The AI-Enabled Workforce | Atlassian President, Anu Bharadwaj

Hi Julie, thank you so much for having me. It's a pleasure to be here.

The most transformative technology shift that I have seen in the 20 years of my career, the kind of advances we hear about on a weekly, almost daily basis over the past couple of years has been really fascinating for a technologist like me. Why is this accelerating? I think the promise of AI, for those of us who were around for a long time, we remember AI in the '90s was really big and I remember starting off my career at Microsoft building genetic algorithms, which was very much supposed to herald the era of AI. But what caused the acceleration is the increase in computational power that's available to us over the past decade, the explosion of data that's available thanks to cloud computing and multiple software as a service offerings that are around, and just the sort of data exhaust coming from a lot of our daily workflows. All of this has culminated in finally AI becoming democratized to a lot of us, even those of us that are not necessarily technical people. You see tools like ChatGPT are now making AI democratized and accessible to a lot of non-technical people, to a lot of lay people who can use it like a consumer tool. I think this is particularly exciting for those of us who are in technology early on, so it's an exciting time to be in technology.

That's a great question, and it's a question whose answer has evolved over the past year or so. If you asked me the same question last year, my answer would probably have been different, just as a testament to the pace of change. At Atlassian, we are a collaboration company, so day in and day out, the thing we think about is how do we make teams work better together? Let's look at just individual productivity as an example. The way I use AI for individual productivity is things like doing research, looking up a person that I haven't met before, getting introduced to for the first time, getting some context about them—what are their interests, what are their current pieces of work that are relevant and overlapping between my work and their work. Using AI tools for personal productivity tasks like building simple automations that help me synchronize across calendars, set reminders, and take actions based on my schedule in the workday, those sorts of things. But the thing that will not change even in an AI-powered world is the element of human creativity, the element of human collaboration. There is always going to be a need for teams to come together to do great things; that's what makes humanity special too. So for how do we make teams more productive with AI? Specifically, the areas where we are seeing productivity improvements manifest right now are in the areas of coding. You've heard a lot about code generation tools that help developers automate some of the more simple, routine tasks, planning tasks in their work. How AI can actually help take a problem statement and break it down into smaller steps has been phenomenal. At Atlassian we have managed to serve over 50% of the tasks raised by employees that have been intercepted by virtual AI-powered agents. This has saved us thousands of hours, thousands of man-days overall, and made our human agents a lot more productive, freeing up their time to do creative tasks. Also, how does AI help people understand who is a domain expert in a given area, who to reach out to when they are stuck with a problem? The promise of how we can use AI to supercharge teams is really high, and it's an exciting evolving area that we are seeing.

Yeah, so specifically for us at Atlassian, we have about 12,000 employees spread across 11 countries. Even pre-pandemic, any global multinational company is by definition doing distributed work. In that kind of a distributed setting, how do you make collaboration simple? How do you unite people behind a common shared purpose? How do you get them to understand the company strategy and how the work they do contributes to that? How do you get them to understand who to contact in certain situations? With AI, you can really bring a lot of those latent connections between people and collaboration between teams to the forefront. A tangible example is we have a tool called Loom, which allows us to build asynchronous videos. You can record a video message, share it with a bunch of people, and they can post reactions. Think of it like Enterprise TikTok. I use Loom a whole lot across Atlassian; I post a weekly video update. It's just a short video that says, 'Here are the customers I met this week, kind of an update in the world of Anu.' And Loom has AI-powered capabilities on top of the video that bring an element of human connection. With distributed work, the challenge is you often end up in virtual conversations, so text doesn't quite convey the humanity of the person or the mood of the person. What somebody says with a smile on their face is very different than what somebody says with a scowl. Loom, especially with the AI-powered capabilities we've added, helps in that kind of distributed setting. More and more AI capabilities will help bring teams together in distributed settings.

Cloud computing democratized the number of players who could come and build a product and have customers use their product. So we went from a world where there were a few big companies providing software to now thousands of companies that went public, thousands that built multi-million dollar businesses offering digital services, accelerating the digital transformation of pretty much every company in multiple industries. As a result, there was a lot of what you would call data exhaust coming from everything we do—every transaction, every interaction we have. Previously, human attention was still a constraint on top of the volume of content generated, but now with generative AI, content creation has become much simpler, as you've seen with Sora and some of the multimodal models where you can produce audio, video, text, multiple kinds of content. So it definitely feels like an acceleration in the different kinds of content and volume of content available. But AI also presents a solution to that in its own self. For example, one of the things you can do with AI is feed it a bunch of content and say, 'Give me a summary of this content' or 'Tell me just the top points.' So a lot of people can make use of it. A good tangible example is in one of our Atlassian products called Confluence, which is a knowledge management tool. We've introduced features that say when you're looking at a document, there is a little magic button called Summarize that can look through the entire PDF or document and give you a three-sentence summary. Domino's Pizza is one of our customers for Atlassian Intelligence, and I was talking to the customer when they said, 'We can now get pinpoint summaries of huge pieces of content, which has really been helpful in understanding the gist of what we need to do and what action we need to take thereafter.' So content management has become a lot simpler. Even within the use of generative AI tools for content creation, AI can help change the tone, the tenor, the content such that it's adaptive and suitable to the situation. So while AI accelerates the problem of content proliferation, it also offers a bunch of different solutions that are very helpful for people worldwide.

That's definitely an interesting question to ponder for all of us, irrespective of the industry we are in. You see that show up in the hiring market. However, for companies at large, I think the trends will be driven by a singular belief that AI is going to be additive—it's going to help augment human productivity—rather than AI is going to automate a bunch of different things end to end that we're doing today. I think we are not yet at that point. So the way to think about it in my mind is really, what are the roles that AI can help support and can help unleash more productivity in, so that the humans in those roles are able to focus on some of the more creative aspects of their work. Coming back to the earlier question about areas where AI productivity manifests already—things like customer support, software development, automation of test cases—there are a few areas where the tools available are really helpful. Companies need to think about upskilling employees in terms of how to use those tools, provide training and enablement and education around using them in their day-to-day work. Second, I think as AI helps automate some of these routine pieces of work, we'll also see new roles emerge. There are never enough things to do; people are always overwhelmed, always overworked. The promise of being able to use some of the AI technologies to help deal with that overwhelm, to make automation more prevalent and useful, I think will automatically help with the kind of skills that employees will value and need in the future.

That's a great question. We are kind of at a point in the hype cycle where I think last year it felt like everything was going to be somewhat AI-washed, so to speak, but this year we have a lot more clarity. We've had close to 18 to 24 months of technology advances, and now the time has come for us to really think about what can be applied and productivity unlocked versus what is a cool new shiny thing to explore. For leaders, it is very important to understand what are actionable, useful applications of AI that you can use in your own company. For example, use cases like personalizing demos of products, creating marketing content, and internal use cases like handling internal HR questions or laptop policy and IT policy questions, handling employee onboarding—that's a use case we use internally a whole lot. We also have a lot of customers using our products. Recently I heard from a customer that they cut down their employee onboarding time using some of the AI capabilities from 4 months to 1.4 months, which is a pretty dramatic productivity gain. So it's useful to identify what are the specific use cases and workflows where you can apply this. The second important thing is to set responsible guidelines in terms of building AI products, and for those of us deploying AI products we buy from a vendor, that vendor should be trustworthy, reputable, and we understand their philosophy and stance around AI responsible use. For example, at Atlassian we have put out a responsible technology template that we use ourselves and help customers use if they are building AI products themselves. AI is only as good as the data that the AI product is able to access, and it's important to be cautious and protective about the data you have and who you give access to. So understanding privacy and security is another important thing. Finally, understanding the extent to which you can deploy some of this AI technology for your use cases, separating hype from reality, from my experience, a lot of the productivity improvements come from deploying even just partial automation of some of these workflows. You don't have to take an entire end-to-end job and try to replace it with one AI agent—I don't think that's a particularly helpful approach—versus being able to say here are use cases that the AI agent is useful for, and we will place the AI tool in the hands of employees who are actually in the midst of that workflow, mostly creating collaboration between AI and humans. I think that's really the most productive edge of AI.

That's a great thing to think through, especially in the current context of the pressure to deploy AI across organizations because you hear every day about companies who talk about operational savings and cost savings by deploying AI technology. Being completely out of that loop and not aware of the technologies available for you to use is no longer an option. A few places where I've seen tech leaders struggle with deploying AI across their organizations: one is understanding who is trustworthy and who is not. I would strongly advise deploying products and tools from vendors you trust, who are reputable, who you've had a relationship with, and you understand their track record and how they've built products in the past. Be very clear about their data usage policy—are they using your data to train their model? How exactly will your data get exposed? Second, there is a fear of 'Will AI take all jobs in my team? Will it replace a bunch of different people on my team?' As with any new technology, there is always a sense of fear of what this means for the future. Given the level of uncertainty associated with any fast-moving technology, this is a natural human reaction. But it's helpful to think about how we start with augmenting our teams with the power of what's available with AI-enabled tools, versus complete replacement. A third area that I think is also a bit of an impediment is understanding the specific areas where AI is especially helpful and some areas where you still need human involvement—human creativity, human coordination, advanced reasoning. Being able to make the distinction between where it is productive and where it is not, and really encouraging employees to try this out on their own, providing them the training and skills, and enabling them to make responsible choices themselves, I think is the most scalable way for leaders to stay on the bleeding edge of the curve while protecting themselves from some of the potential harmful ramifications.

From the perspective of technology builders who are building this for our own customers, the single most important thing is to put these features and products out to our customers and get user feedback early. In our own case at Atlassian, last year we announced what we call Atlassian Intelligence, which is a layer of intelligence across the data collected by 14 different Atlassian products, helping our users unlock insights based on their own data. There were some use cases that were really helpful and some that were not as helpful. As we ship these features, we haven't yet realized the full promise of what AI can be. So doing quick feedback loops and understanding from users what the real-world value of these scenarios and how users are using these capabilities is critical. It helps distinguish the hype from reality because product analytics and usage data don't lie, so you can compare reality with what your aspiration as a product builder was. A second lens for consumers to distinguish between hype and reality is to really use those products in the context of somewhat analogous data to what you have. Vendors will always show you the best possible light, but when you bring it to your own workflow and try using it inside your own use cases, it should make sense. The amount of effort required to deploy and use it should actually be far lower than what you're expending right now. I think that separates the wheat from the chaff. In my experience, having used a lot of these tools almost daily, I can clearly see that well over half of it is really a lot of hype, some of it is early on, and very few are really at the point where when I deploy it, it offers some real productivity gains that are game-changing, like the virtual agent example I mentioned earlier.

I think that's an important question for our times. It's a generational shift that we're seeing with artificial intelligence that is not only intellectually interesting to think about but also behooves us in terms of responsibility and our ethical duties to think about what this means for humanity, not just for individual productivity or team productivity or companies, but what does it mean for humanity? It's a turn of technology that is so foundational to the way we live and work. There is a lot of debate—I'm sure you've heard it—about when artificial general intelligence is going to come forth. We've been wrong in the past, but now it feels like we've built upon multiple foundations that it feels pretty real that we are at a moment where perhaps autonomous AI might be within grasp. But even without getting to AGI, there is a lot of uncertainty associated with that, and there are multiple philosophical takes you can take on that. But even before that, what is certain is that this is transformative technology for all of us, irrespective of which industry we are in or whether we're building consumer software or business-to-business software. The ability to take an idea from a thought in your head to a reality in the universe just became available to a lot more people than before. I am a computer science engineer, so earlier you had to go through technical training, teach yourself how to code, really build something that was a thought in your head into an application out there in the world. But AI really helps make that possible for a lot of people without being technically trained. Throw that in with robotics, and it makes it also possible to bring your creations to life in the physical world. This is like magic that was available only to a small group of people before. I think that's the exciting part of the technology, which is balanced with the fact that it is important for those of us involved in the creation of these tools and capabilities to make sure we do it in the most responsible, most ethical way possible, so that we don't build biases and amplify biases in these products, and we are thoughtful about what this manifests as in our technologies.

Thank you so much for having me, Julie. It's been a pleasure.