that are instantaneous. And I think you'll need both.

And it's not that one's better than the other. I think today we only have the capability of instantaneous responses. Uh, I think the ones where the, the models as they get more advanced and can do deeper thinking, um, will get you far better results.

And I think the more complex tasks will be able to assign to agents in, in that world that wasn't 30 seconds. So are you another 60 seconds? No.

You Have to throw in a hitchhiker's guide. You Have to have to Okay. That, that was a short answer.

That was A good one. That was good. Yeah.

Alright. So it's such a pleasure. Hey, congratulations on success.

Thank you for you and also Atlassian and I wish you the best. Let's keep up and see how things are going. I wanna see the progress.

Thanks for having us. Appreciate It. Sherif Mansour a uh, a last in veteran in a good way, in a long history, been working on all the products and you know, look what he's leading today, making a real difference in the world.

So thank you Sharif. Thanks Mitch. We'll be back with more great conversations here from Atlassian.

Team 24 in Europe.