Atlassian's Sherif Mansour On Why Context Will Define The Future Of AI

And most importantly, it does it without adding deployment headaches or complex onboarding. You get things like browser based data loss prevention, SAS access control, and zero trust browsing, but delivered in a way that your team can actually use. So if you've been trying to simplify your stack while improving visibility, please check it out at Nordlayer.com/slash browser.

Damos at work in front of thousands of people, and sharing what the company believes the future of human and agent collaboration could look like inside modern organisations. And this conversation goes much, much deeper than product demos or feature launches. We actually unpack why organizational context may become the defining moat of sorts in enterprise AI and why Atlassian is opening its teamwork graph rather than locking customers into a closed ecosystem.

But what happens when AI starts moving from isolated assistance? into coordinated operational workflows. Well my guest will share with me some of the lessons that he's learned from Atlassian's own transformation journey from on-premise software to cloud and why he sees the rise of Agentic AI.

as another major shift already underway. And we'll bust a few myths along the way around some of those big narratives surrounding AI right now, from fears around replacement and automation So to the reality of human and agent collaboration, governance, trust, and why keeping humans in the loop, yes, it is a buzzword, I know, but this does matter more than ever.

So if you're trying to separate meaningful enterprise AI progress from all the noise out there, then I'm hoping this is a conversation you're gonna want to hear. But enough for me. Let me introduce you to Sharif right now. So thank you for joining me here on the first time. Recording uh live from uh Lassian Team Twenty Six. Now you've been on stage today, natural bantou with uh the CEO there as well. But tell everyone missing a little about who you are and your role at Lassian.

What is my differentiator? Like if I have the same power to intelligence and you have the same power to intelligence, like what's gonna set my business apart and my teams apart? And so our m I guess message to the community really is like your context matters and um your knowledge, your organization, your learnings.

um your values, like your beliefs, your system, um, yeah, all of that matters and that is going to be probably the most sustainable thing in the long term for most businesses. So uh what is your context? Or your context is um whatever is digitized in your world. that you have in a secure and um safe way given access to AI to use.

And so we were then talked about okay, if if context is the biggest uh differentiator, how do we help you grow that context? Like how do you make sure that you have a uh uh organization of learning and continuous learning but also sharing? And how do you make sure that um the data isn't locked off away? And then the second thing we talked about is how you can help use that context to actually accelerate your team's work and actually achieve a business outcome.

And the third thing we talked about is how to make sure that you don't have vendor lock in and you can take the context with you. And so we made some announcements there. We can go through each chapter or whatever you wanna go through or

a personalized way um for a uh AI to use. And typically they'll connect data sources to do that. So we made a bunch of announcements today around new connectors for all the SAS apps customers use, including a new one we made about code intelligence. So you can connect code in where you're trying to connect the business context with the technology context in your organization.

Um so the most practical thing organization leaders need to do is to think about what's my data strategy, how am I thinking about which data do I want to use and how do I make sure that's the right data set up in the right way with the right security protocols. um to use it to enable my team to do that. Because that's really what's gonna give everyone a unique answer, uh, rather than everyone getting the same generic LM response.

CEO of Pendo uh a few weeks ago. And he said he said because of uh the journey that Alassian have been on, it wouldn't surprise me if they went at Fully Argentec within a few years, because they've already learned so many lessons from the previous transformation, but how do you see that at

You know, name it. My code is the most important asset. My juror work item is the most important asset. Like whatever it is, that's never going to the cloud. Governments as well. Um and we were still building the cloud. So I think one of the things we learned back then is that as a vendor, for us to be relevant and continue to be relevant with our customers.

is to help them transition the next wave. And so we need to skate to where the puck set is. And so often we're building ahead of where customers are at. That requires making some bets. Uh unknowing the market really well, understanding the trends, um, and making sure we can be ahead of that. It's just a simple example. It's been nearly two years since we have had agents on the Reverb platform.

Yeah, yeah. I th actually I don't have the ex exact things, but we I'm confident we were one of the very first few vendors to actually have agents in our platform. Um, I was there at the time and worked on it. And uh we had this idea of like, well, what if you could package an L L M call with specific tools and specific knowledge and put it together We need you know, there needs to be a name for that. And we didn't create the word agents, um, you know, et cetera. But like um

Uh, you know, and customers when we announced it, it was uh some customers were like, I don't know about this agent thing. I think it's gonna be a fad. Um as as they should. Like everyone goes through different journeys and our job is to be a little bit ahead of customers. We went through this with Agile as well back in uh for for older listeners such as myself.

Um to be a little bit ahead to help customers during transition um there as well. So um spot on, I I think that the future of teamwork is going to be humans and agents collaborating together. And so we have been making a ton of work to make sure agents are First class to use a buzzword, it's probably a better word, like just hu a agents to be able to operate at the same level of humans in our platform, but ensuring that humans are in the loop.

Uh and so we've been putting agents throughout all our product experiences from assigning juror work items today. You know, the classic example, everyone um Sees how developers work in their consoles and a uh and they can see all their agents, you know, you see the stories of like, I farmed off an agent to do here and I'll come back tomorrow to work on it and another agent here, another agent here. Why do developers get to have all the fun?

Knowledge workers have this today in their JIRAs and their conferences. They can actually farm off multiple agents to go do stuff. switch context, do something else, come back and visualize it on their board and move t move stuff around. And so yeah, we are slowly uh helping our customers transition to a human agent collaboration world where humans are in the loop.

And she works uh at a company that runs data centers and she was responsible for a particular data center and she's in the IT department and her role was ensuring that uh when new procedures came into the data center about installing new racks and that kind of stuff, she had the standard operating procedures. And she came to me telling me how amazing Rovo is and how amazing it was. Um, and I, you know, had a big ego and my head was getting bigger. I'm like, Oh, thank you, we'll take the credit.

Um and you know, I think Rover's pretty good. I work on it all the time. Um and then she shared uh her laptop and the experience with me that she went through and she asked Rover to create some new operating procedures for a particular server mainframe that was changed, went beyond my head a bit. Um but She was like, These results were really specific to my company and it was amazing and I said to her, Oh, let's just check the sources or whatever.

And why was that result amazing? It was because someone else in her organization had already created context that was in their teamwork graph. And she was blown away by it and she didn't know who this person was. There are thousands of people in this organization. And so why is the T-Mograph important? Is that because it helps accelerate actual teamwork in a unique and very specific to your organizational way, not just generic LM responses, right?

Uh and that's why we've opened up the Timo Graph with Temograph.com and customers can use that graph in any tool uh that they like as well as add more things to that graph.

The F one race in Melbourne a few months ago and uh was fortunate enough to go into the uh is it cockpit, is that what it's called? The where they they park. I'm car illiterate by the way. Not always fascinating. Yeah. But I learned so much that week. Um yeah. And watching them m manage the car parts, put the wheels on all that stuff. And when there's a faulty car part,

They turn around and report it in service collections, your service management, and an agent AI agent picks it up, runs through all the engineering context that was existed and all that stuff. So Williams have their car parts modeled as objects in our teamwork graph. So that means someone can say, hey, you know, what happened to this wheel in this race, etcetera? That i Rovo understands what that is.

And so when you hear these physical world stories, um, we have Ford that um capture all their uh test car parts as well there, et cetera. Um Uh so you always hear the like not just IT assets, you know, is typical use of our graph, like oh laptops and we maintain all the computer systems so we have each asset tracked, you know, the b under your laptop is probably an asset number or that kind of thing.

That's typically the hardware that a l uh, you know, your average IT department would track. But when you hear the physical world products, you're like, ah, oh that that's pretty cool. Um, you know. Um so it was exciting to see you know, them using an agentic workflow in in Jura Service Management. that use the context of the graph to help them accelerate their their time to respond to faults on car parts. Um which is pretty cool. Yeah, they're always cool stories.

Um and that language is important. Like I I even s talk to my teams about it'cause some will say, blah, blah, blah, our data. I'm like, No, sorry, it's the wrong language. It's not our data. It's our customers' data. They've trusted us with their data. Um and I think that's the right thing'cause that's the customers IP, it's not our IP. Um yeah, so customers can now go to T Mograph dot com, take the uh rovo MCP or the C L I tool if they're a bit more advanced.

Connect the MCP to you know, I gave a demo this morning, which worked, thankfully. Uh with our friends at Figma, our partners there where I connected Rovo and the Teamograph to uh um uh Jigma make. And I generated a prototype based on a a requirement. And because I used the context of the graph, the prototype was way richer than it would have ever been with my specific requirements across

Like a it was a Google Drive, a conference page, um a sh a SharePoint uh content all in one place. Whereas if I just went to Figma Mac and say, create me a prototype for this thing It would just create with only the access the you know, very basic prototype. But I could speed that up. But again, Pendo, you can connect, you know, rubber MCB to Pendo in your example, et cetera, and use that context in any app.

And the CLI is really for anyone that's building agentic workflows with things like Cloud Core Code, uh um OpenAI's codecs, they can use the C L I uh there as well, or our very own Rovo uh Rovo C L I or Rovo Dev. Um so Meet customers where they're at is a principle and uh and we'll I'm sure we'll be fine as a business if we keep following where c where customers are at.

deployment inside enterprise and I think that's why your demo really stood out today because it brought it to life in a real authentic way and it I think it resonated with everyone'cause it w you could just say, Oh, okay, and and it brought it to life instantly, so kudos there. But From what you're seeing across your customers, where are organisations getting that real measurable value today and and where is the hype still outrunning reality?

I think what is uh been uh high value, very specific uh v um and pointable to is specific agent deployments that are done where for example you may deploy an agent And this agent reviews sales contracts for a particular um set of criteria to make sure the sales contracts meet some criteria.

That is a very measurable outcome. So customers can say for this agent, every time it runs, it saves me ten minutes and therefore, you know, it ran a thousand times last month and that's that's many minutes, et cetera. My mass is terrible. Don't ask me to calculate that math. But y uh so they can do that on our platform and measure the value and that's at a team level. So every single team can quantify the value.

of the agents that they're built or the agents that have worked and that haven't worked and identify what isn't returning the value that is. And so not always, by the way, um the easy thing to talk about here is, oh yeah, agents are all about saving time and saving money and reducing that. Agents also create opportunities. So example, there was a customer the other day that has an agent that identifies new sales leads for the sales team to look after.

So it's connected to external data sources to then say, Hey, we should go reach out to these people as potential new opportunities. So that's an agent that's creating new business or helping create new business. uh for that organization. So that's a very quantifiable uh set of measures there. The last big category is our in our service collection, which are our uh collection for teams who provide service to other teams, typically starts with IT teams.

HR teams who provide an internal service for employee onboarding and that kind of stuff, employee experience platform. Um uh that is a uh very measurable set of agents that are available out of the box to help people deflect work or automate repetitive tasks. You know, no one wants to talk to a human for a password reset.

The agent can do that for you. No one wants you have VPN issues, it's likely possible these things, et cetera. So that that whole uh collection is likely designed around uh very measurable uh service we call it service based work, where someone asks something of one team and that team provides a service to another. Um, often they provide either knowledge to try to answer that question or they'll provide or they'll take action like

I'm gonna grant you access to the system that you requested access to. So they'll so um that's often how we think about it. So th there are multiple different ways to measure value in in those areas.

Well one of the things that really stood out as well from uh one of your presentations was uh when you outlined the maturity curve and how it moves from assistance to orchestration between multiple agents. So what does the final stage actually look like in practice and how close are enterprises to running coordinated AI agents across departments? It seems like the utopia but how close are we from that or are we there?

Um so like, oh, you know, you should change this in this app. So imagine customer feedback for your app. Then you have the agent another agent reading the feedback and then suggesting a specification for what to build, another agent picking it up and building it, another agent Uh building the code, deploying it, testing it, another agent, checking that it works and publishing um some market material.

That whole loop today in Jira, Rovo, Confluence, you can automate that whole loop. It's fascinating to watch. Yeah. You you could just do a very basic thing. After a few loops, it starts creating slop and going down the drain. Yeah. Oh, why is that? Well, turns out if there's no humans in the loop and humans actually providing guidance and taste and making those decisions, you're just like it's just AI feeding AI.

And so I I you know, I I think philosophically, I think that's there to today. Like you could seriously build a system that just loops agents in infinitely to do build something, listen for feedback, suggest the next proposal, build it again, listen for feedback. That I think that's possible today.

Is it all the hype that'll build, you know, you a million dollar business? Absolutely not. It's nowhere near there. Um it actually highlights more and more important why context and humans in the loop uh are key. So how far are we from away from that? I would say we're here today for teams that have maybe smaller context that's manageable for agents to handle there. Um it does feel like it's gonna be a multi year journey before anyone deploys this at any scale.

That's a serious scale. You be if we're talking if your your specific context is like doing that at Crazy Loop. Um and who would want to do that? You know, cloud migration journeys that customers went through to bring it back to the start of the podcast.

Customers that did it well were the ones that cherry picked small teams and tried at first with a few others, then migrated others, etc. And the same thing's happening with agents. I might I'm I know this part of my loop, you know, one example, one thing I'm trying to do with the product teams at Alassian is I want to go from

Uh a project finished to generating draft release notes, that should be an automated agentique flow. And we have all the tools to do that, and some teams are doing that license, and other teams are not doing that.

So I'm like, okay, that part of the flow, you guys can speed that up like crazy and you can actually put you'll get better results. You'll be able to actually write better release notes for our customers to understand what's changed. So we the best way for any customer to build that whole loop is

Piecemeal, building confidence, you'll learn something. And then next one you'll realize, Oh, we shouldn't have done it that way, we should have done it a different way and so forth. So um so it's there today if you want to build a very basic snake game or something like that. But to do it at scale, I think you're really gonna zoom in at one step out of the loop at a time and it'll take it'll take a while.

Practices are actually you know, they always say people people, processes and tools. The thing that's hardest is the people. Yeah. Then the processes, then the tools. And so everything we try to do as an organization, we're often sharing our practices as well on how we're trying to solve some of these problems around

Processes, governance, etcetera. So we're always sharing case studies internally at Lyssian because we're going through this journey too. Every vendor is going through this journey. What has worked for us? What isn't working well for us? What are things you need to be aware of? But at multiple levels we need to be doing this, at the education at the people level, at the process and framework level and at the tooling level.

Um, I don't have a silver bullet answer for you to say that, oh, we have sold it. But the reality is, yes, we need to make sure at the tool level you have as many as much control and transparency as possible and multiple failover levels. So we have we have ways where you can ingest data, for example.

and say, I don't want to ingest this kind of data. But even when the data's ingested, we have another solution that lets you redact data if it shows, you know, if you said I don't want to inject ingest customer addresses. They don't get ingested, uh with a lysine card coming soon. It's not available yet, but it's coming soon. And then even when it comes in

when the data is rendered, um you can say, I d uh if there's anything with a customer address, don't ever render it. So there's multi so what we can help with is a vendor is provide multiple layers of defense, uh if that makes sense. But really the real thing we need to be able to help customers with is

all the way up the stack with the frameworks and processes that we're learning uh and uh education uh at the people level. So don't have a syllable to answer. I think that's just um something that every team and every company needs to be aware of. Um and we need to get better at it as an industry, uh

The reality is if that's the business, the challenge of saying we're just happy to maintain the current pace of growth or whatever it is, the next player out there is probably gonna put you out of business. So the I think the smarter ones are always gonna be thinking about, hey Humans are my my bottleneck and they should be my bottleneck. Like that's what makes me differentiated. If I just got AI to run in my business, I don't know how you would do that. That's another

My mind breaks in that territory. Like what is the what is the business if there's no one deciding what that is, etc.? It's just AI slop, like you'll need to work that out. Um, you know, uh the example you have is one developer is now doing ten things. And the immediate uh uh assumption is, well then I don't need any more developers. I mean like well hang a second, but if you had two developers you can now do twenty things.

Oh wait, yeah, yeah, yeah, I can. And so actually the reality is it's as it's as it's as ambitious as the organization wants to grow. Because your throughput's more, yes. Could you do more with less? Yes, there's no doubt about that.

But has hasn't that just lifted the the the floor for everyone? Like if that makes sense, that's just this has raised the bar of how mu how it's just harder to build a business now. I think I generally think it is. Um Uh and so to do well as a business, you need to be uh making most of all the tools and services you have that will make you differentiated.

And sustainable and part of that is uh working with AI to work out how to do that. But also working out how your humans can help you accelerate use of AI as well. So um yeah, I don't know if I have a very specific thought there other than that. Like it just feels like it's Uh if you're happy to remain where you are as a business, you it'll probably be tough in a few years' time.

Um and I've heard this from many customers actually. It's a it's a department I never thought we'd be talking a lot about. Um It's a very uh if you understand the domain, understand what your cut your company's requirements are when you onboard a new vendor or what the contract requirements are, that is a very um repeatable piece of work.

And so um being able to adopt and set standards for that across your organization, have agents review the first draft of these crazy long documents. It's insane. Until you you should pay pass it back to the human and say, you know, these are the sections you called out as concerning. I have scanned this whole thing. You should work focus on these areas and here's what I can do next is incredibly powerful. I think that's uh been a a pretty uh amazing use of that as well.

Oh I actually got an inst uh another use that's surprising us. What is the th the uh we have a set of agents that always spike at a particular time of year?

All their growth profiles. So, like, hey, this is what it looks like to be this level product manager, designer, comms, analyst, et cetera. There's like a whole bunch of profiles. And so people love to use agents to be like, I worked on these things. It actually Rover already knows that through Timograph.

Help me write the language to frame it appropriately so that when my boss represents me, I'm fairly represented, right? Like as in I want to feel confident that, you know, AI has helped me to do that. Um it's a great use of AI because th like Your performance shouldn't be disadvantaged because you didn't get the right words down, right? Like AI should help you do that, right? That's a good use of AI to do that.

Um and it you know, you you you you do it you you're improving that experience by knowing that hey you actually had a really good tool to help you put the best foot forward for a promotion as an example. So um yeah, fun fact uh uh you know but that's a it's also indicative of when AI is used for personal reasons, people will want to use it more.

Yeah, exactly. Oh, which is great. Like you know, I think I think it's a good thing. So, you know, trying to find ways that AIs can help every individual and every team is something we always think about. Like, how do we make the team the individual feel like a rock star? And if we could do that, that'll help the team feel like a an amazing team, right? So Um so anyway, it's just a a fun fact.

One of my jobs is one of the things I I want to make sure I do a good job as is communicating that excitement enthusiasm that that our customers have here back to our teams.'Cause you may it can only take so many hundred people to this conference and there's thousands of people that work on this. So I'm like

If if you if you're a team member working in a team on something that, you know, got announced and was used, whatever, you'll feel so much more engaged knowing how your thing's being used. People want to build stuff that's used and excited and stuff. So that's probably like my first priority is just

If I can enable them to feel more engaged and excited about their jobs, that helps us uh all feel feel better. The second thing I'll hit them up with is a bunch of asks and changes to our products. So, you know, give'em some encouragement before giving them the next set of roadmaps.

But you know, they they probably already know most of that stuff themselves. It's probably just me slow and catching up to the feedback that they've already heard from customers. Um, so I'll probably d uh do that as well. Probably the main things there. I usually lose my voice by the Friday, so Maybe I'll also just just chill and relax for a few minutes.

And that is how do you organise knowledge? How do teams collaborate? How do you create trusted context? How do you keep people aligned while the technology changes around them? And this idea of context as the long term differentiator, I think, is so important.

Because as I said at the very beginning, if every company gains access to similar AI intelligence and of all using the same tools, then perhaps the real advantage comes from how well organisations structure their Structure, share, govern, and operationalise their own knowledge and workflow.

And I especially appreciated Sharif's honesty around transformation itself. It and some of the most fascinating moments from this conversation I think came from that some of those unexpected use cases, procurement teams.

be uh becoming AI power users, employees using agents to help frame performance reviews. Formula One teams connecting physical car parts into teamwork graph workflows. These are just The kind of real world stories that make that shift feel incredibly tangible rather than theoretical.

But as always, I'd love to hear your thoughts. Are we heading toward a future where context becomes the real competitive advantage in AI? And how prepared are you and your organization for a world where humans and agents increasingly work? Side by side. Let me know your thoughts as always. Techtalksnetwork.com. I thank you for your time today and I invite you to join me again tomorrow. But that's it for today.