Inside Atlassian’s Governed Agent Loops for AI Native SDLC

Hey everyone. Welcome back here to Tectron TV. I'm really happy to introduce you to our next guest.

It's her first time here on Tectron TV, so let's give her a warm welcome. Ming Wu. Ming is head of engineering and Dev AI over at Atlassian.

Ming, welcome to Tectron TV. It's great to have you on here. Great to be here.

My pleasure. So Ming, I told people your title and you're with Atlassian, but you've done more than that in your life, I'm sure. Why don't you, if you don't mind, give people a sense kind of your career path, of your journey that led you to become head of engineering here.

Yeah. I've had probably about 20 years professional experience after I graduated from Michigan State with my PhD degree. So I joined- It's cold up there.

Yeah, it's a tough time, but, well, I survived that one. It's cold up there. Yeah, it is cold.

Well, I love the snow. I was from south part of China. Okay.

It was, I never saw that big snow, so it was different. Then after school, I joined Microsoft at Bing Search. So spent a good amount of years in Microsoft, navigating through different orgs including Search and Office and Azure.

And then couple years ago, I went to GitHub Copilot. So that's where I changed my previous specialty area in natural language processing, search and ranking, to actually the dev tooling. So at that time it's about when ChatGPT, all the GPT models came out.

So that was like, when I see that, I see the opportunity for the generative AI and bringing AI to a different field. So that's where I changed to switch to GitHub Copilot, and then two years ago I joined Atlassian because I see Atlassian has a unique position in the industry where the product line and the ecosystem are quite different. So it's a very good, I would say, platform for actually developing enterprise dev tooling solution.

I love it. Yeah, you're right. You know what?

It's been a very interesting career path, though. It's a great experience. Look, people go to Microsoft, and they get a chance to really jump around a little bit- Mm-hmm ...

to all of the great things Microsoft does, so that's good. Ming, I want to spend a little time on Atlassian. Look, our audience are DevOps people, cloud native engineers, DevOps engineers, platform engineers, cyber security.

They're very familiar with Atlassian. I bet you most of them are probably using Atlassian tools, whether it's Jira or Trello or something else. But this whole AI thing has kind of changed how we're using tools.

It's changing what tools we use, how we use them. Copilot is a great example, right? Think about how Copilot has so integrated into the Microsoft offerings now, and GitHub for that matter.

Let's talk about how Atlassian has changed with AI, right? Because your title is not just head of engineering, but dev AI too, right? AI's right smack in the middle of it here.

How has AI changed Atlassian? I think significantly. Actually, I would say, I wouldn't take all the credit for AI in Atlassian because before me, I think Atlassian already started this journey of adopting AI to their product line and also the, how to say, like bring the internal developer productivity to the level where much higher than before with AI tools.

So there's definitely a huge amount of efforts. I think before I joined, and especially after I joined, the past two years, as everyone knows, industry has progressing so fast. You hear new things all the time and new models.

So, I can share the past two years what I'm seeing because I've been in company for two years. Mm-hmm. Dev AI is org where I come in and I establish an org.

So this is actually a very new org. Before that, there's the AI org and they're working on, you probably heard the robo search. Right.

That's like the AI efforts. So two years ago, the company decided, you know what? The developers are at the center, at the core of the product from the beginning, right?

Like that was for software team. And with all the LM and the dev tooling at the time, with all the LM models coming in, people in the industry seeing a huge opportunity for really kind of focus on the software team because I think that's the area where you can see LM kind of bringing the biggest, at least at this moment, you can see tangible biggest velocity gain, with all the LM models. So the company decided, you know what, we need to actually triple down, double down the focus and investment to help the software team with a solution, AI native solution, that can really significantly improve their velocity.

So it's definitely the very exciting and new efforts there. So the approach, I would say there's two folds. One is internally, we have a separate org just taking care of the internal productivity.

They work with my org pretty closely, where my org is external facing. So what we were doing, my org's mission is facing our customers trying to bring the AI SDLC solution to our enterprise solution so our customer can have a playbook and have a good solution to actually adopt those AI tools. But we're not just bringing the tools to them.

We also apply the learnings from internal. We have 6,000 developers inside of the- Wow ... I think it's a big community.

So all the tools we develop, we actually put into our internal audience first. We call it dogfooding in the earliest stage of fish- Yeah, some people call it eating your own dog food. Other people say drinking your own champagne.

I guess it depends- Exactly ... how you look at it, right? I know, totally.

Fantastic. Yeah. No, totally.

So we have our own people trying it out, giving feedback. We actually have to compete with third party tools as well, because Atlassian, we actually encourage folks who are using different type of tools. That way you get different opinions and comparison, all that.

Yeah. Yeah. Love it.

So this approach actually so far works well for us. So you've been a real catalyst for change there. That's fantastic.

Good stuff. Ming, if you don't mind, I want to start to kind of pivot. Recently, Atlassian announced something they call governed agent loops for AI native SDLC, software development life cycle, and that's a mouthful.

Mm-hmm. For people like you and I in the industry, I know just what you're talking about. People out listening, watching, they may not.

What exactly are we talking about? What does that mean? Yeah.

I think that's a great question. Honestly, the naming is the hardest part I found. In all my technical jobs, you know, like you're finally- Oh, Lord, I know it.

Yeah, totally. So you're resonating with the customer and also makes sense in the long run. So it's a hard job.

So I would say governed agent loops, really the key part landed on a few points. For one is governed. That's one of the very, very big point.

That is one of the core pillar Atlassian always trying to push. We want to bring the customers, give them the visibility, and the capability to control what the agent is doing. That's the governance, right?

Like you have full control on whatever the agent is doing. Agents are not responsible for any kind of thing they deliver. Humans are.

So we need the humans to be there to govern the process. So that's kind of one, the first point. Second, the loops, agent loops, really kind of the core part of that, however you call it.

It's actually about automation and scalability. So the core part of the Agent Loops is kind of one of the automated feature we're going to bring out very soon. So the core part of that is really allow you to define your automated workflow to actually, instead of manually trigger the agent for every job you see, we set the conditions and allow the agent to perform on the big batch of jobs.

But also you have to set guardrail and enforcement, so making sure the agent delivers things within your expectation, not something unexpected. And we also, Jira as the center of the control plane, you can see everything happening. You have control on everything, and the traceability, visibility are all there.

So it's really landed on this two point, I would say, however you call it. Excellent. Now, is this built into Jira, basically, or is it a standalone thing?

Or you've got, I mean, do you need to be- Very correct, yeah. What, yeah. Yes.

Our solution, I would say, in general solution, we want to develop this platform capability. So in general, Atlassian, as you know, the product portfolio is pretty large. There's Jira and Confluence as the bigger, so Jira is the flagship product and Confluence.

Okay. And there's a tail of the others, and it's a very big family. So our philosophy and the solution here is we do want to develop platform-level capability in the agent that can actually handle certain type of the flows.

For example, taking a task, perform the job, and do the PR reviews and along the SDLC stages, we want to actually have the agent capability to bring the task from the beginning to finish. And we set a foundation. Then we have to look at the product surface.

Okay, so for the Agent Loops right now, we see Jira as the best product surface to bring that capability in, because Jira is where your system of record is. People set up, it already has the previous automation feature, which where you can set a flow workflow. It's not agentic, but there's also all the boards and the different configuration.

So Jira, in many sense, people use it a lot. So that is kind of meaningful surface for us to bring the capability in. Maybe in the future we'll see, maybe, you never know, Loom or Confluence people start doing tasks there, that as the entry point.

We could actually bring the capability there as well. But right now, Jira is the natural. What we see is the very good centralized control plan for our people to see all their tasks at the team level or org level.

Yeah. So, Ming, I've been covering Atlassian here at Techstrong for a long time, right? One of the strengths of Atlassian was always the partners, the ecosystem, right?

And Atlassian, Jira, as you said, is the flagship. So many solutions integrate with Jira. How does these governed agent loops and the AI stuff, do the partners also integrate with that or standalone kind of thing?

Partner, you mean the, not the customer, but the partner- Not the end user. I mean the partners who have third-party apps that work in Jira and so forth. Yeah.

No, actually, that's a really good question. So I think our strategy is we do want to allow the third-party partners work with us on the platform to provide a very meaningful solution for our customer. At the end of the day, what we want to do is really bring the customer value.

So whatever pieces in the solution that works for our customer, we want to enable it. Third-party partners is one of the important piece on that. So, for Jira features, we're rolling out last month, the last couple months already announced, I think, the support for our 3P coding agent.

So we do allow people bringing in their agent. We actually share the important thing, not just letting them in to bring the third-party features on our platform. We also share the context layer.

So all this, even though it's from our partner, but the Atlassian's advantage of TWC is the context layer that actually would allow all the agent, regardless 1P and 3P, have the shared understanding of your team or business intent, all that. So that is actually a very important foundation, to allow us enable 3P or 1P. It doesn't matter what tool we bring in.

What's important is the shared context and knowledge, what you need to do and how you should be doing that. That is shared understanding. Everyone has it.

All the tools has it. So that's kind of Atlassian's philosophy on the solution, a strategy there to... At the end of the day, customer is the most important thing we need to serve.

We need to bring the valuable solution there. Absolutely. Ming, let's talk a little bit about customers, right?

Look, AI has changed things, and it's changing things every day, every week, week to week. Right? We hear such different, this week everyone's using it.

Next week, everyone's afraid of it. The week after, we love it. It's like a soap opera.

But from where you sit, you're hearing, you're dealing with real-life customers, thousands, tens of thousands of customers. Where are you seeing kind of the rubber meet the road? Right?

How would you describe the adoption curve, let's say, for AI in this with end user customers now? I guess you refer to the spectrum of how the adoption look like? Yeah.

Because you see it, you really do see a spectrum, right? Yeah. I would say, it's definitely, they're not the same.

Some companies, especially in the software industry, by nature, with developers, they're more exploratory, and some of the companies already tried a lot of tools, and they already know a lot. So there's definitely advanced customers. Like early adopters.

Early adopters, and they're great, and we actually learn a lot from them, how they use it. And the interesting thing is every company, they are so different in terms of the workflow. So the type of customization and the features they care and it's very different across.

Obviously, there's more traditional industry companies. They're in different business than the software. Every company has a software team and like this, supporting their internal IT, and they also trying to get on this AI adoption.

So I do see a very different spectrum of the AI adoption. But I would say over the past two years, though, in general, everybody has moved big step forward, just within two years and from all the customers I've seen. " Now they actually have pretty customized ask.

How can I bring it to a more scalable... I think scalability is one thing people always ask, how can I scale this, right? How can I see real velocity gain rather than I use AI, but how much I really gain?

The visibility to your ROI, that's actually one of the big thing, the industry is getting to that. DX from the Atlassian bring you the reports and they help you understand where your money is spent. Does it actually make a sense?

Did you get anything back from the money? So those are the areas I think people will start looking into more and more. They're just like, "We want AI," but it's no different from any other tools you were using before.

You want to see the gain from that, right? Absolutely. I get it.

And that's an important message, too, right? Ming, at the end of the day, it's just another tool. It's a very powerful tool with all kinds of potential, but it's another tool, and sometimes I think we lose sight of that, right?

We get so caught up in the media and the spectacle of this. Yeah. Yeah.

That we forget right now, especially if there's one industry where AI has really kind of gone out ahead, it's in IT tools, right? Coding and testing coding and deployment and all of these things. And so within the IT industry, we probably have a different view of it than, let's say, someone in accounting or dentistry or something like that, right?

" You never know. We've heard these. What are you going to do?

I didn't say anything. " I didn't want her to mess my teeth up. But anyway, Ming, I want to thank you for coming here on Techstrong TV.

This is great. com. Is there any place within the website they should be focused on?

com is probably the biggest thing. For our customers, we do send out email notification for all the new changes, and we will be doing, I believe, Mia can correct me, we're going to do a bunch of the marketing campaign to bring awareness to our AI native SDLC solution. So you'll be hearing from us a lot.

Good. Hey, don't be a stranger here to Techstrong TV. Come back and keep us posted, okay?

Definitely, yeah. Very nice talking to you. Thank you.

Nice speaking to you. Ming Wu, head of engineering, and not only head of Dev AI, but she actually started the Dev AI program there at Atlassian. Fantastic.

We're going to take a break here on Techstrong TV. We'll be back with more in just a little bit.