Episode 291: How Atlassian Connects Context, People and Work to Create AI Value (with Avani Prabhakar)
“The technology itself is no longer the competitive advantage. Your organisational context is.”That is a central message from Atlassian’s Leading with Context report. Its State of Teams research found that while 85% of knowledge workers use AI, only 6% of executives can point to clear organisation-wide ROI.In this episode of the Digital HR Leaders podcast, David Green is joined by Avani Prabhakar, Chief People and AI Enablement Officer at Atlassian, to explore how the company is working to close that gap.

“The technology itself is no longer the competitive advantage. Your organisational context is.”
That is a central message from Atlassian’s Leading with Context report. Its State of Teams research found that while 85% of knowledge workers use AI, only 6% of executives can point to clear organisation-wide ROI.
In this episode of the Digital HR Leaders podcast, David Green is joined by Avani Prabhakar, Chief People and AI Enablement Officer at Atlassian, to explore how the company is working to close that gap.
With a remit spanning HR, internal IT, data science and research, customer engineering and customer tech support, Avani shares practical lessons on connecting organisational knowledge, redesigning work and helping teams build new capabilities.
Join them as they discuss:
· Why Atlassian brought people and AI enablement together, and how it is testing changes to the HR operating model
· What organisational context means, and how connecting work, knowledge and communication makes AI more useful
· How leadership behaviour, experienced AI users and regular experimentation build confidence, and why usage should not determine performance ratings
· What unsuccessful experiments revealed about human relationships, deep professional expertise and the value of graduate talent
· How Atlassian approaches tasks, skills and roles, and is exploring workforce capacity across people and AI
· How the People Insights team is redesigning its own work, and how leaders can assess AI progress through capability, business outcomes and governance
This episode is sponsored by Eightfold.
Workforce planning is only as good as the insight behind it.
Eightfold’s talent intelligence platform draws on career data at a global scale to help HR leaders understand how skills are changing, spot the capabilities their organisation will need next, and make more confident decisions about hiring, development and internal mobility.
Find out more at: eightfold.ai
Resources:
· Connect with Avani Prabhakar on LinkedIn
· Leading with Context: Lessons from Atlassian’s AI Journey
This episode of the Digital HR Leaders podcast is brought to you by Eightfold.
[00:09] David Green: AI adoption is rising, and employees are reporting that they're getting more done, but those individual gains aren't necessarily translating into better results across the organisation. Atlassian's Leading With Context report highlights the gap. While 85% of knowledge workers use AI, only 6% of executives can point to clear organisation-wide return on investment. My guest today, Avani Prabhakar, Chief People and AI Enablement Officer at Atlassian, has been working to understand and close that gap. Alongside HR, her remit includes internal IT, data science and research, customer engineering, and customer tech support. Bringing those responsibilities together reflects Atlassian's focus on changing how people work, collaborate, and make decisions with AI. Avani argues that organisational context is essential to that effort. Connecting the systems where work happens, the knowledge in documents, and the conversations behind decisions helps AI produce more useful, accurate results. We explore what that requires in practice from documenting decisions to redesigning critical workflows. We also discuss why leaders need to demonstrate their own learning, and why linking AI usage to performance assessments can encourage the wrong behaviours. And for people analytics listeners, Avani shares how Atlassian's People Insights team is building agents and rethinking its own work, alongside a broader discussion about HR's role in managing capacity across people and AI.
There's plenty here for anyone looking to take a more active role in their organisation's AI transformation. So, without further ado, let's get into it with an introduction from Avani herself.
Avani, welcome to the Digital HR Leaders podcast. Absolutely delighted to have you on the show. To start, please could you give listeners a quick sense of your background and current role as Chief People and AI Enablement Officer at Atlassian, and what kind of sits within your remit today?
[02:19] Avani Prabvhakar: Hi, David. Thank you. I'm really excited to be here. I'm a big fan of all the work you've been doing. So, about me. So, my background, I've always been in HR as a craft, but before joining HR, I did spend time in the business and then moved into HR. This was back in 2000 when I was working with GE. And Atlassian, I've been here for the past, seven-and-a-half, eight years right now. I have always been in the HR space, but working closely with both the founders in terms of all things people and strategy. In my current remit that I have, I've got HR, IT, internal customer engineering, as we call it, data science and data research, and customer tech support. So, that's my current remit. And prior to this, within my HR remit, I had facilities, real estate, teamwork lab, and everything that falls under HR. So, that's my existing remit.
[03:11] David Green: Perfect. And as you just explained there, Avani, your role has expanded recently, I think. So, from leading HR, as you said, you've brought in internal IT, data science, customer tech support, customer engineering I believe as well. We've seen this as a bit of a trend in certain organisations. I'd love to hear, why did Atlassian decide to put responsibility for people and AI enablement under one leader; and what has that changed about how you, as a leader, lead transformation?
[03:44] Avani Prabvhakar: So, for us, we've been on this journey for quite some time. And it's good to get some context. So, even before we made this org change organically, the internal OKR or goal about AI transformation was co-led by our CTO, CIO, CISO and myself, four of us kind of working on the three things together, starting with responsible tech, of course, core partnership with the CISO organisation there in terms of safety, security, governance, that was the basic foundation. Second came all the critical workflows, so partnering very closely with the CIO and CTO, both in terms of the technology to deploy and what workflows to look at. And then, of course, the third one was in terms of how do you drive the cultural transformation, which was primarily led by my team. What we realised over, I would say, after 12 to 18 months as we were on this journey, to say like, AI adoption was not a challenge; deploying a lot of AI technology wasn't a challenge. Where everyone was kind of getting stuck was like, how do you really redesign how the work happens; how do you redesign and change people's behaviour; and how do you use AI, because it's such an intuitive technology that people use it very differently. It's not one size fits all, "Here is a new ERP you deploy". So, I think it's a very different playbook than what we have seen in the past when we were doing digital transformation, so to speak.
So, for us, the main pivot was we were already working together, and when the shift happened, it was very organic, because the teams were working closely for so long and they were like, "Yeah, this makes sense". So, for us, it didn't feel like, "Why are we making such a big shift?" But I think what led to that change was that, for example, you deployed Claude, right, inside a particular team and everyone is using it, adoption is high. We observed things, like personal productivity was really going up. People would personally say, like 89% of individual people will say, "I'm getting productivity gains". But then, when you look at the teams, at a team level as an organisational level, only 6% efficiencies were really being delivered by everyone using AI and adopting it really well. And we did a lot of this research internally and externally with our teamwork lab to really study, in terms of what is coming in where. And that was one of the reasons why we made the shift to also make a statement to say like, "Hey, it's a people-led transformation. Let's try and create faster learning loops and solve that problem versus technology problem".
[05:59] David Green: That's really good. And we're going to get into depth into some of the research that you've published and you've co-authored recently in a minute, Avani. One of the things that I got from speaking to you previously is your kind of belief and purpose around HR. And obviously, as you said, you've been in HR. I understand you've got five core HR beliefs, and I think that could be quite inspirational, I think, for some of our listeners to hear those, if you'd like to go through those?
[06:29] Avani Prabvhakar: So, before I talk about HR beliefs, this is one thing we are doing with each and every craft inside the building. So, we will go and ask a CRO and we'll go and ask whoever is leading that craft to say, "What are your core beliefs that you think are going to be true with all things AI coming in? So, we did the same thing for the HR team and we did it in a workshop fashion. Some of the core beliefs we have is, in the age of AI, there will be one pivotal role that will come up which will be more around, call it AI human capital management, right, in terms of what is the right mix. Currently, people think about that as humans in the mix, but what will that look like when you have tokens in the mix? I'm not saying agents in the mix, but the real currency will be token versus human cost spend. So, that will become one of the core capabilities within the HR team. And our belief is it'll rightly sit within people space, not finance, not IT, for all the reasons that I've outlined before with you. So, that's one core belief.
Second belief, we also feel like the current model that we have around HR, which is COEs and the standard model that we've been using, is not going to be relevant in the age of AI, and we are experimenting heavily internally to prove that point. So, for example, you don't need a talent COE, you don't need a compensation COE, you don't need the COE experts to be servicing and helping an HRBP who works with the customer. In fact, I think the roles are already getting blurred. And we have done some experimentation internally where you need to find out more AI-forward people within your team who work really well with the technology, who understand what skills to be used for what agents, and give them a client group. It doesn't matter if they come from the HRBP experience or not. And then, experiment with them, and you will see, if you find the right ways of working on how to engage, you will not need this current model that you have. It'll be a more fluid model where AI will be able to give you most of the SME expertise, but you need someone with a very different skillset who can combine it all and go to the business. That was, I think, one of the controversial ones, because everyone is still very married in terms of how HR is organising.
[08:41] David Green: Yeah, well let's start with those two. I'm sure the other three come out through the rest of our conversation, Avani, because I really want to talk about the research that you co-authored, the Leading With Context report, which came out a couple of months ago. As you probably know, I share every month a collection of resources that particularly inspired me, I guess, from the previous month, and that was one of the reports that I shared. And the response to it has been overwhelmingly positive. I'm sure you've seen that as well. And it provides lessons from Atlassian's own AI journey. Again, listeners, if you haven't seen it, we'll put a link in the show notes. And one line in the report really stood out to me, and to quote, "The technology itself is no longer the competitive advantage; your organisational context is". Love for you to kind of walk through, what does that mean in practice; and why is context becoming so important to business value?
[09:41] Avani Prabvhakar: So, I'm glad you asked, because the way I've been working with some of our peers and customers in terms helping them through, I always tell everyone to start from context, then focus on workflows, and then culture. That's how you should go about it. And I'll probably go deeper, what do we mean by context? Because it's been overused, I guess, for the last six months. So, let me take you through it. In any organisation, you will have probably about six or seven different types of contexts, and I can take you through what I mean. So, when I talk about six different contexts, one context would be where the work happens. So, there's a lot of work context. So, if you're a salesperson, all your work is happening in Salesforce. If you're a software developer, it's happening in Jira. So, I think there's a lot of work context in terms of what is happening over there. You move to the next thing, then there's a knowledge context. So, all the work is happening there, but where do you really record all the knowledge that is sitting inside the organisation, which it will maybe be Google Docs, Confluence Pages, or whatever is your tool that you do. So, there's a lot of knowledge repository that is there in every function that you go into.
Third one, which is the most critical context for me, is communication context. So, all the actual communication happens on a Slack, on an email, or on a Zoom call, or in a team meeting, or a call like this, and that is a very important context right? And then, their assets. So, based on whatever business you are in, there are a lot of assets which you manage, based on whatever industry that you're in. So, that's one context. And then, there is context around code in terms of where the actual code and how you're building your products, that kind of decides. The reason I'm taking you through all this journey about different context is if you don't connect these into one another, so for example, if I'm doing all my work in Salesforce and if all my knowledge is being written down in a Confluence page, all the communication around a particular deal that I'm working on is happening on a Slack message. If these three contexts do not come together, you're not really getting the magic out of AI.
Right now, it is like if you give a command and if you haven't built this interconnected context layer, where everything is connected, and if you give a command to Claude to say like, "Hey, I need to do this code review, and can you just help me?" and we ran this experiment side by side internally. What we found was if you don't have a context layer, we call it teamwork graph, you can call it something else, whatever context layer you build, when you connect all of these, where the work is happening, where the communication is happening, where the knowledge is restored, if you connect them well, and then you send your AI to go off doing certain things, it does not sprawl in your billions of objects which are not interconnected, which means it takes way much longer time when it has to go and find, because the context layer isn't built out. It takes more time, which means it's burning more tokens. In our case, we felt we were able to save about 48% tokens because the time this thing was there. The response that you get is much more accurate, so it does not start hallucinating because it's not having to go and sprawl in that.
So, I think I'm getting a little bit technical, but just trying to give you in terms of creating that context layer is the mode, and that is the main thing where the real unlock for AI happens, rather than just deploying tools endlessly to say, "Yes, it is connected, we have built connectors with all the MCP behind the scene and it is pulling the data". It doesn't really give you the right outcome. So, for us therefore, I've been saying context is the real mode here, which is to help you use your AI very differently. If you haven't built that, I feel like you will have to redo the whole cycle. And a lot of other companies are getting to that realisation, and now they are investing in building the context layer.
[13:18] David Green: Every plan assumes a future you can predict. None of them can. Hire ahead of it? For what skills? There's no software for that. Until now. Eightfold AI holds 1.6 billion career trajectories, intelligence that shows how skills evolved before you need them. The plan changes, the workforce changes with it. Build the workforce your future demands with Eightfold AI. Talent, intelligence, human-led. Visit eightfold.ai.
What have you learned about making that knowledge usable as well so that AI understands not just what happened, but the decisions and the relationship and the goals behind the work? This talks to your earlier point, I guess, where, where do you move from personal productivity to team productivity, I guess?
[14:22] Avani Prabvhakar: So, here comes in terms of how do you need to redesign the work and redesign the ways of working, right? So, for example, AI works well because it can read a lot of content right. So, do you have a written culture inside the company or do you have a presentation culture? If you're only creating decks and you're only presenting it, it doesn't give you rich context for AI to go and read. So, a few things I would say for this knowledge to be usable, try and pivot ways of working in terms of start writing a lot more things if you can, or transcribing a lot of things. If you and I are in a meeting and if it can be transcribed and come in a written format, it's to write this thing. So, you should rethink the ways of working, do you have a written culture; do you write everything? That is one.
Secondly, how does work happen? Do you have a system where you can tag how the workflow is happening, in terms of giving a status tag to every meeting that happens and create a goal attached to it, which is tracked in a system, because then AI can see this is the system of work, how the work is flowing, for every project that you've created. Things like 'open by default', like are you an organisation where the information is open by default? Do you create documents which are default open, and then you lock them only if you need to, or it's the other way round; everything is locked and then you open it when you need to? It's a very slight shift, but open by default, you can ask every other company who's working on it to say like, "Hey, if AI cannot access and you don't have an open-by-default culture per se, it blocks". So, we have a list of things in terms of what are the ways of working, which gets into like really next level of details, which helps you kind of make sure that you are AI-ready in so many ways.
[16:01] David Green: And I think what I'm really hearing there is that openness and having those documents open, so that AI can use it and then deliver that productivity and those business outcomes and presumably the employee experience as well. You've talked about culture there, and then obviously there's that trust element that absolutely underpins everything. And I guess in your role, part of your responsibility as Chief People Officer is to make sure that that culture of trust is there so that actually happens?
[16:30] Avani Prabvhakar: Yes, and while we do all of that, not forgetting the permissions, it's not open by default and everyone can see everything inside the company, but you have the right perms that you've already logged into the system. Second thing I would say is just a very simple example. A lot of companies have company-level OKRs and goals. There is no system that can break it down to the next level, down to an entry level to say, "Hey, this is what the casket looks like", in a systemic way, or they're using a system to do that. Once you break down the company-level goals, you break down who's working on what in terms of every project and every workflow is captured. So, it's a big change in ways of working. Usually, all the work right now happens either on emails or meetings. So, people are either writing emails or attending meetings. So, it is a shift in terms of, what tools you're using, how the working practices need to change, for AI to be a part of it.
[17:23] David Green: Well, we're going to talk a bit more about culture in a few moments, but the research that Leading With Context found, that 85% of knowledge workers use AI, but only 6% of executives can point to organisation-wide ROI, you talk about ROI in the report. Where is that gap between activity and impact coming from?
[17:48] Avani Prabvhakar: I would say one of the biggest gaps that we are seeing is, one is the context layer. Like, if you haven't built that layer, then definitely you're spending way more tokens and not getting the ROI. So, that's the first thing when it comes to ROI. The second thing which we have found in the ROI is, if you haven't really done any top-down prioritisation of what workflows are really business-critical where you're going to be deploying it, and if you've just kind of deployed AI like another tool to say like, "Hey, let's drive adoption", I think that's where you get into the trap of like, "Hey, we've deployed it, no ROI, we don't even know how people are using it", and I would say, there's no responsible innovation that is taking place in that environment. So, one is context layer.
Second thing, I would be really identifying and prioritising the critical business workflows and kind of redesigning the workflow. Start from there. Don't think about, "Can I use AI to make this better and drive this much efficiency?" I would say the starting point should be, "This is the most difficult problem we need to go and solve. How do we do it with AI?" It's a very different approach. To give an example, like employee onboarding inside Atlassian, I mean onboarding wherever you go has always been one employee experience everyone wants to really enhance. And we didn't go in a traditional way to say like, "Let's map the onboarding workflow and see where we can reduce this step and save the time". We did it the other way around to say like, "How do we fix the employee experience in this one?" and the team went and created an agent. It's called new on New-hire Onboarding Rovo Agent, Nora. Because it was created by the team, first of all, it was created by HR team, so there was no one from it or anyone who was involved. So, it gave a lot of confidence in the team to say like, "Hey, I can build my own skills, my agent can write my own skills, they have fixed all the perms". So, it gave a lot of confidence and pride. But what it did was, it has been built in a very intuitive fashion and very personalised, based on which employee kind of works in. So, as you enter Atlassian, that's the agent you speak to and they do your onboarding end to end for the next 90 days.
The third thing I would say when you're talking about the ROI, and I was talking about responsible innovation, we have something called AI wallet. So, that's one thing I would advise everyone who's on this journey to do is, build the AI gateway, build the AI telemetry, and then find ways where you can track how AI and tokens are being used. Come up with your own analytics where you can say in terms of, "Hey, these types of models are being used to solve this type of problem. Can we give a structure and can we build an in-house kind of a model routing place before I go and ask Claude to say, 'Can you do this?'" It should not be using the most expensive model, so there's a bit around that. And then, AI wallet, what it did for us was, if I'm in HR, I get x number of dollars for a month to spend on my tokens, and I can monitor the usage. If I have used it up, I have to go through a certain process, explaining in terms of what's the next thing I want to use it for.
What it has done is not just more governance around the monetisation of it, but it's more around more shared innovation. So, before me going and creating my own agent, I'm asking ten other people in the team to say, "Hey, I'm doing this. Have you done this? If you've already spent your tokens, I don't need to spend on it". And then, a lot of sharing happens that way. So, we have kind of deployed these two, three things internally to make sure that from an ROI perspective, we are thoughtful, intentional. I think the main balance you need to hit here is that by putting too many checks, you are not killing the innovation. So, I think that's the fine balance you do. We call it responsible innovation, like it's still innovation, but give some guardrails so the team knows how to operate within that.
[21:29] David Green: Very good. Let's come back to culture, Avani. So, let's look at two elements of culture first. So, let's look at role-modelling by leaders first, and then we'll look at the culture of experimentation that you have at Atlassian. So, link to adoption, I guess, and making AI part of the culture, I found it really interesting in the report to read that when a leader uses AI live in front of their team, those teams are four times more likely to make it part of their daily work, which I thought was quite a big finding on there. What is it about your research, as you found, what is it about visible leadership behaviour that changes adoption? And what else have you learned about creating the confidence to experiment?
[22:15] Avani Prabvhakar: So, I'll tell you what we have found out. Initially, we were also just tracking adoption, how many people are using AI, right? Then we learned after three months in, this isn't really the right matrix. And then, we came up with something called superusers, is what we call internally. But we started tracking in terms of how people are using AI, and you have to come up with the definition which is clear to each craft. For example, if you are in marketing, if you have used AI more than 40 times in a week to work on a very critical work and resulted output, then you qualify as a superuser. So, then we started measuring and tracking our superuser population. And we also learned in terms of how -- and I'll give you the fantastic research that we found out, was at any given point in time, I don't expect and I don't want anyone to be chasing like, "Hey, we want all our employees to be super-AI-forward users". That is not the solve and that is never going to happen. I can tell you if you get to about even 10% or 20% of your population being AI superusers, I feel like you should give yourself a big tick to say, this is the North Star. And the reason I say that is if you drop one single AI superuser in a team, the overall team adoption goes up by 56%. That's what we found out. So, when we saw that, we were like, "Hey, this is a better way of driving adoption and changing the culture, rather than expecting all will become overnight AI native". It's more like you'll probably have 10% to 20% of your superusers, but make sure that they are distributed in the teams in the geos in a way that they are driving more adoption. So, that was one thing.
On the leaders, we found out leaders who are kind of running their meetings, starting their meetings, how they are conducting themselves, if they are more AI-forward, for example, if they are doing more asynchronous Loom recordings and just sending it to the team. Or for example, if they are using AI features to transcribe some of their thinking and summarise. The more they use it and make it a part of team rituals, the team adoption goes up by 4x. They feel like it's not a mandate, because that's the other thing which I would say. Like, these are the dos; what are the don'ts? The don't that hasn't worked and will never work is the mandate. So, the moment you say, "We need to get to this point", two or three things will happen: your token cost will go up, but also people will learn how to gamify the system, because they know this is how I'm going to be assessed on my performance. So, please, please do not link AI usage, AI adoption, to your performance outcomes, because then you're killing innovation straight off and creating more problems. At some of the other organisations, I think one of the CEO founders said that people are throwing AI grenades at each other, which means I'm doing more and I will churn out more work and throw it over to you because you are the next one to catch. And then, it's creating more AI slop in the organisation, because they went too hard on mandates when they started on this journey. So, I would say the don't will be like, don't do the mandates, it does not work.
Second thing on the leadership side, you asked me, I would say the one thing you can do as a leader is to be absolutely vulnerable and honest, and we do it all the time. The pace at which AI is changing, everything that I'm telling you is only relevant for the next four weeks, right? And then people want to know long-term answers and certainty. So, just be very honest about it. When you don't have an answer, just say, "We don't know". And I think the more you say it, people start to believe you to say you really don't know what is going to happen in the next four weeks. And it is okay to say that, "We are going to collectively find out together. We'll learn, we'll unlearn", and you create much faster learning loops in the process. So, we have seen internally also, teams which are able to do much at a faster pace are leveraging AI in a very different direction versus those who are kind of doing the traditional playbook to say, "We'll deploy it, we'll train everyone, and then let's go in it together". That's not really yielding the outcomes.
[26:10] David Green: I want to take a short break from this episode to introduce the Insight222 People Analytics Program, designed for senior leaders to connect, grow, and lead in the evolving world of people analytics. The programme brings together top HR professionals with extensive experience from global companies, offering a unique platform to expand your influence, gain invaluable industry insight and tackle real-world business challenges. As a member, you'll gain access to over 40 in-person and virtual events a year, advisory sessions with seasoned practitioners, as well as insights, ideas and learning to stay up-to-date with best practices and new thinking. Every connection made brings new possibilities to elevate your impact and drive meaningful change. To learn more, head over to insight222.com/program and join our group of global leaders.
When you shared the research a few months ago, you talked about -- because we've got a lot of people that listen to this podcast who work in the people insights or people analytics space, and you gave the example of your Head of People Analytics, or Head of People Insights as you call it, and how she led some innovation weeks, I believe, with her team, and their confidence went up, rocketed up, didn't it, for using agentic AI? I think you said that their confidence with agentic AI went up from 27% to 91%, but more important, they shipped six real workflow automations. So, our research finds that insights or people analytics is an important part of driving not just adoption, but getting to the outcomes from AI and HR. I don't know if you wanted to talk to that at all?
[28:13] Avani Prabvhakar: If you're focused more on the people insights team, I can tell you our team has been spearheading it inside the team, and that was more because of the leader. They set a North Star in terms of like, I asked about the core beliefs of HR, like core beliefs of what will people insights be doing in this journey? I think that's very important, anyone who's leading that team, to write it down, "What does it mean?" Secondly, is kind of do more and more experimentation inside it to test the hypothesis. So, also see like, currently the kind of reports you're sending out to the business and the work that you're doing, kind of rethink to say, "Can this be done without us? How will the business get to self-sub-model? How do you rethink about it?" I think that really helped us in terms of driving innovation. Right now, the team is at the cutting edge of, you give them a challenge, and the first thing they'll say like, "Have you rethought the whole thing?" You have to build that, I would say, that muscle memory to say like, "Hey, let's rethink it, let's start from scratch", rather than knowing how we've done it all the time. I think that's the more important shift. And you only build it once the leader and you all have created it.
We did an AI innovation week just with that team, went deep in terms of everyone from the team came and said, "These are the three use cases we want to experiment." Then, you just go down and only focus on that. What we also found out is when you do an engagement like that, that effect of that engagement stays for the next eight weeks, which means people will still be tinkering with it, working on the use cases, and then you repeat it after eight weeks. So, you have to kind of build that motion. Regular in-house training or courses, it's not really the way to get yourself up to speed on that.
[29:57] David Green: Let's stay with experimentation, Avani, because it's so important. Not every experiment works, of course, if we go back to our days in the classroom in the science lab. And I think it's worth being honest about that on the show. What I understand is that you've scoped 14 initial use cases, scaled ten and paused two, which is a pretty good success rate, to be honest with you. That talks about coupling experimentation with prioritisation as well. What did the ones that you stopped teach you about knowing when to persevere, when to redesign an idea, and when to walk away?
[30:36] Avani Prabvhakar: Excellent question, because I always tell people in terms of we get the most amount of learnings where things did not go and our hypothesis was not what we thought going in. So, we do a lot of retrospective internally, in terms of for every experiment that we ran in terms of what worked well, but what did not work well. So, in our experience, some of the experiments that did not go well was the amount of process re-engineering that was needed was way more than what we anticipated. And what I mean in different words is that we did not have enough, I would say, foundation about data foundation, data layer not ready, context layer not ready, before we went and started making changes. So, we had to pull back to say, "Okay, fine, not the right time. We need to pivot and go and fix that before we think this is where we'll go". Few things, I'll give an example. There are a lot of jobs that AI is claiming that will not be required, right? And there's been a lot of new AI startup that is coming in that space. Let me pick like, SDR role in the sales team. Everyone is saying like, "Hey, the entry level jobs in this function will not be required". And there's almost kind of, people say it with such strong belief that you believe, "Oh, fine, maybe let's also run an experiment to say what that means".
What we learned, opposite to our hypothesis, was if you don't have an SDR speaking to a customer in the entire lifecycle, you will lose accounts, like we started losing customer contacts because there was no human involved in the process, right? So, this whole hypothesis of like, SDR auto renewals, like there's a list of jobs and process flows which made you believe like you will not need human in the workflow. I think that was a wrong hypothesis. So, in fact, we have learned it on all our job families right now. Where we are spending time to give an example is our designers. There was a lot of talk about designers can commit codes and engineers can write prototypes, and hence jobs are blurring, you may not need these different types of roles. And it is true to a certain extent. So, we ran this experiment for the last nine months. We have these frontier teams. A team of five is what we think is the right size, if you have to really create AI-native teams inside a big organisation. Find some greenfield projects, give it to this five-member team, and then we studied them.
What we found is because the psychological safety is more in smaller teams, people don't mind if there's a lot of crossover that happens and the roles get blurred, which is true. But beyond that, the deep craft expertise is what is emerging to be the most important thing right now. So, like a designer can only do so much of committing codes and engineers can do some prototyping. But when it comes to once the code is shipped, the question is who's responsible, who's accountable if there's a bug that happens down the chain? How do you maintain this code? And hence, it will always be the craft's responsibility, which is engineering craft in this instance, which needs to go deep and hold that bar. So, we learned a lot in that experiment because we went in thinking like, "Hey, of course roles are getting blurred and there'll be more overlaps in the role and you may not need…" but it is changing our thesis now to say, "If anything, you need actually more deeper expertise in each of the crafts to be able to review the work done by AI, to be held accountable in terms of in the process when you're talking about human in the loop".
So, we learned a lot, I would say, in most of these experiments where our hypothesis didn't work well. Actually, the findings were much richer. I can share more examples where what we thought was true going in did not work. For example, AI adoption matrix, like it just fell flat in the first three months. We are like, "Oh, it's simple, we'll track the usage and then we will change the matrix of usage". And then after a while, we're like, tracking usage and adoption is not driving any transformation.
[34:21] David Green: You learn just as much sometimes from the experiments that don't work, the hypotheses that don't pan out how you think they're going to pan out, as you do from the ones that do work. And it's all about to learn and build the next time. And that seems to carry into a little bit on how you're thinking about roles within the organisation as well. So, I understand that you go tasks first, then skills, and then you only touch job titles right at the end. Would love to hear a little bit more about that on how that plays out at Atlassian; but also, maybe why you think most companies do it the other way around?
[34:56] Avani Prabvhakar: It's a few things that I've shared throughout the process. I feel like that is the most important ingredient, and what framework you use internally will help you drive the trust. So, right now, what we are seeing everywhere else is they start on the AI journey with a business case, say, "Tell me how many dollars you need to spend, what is the ROI, and hence, what is the efficiency that you're looking at?" I think if you start in that order, there is no way genuinely you can build trust, because everyone knows what the end goal is you're going towards. The way we do it is we have a framework, so tasks, skills, and then roles, and we are very open about it. So, when I talked about the 14 use cases, that's the work. And I can also tell you in this forum, anyone who's telling you that AI will take away these roles and you don't need AI in this thing, there's a lot of garbage out there. I feel like there's almost kind of a race to make these statements out there. So, my honest ask from all the HR professionals listening to this will be, "Can you just kind of stay away from that and not fall into the trap that I need to go out and say, 'Hey, we don't need this job anymore because we did AI transformation'?" So, don't fall into that trap.
We start from task because right now, all the real AI transformation is still happening at a task level. No one can tell you, even software engineering, that is one space everyone feels like, "Hey, now it can write codes". Coding is only 16% of the software development lifecycle. There is another 84% of the job that is getting redesigned at a dramatic pace. So, can you focus on that versus making this thing about, like, what does it mean for SWE? Do we need less; do we need more? What do we do about grads? And I'll probably talk about grads in a bit. So, the way we look, focus is just task right now, in terms of what workflows, what use cases, where are we kind of really using the AI and getting a lot of efficiency out of it. And be open sharing about that.
The second thing is when you talk about skills, and skills again has to go hand In hand. So, this is the hard part. So, for example, if you were to figure out what are the managerial skills you need in the AI era, right? People right now are saying, "The span of control will change, you will not need more managers, that is not needed", and all of that stuff. Again, you'd build some core beliefs for some certain skillset to say, "We believe, if anything, manager's job will get even more critical, because you need deep expertise to review the work of multiple people". What does that look like? What are the skillsets for a manager? It is more about ethical leadership. A manager is somebody who is able to be good at orchestrating the AI layer of work inside the team, right, someone who has more context and breadth outside of their role. So, if you're just a good engineering manager, it does not help. You need to be able to go across the life cycle and understand product and design as well. So, we've invested a lot of time writing that skills down, because people care about it. A lot of them want to know, "Hey, what are the skills I need to really develop to get to the next level?" So, I would say that helps you build trust and you have to do it through the way of experimentation.
Third is where the role comes. And I can tell you it is the most delicate, sensitive, and also the most difficult part to really crack, and it requires a lot of hard work. So, for example, I gave an example of designers, and we are doing it for other job families as well. You have to identify your more AI-forward, AI-native folks in the team who are using technology very differently, based on their consumption or are they superusers. Give them a very different task other than what they're doing in their core job. Change the toolset, like don't let them use their current toolkit. Give them, "This is how you'll be working, only with these five Claude, whatever you're using role, work with that". And then, study how their job is being reshaped. Like, are they reviewing their work the same way they did before? What are they doing throughout the day? You write it down, and you have to write down almost like a new job description after reading and researching their role for four to five weeks. Once you've written it down, and therefore I'm saying it's painful, it takes a lot of time, then look at that to say like, "Hey, this is what a new role will look like before what it looked". And then share that, and that's what we did. We call it design technologist, and we have shared it with internal teams.
So, now people are more focused on, "Okay, fine, that's the new role, that's the new job, this is the skills that are required". And then, you create the full learning loop around how do you get there. That has really helped us immensely in building trust with our teams. And no one is worried about experimentation, or no one is worried about, "Oh, my AI token consumption is so low this week. Does that mean I'm not innovative enough?" I think you have to drop all of that and the focus gets on this framework. So, that has been one thing. And I would say that is one takeaway I would ask for the team to think about, how you're driving it and whatever is your framework. It may not be this, but create a framework that the whole team can anchor around it, and which is honest and authentic.
[39:44] David Green: That's really interesting. And you talked also, Avani, about the role of the manager and how that actually becomes more important in terms of setting direction, exercising judgment, getting involved in redesigning workflows, and looking at some of these new roles and new tasks that are going to be created as well.
[40:04] Avani Prabvhakar: Yeah. And one more thing I forgot to mention about grads, because again there's so much noise like, "Early-entry jobs are gone", and everything, we studied our ENch cohort and we kind of studied their patterns and how they are using AI and everything, and we have facts that the new grads who are coming to the building, they are able to make decisions by using AI 19% more effectively than some of our senior folks inside the building. And there are a lot of such metrics that we've been looking at measuring. And that has helped us build our core belief to say like, "Hey, you know what? If anything, we will have more grads inside the building at any given point in time". Where we will focus on is currently, if they take about two years to get from here to here in terms of seniority and experience, we are focusing on how you reduce the time frame by using AI to say, "Can it be six months, one month?" so you can accelerate that learning, rather than saying that you don't need this incoming talent. So, I think you again need to have some core beliefs based on your own thesis and organisation, but that's where we sit with all things grads.
[41:04] David Green: You anticipated my question, Avani, I was actually going to ask about that. And yeah, if you just think about where's your succession coming from, where's your innovation coming from, and actually there's a social responsibility around this about supporting early careers. And it's interesting that your data suggests that they're actually going to be an even more important part of the workforce in the age of AI, so it's really interesting on that. One of the more provocative ideas maybe in the report is that HR may need to move from managing headcount to managing capacity across people and AI. What would that shift mean for workforce planning, organisational design, and maybe the capabilities that HR professionals themselves need to build?
[41:55] Avani Prabvhakar: So, we are experimenting. Full disclosure, we have just created a new role called Capacity AI Capital Management person. We are still figuring out even what we call that role, but we have one person, and this is what I meant when I was sharing the HR beliefs, because we believe that the HR role is going to shift from workforce planning management to actual managing the entire AI capital workforce inside. What it means is, and I'll probably step back what is happening at this point in time and then I can explain the future state, right now what happens is there is a strategy organisation in a company which is kind of looking at the key priorities, how the work gets broken down, and that's sitting in one part of this thing. Then you look at all the token spend. That is usually sitting in a CIO, CTO space, and they're kind of tracking how much the AI spend is happening.
When it comes to in terms of people upskilling, like how are people kind of ramping up with AI and adoption, everything we spoke about is sitting primarily in the HR space, right? And then, of course, any spend-related human capital is sitting either with HR of finance. That's the typical layout right now. In the AI world, I feel like there has to be one team, one person should be looking across all four things, which means do you have your strategic priorities at a company level being done by the most important people, which could be agents or HR internally? And then, how is the spend looking both for token and for human spend? And then, how are you upskilling not just humans, but also your agents right now? So, you should be able to see the skills of both agents and humans side by side to see how we are tracking. So, that role, I feel, is going to become more and more important.
I have a strong belief that HR is the sweet spot which can do all of these things across. And that's the role that we're experimenting to say that will become such an important ingredient of every board discussion, every company in terms of making any strategic moves. They would want to know how much you're not just spending on token versus human, I find that's a very deprecated view. It's more in terms of, what does the overall skills repository look like between human upskilling and agent skilling; and then, how are you aligning strategic priorities that you have in the organisation, while maintaining the cost factor behind the scenes? That's how I visualise it. That's what we are working on. I'll probably be able to share more once I have done the experiment, and share the job description with more details of what that role looks like.
[44:25] David Green: What should leaders measure if they really want to know whether AI is genuinely creating value, rather than simply, as you talked about earlier, just generating more activity and just creating people doing more work, sending it to other people to review and add there, and it just creates a vicious cycle then, I guess; how do you really measure if it's generally creating value?
[44:48] Avani Prabvhakar: I would say in terms of more value creation, there are a lot of AI maturity indexes. One way to measure it is that you are not measuring the outcomes, you've defined your AI maturity index internally as a company, and then you're measuring yourself how you're kind of climbing up that ladder. That's one approach you can do, which is not outcome-based. And in that maturity index, you can say things like, "Hey, at this point in time, we are using AI to just review the existing workflows, which are like P2 or P3 level. We haven't gone down to P0 level, right? Has AI been embedded in a way where every superuser inside the team is using it to X% of capacity?" So, for example, I'll give you a real example.
So, if you're a market researcher, right, you could be an AI simple user, which means you're asking search from AI and you're helping AI to kind of collaborate your work in terms of looking at your emails, doing some research for you. So, that could be one category. If I'm a market researcher and I need to step to the next level, which is of being an AI strategic collaborator, what it means is I'm not just using AI to do my work and pulling the reports, but I'm building hypotheses with AI, which means I'm giving it variables and then we're working and sparring on it almost as another colleague of mine. And in order to do that, you will have to embed AI in all other cross-functional workflows, because you can't do your market research just sitting on your own with this thing. You'll have to work with the finance team, you'll have to work with the product team, so on and so forth.
The reason I took you through that example is that's another matrix to measure in terms of how your employees or how your workforce are coming up that maturity curve, and that could be one way to measure it. The second bucket of measuring could be purely outcome-driven, in terms of, "Hey, I've identified these 14 critical workflows and if I can do ten in six months' time by keeping my token cost in this guardrail, then I would say this is …" which is a much easier one, I think, finance and people reporting to the board that matrix, because it's more quantifiable. And then, the third one is more around the governance one, right? So, it's like a multifaceted thing. One is on the people maturity; second is how your processes are this thing; and the third one, which everyone cares about, is the safety governance. Like, have you built more measures than you did six months ago to make sure that in the event AI went rogue, what measures have you put in place? I would say it has to be like a three-dimensional thing.
[47:12] David Green: If you had to boil all of what we've spoken about today, Avani, and the additional research that you've done, for a leader that's maybe a year behind you, 18 months behind you and Atlassian on this journey, maybe someone who's listening who's still treating AI as an IT rollout or a technology transformation, what's the one thing you would tell them to do first?
[47:35] Avani Prabvhakar: Because the audience is HR, I think the first thing I will tell them would be don't use AI just for search and chat, which is what most of us use right now, or summarising my emails or making them look efficient. If they can go behind the terminal, like I love working behind the terminal and I love it a lot, you really understand in terms of the whole mechanics of what agent skills to build, how does the whole AI telemetry work. So, I would say because the audience is HR, make extra effort to really understand the overall AI infrastructure that is sitting behind. And the reason I'm saying this is then you will be able to invest more time building the context layer that needs to be built, and not just stay focused on like, "Hey, people upskilling and is it AI training?" I would say like, yeah, although that is good, but since you're in HR, I would probably ask you to move a little bit more on this side of the house and kind of learn and understand that, because we'll all need that in our jobs sooner or later, and that proficiency needs to be build. That's one.
Second thing is within HR, Identify two or three workflows, not the usual suspects, not HR Ops, not TA, go after some of them, like maybe comp-related, which is the most sensitive, right, and try and see how you can build a more intelligent interface between what a manager wants to see, what a leader wants to see, and kind of try and solve the hardest and gnarliest problem. Give yourself as a challenge and try and do it within the HR team, right? You will learn a hell of a lot. That's what we learned when we were kind of trying to unpack onboarding, and we've done many more such use cases. I think it just takes your own literacy high, but it also gets a lot more momentum within the HR team to start thinking like, "Hey, you know what, we can move beyond this Chat, or making a PowerPoint look better"!, or something like that. So, just go a little bit deeper. I would say those would be my two things for HR. And if it is a parting note, I would say, I feel so excited. I've never felt so excited about HR craft in the longest time in my career. I feel like this is one time where we are genuinely sitting on the driving seat.
So, it is for us to drive, right rather than wait like, "Hey I'm just going to do the upskilling part here and I'll just look after the cultural transformation". I feel like this is the time to shift gears and move into the other side, and kind of own the whole space end to end to say, "You know what, no one better than us to be driving this". And I understand every organisation might have a different top-down mandate, but I feel like we can create our own opportunities by proving it to say like, "Hey, I looked at this workflow outside of my purview, partnered together, and we delivered it with less IT intervention". So, I don't know, but just kind of have that belief and ride this wave. That would be my one message to all HR folks listening to it.
[50:25] David Green: We've had some of your peers on recently, Kelly Jones at Cisco, Sharon Doherty at Lloyds Banking Group, Yanine Vos at Rabobank, and I always ask them this question, because they're in that Chief People Officer expanded role. People insights, your people analytics team, and we've had the pleasure of seeing the growth in the people and insights function at Atlassian, being members of our People Analytics Program at Insight222 for a number of years, how does the People Insights team help you as a senior executive in the organisation, having conversations with the CEO and the rest of the C-suite?
[51:03] Avani Prabvhakar: Our team has been, I can say very proudly, one of the most forward-thinking teams, even within the People Insights team, so they always experiment heavily. So, we have moved past that stage where I'm just giving great insights to the CEO or to this thing. I think that, for us, is like table stakes that has been done by the team independently. I think where the People Insights team is really leading the way right now is setting the benchmark of what the future of this team will look like in the AI space. So, they are the thought-leader there, working internally and experimenting heavily, but also meeting with the peers externally, like Insight222. So, for me, I feel a lot more pride because they've created a lot of agents internally within the Insights team, which is unheard of. They're not just reporting agents, they are not just pulling stuff from Databricks and telling you that you can do this smarter, but they are kind of thinking about it really in a different way. So, for me, Insights team is less about just giving smart insights to the business, because I think they've hit that benchmark already. It's now in terms of how you rethink this craft in the AI world, is what they're doing. So, I'm super-proud of them.
[52:12] David Green: Last question, Avani, this is the question of the series. So, we're asking all our guests in this series this particular one, and I think it talks really nicely to our whole conversation so far. How can the HR function help build the capabilities that the business needs next?
[52:29] Avani Prabvhakar: We can help first by building it ourselves. So, we need to show and then tell in terms of what it means to be AI-forward and AI-native. And that's why I said go behind the terminal and do work like an engineer and go deep, because honestly AI has lifted the barrier. I don't need to know code, I just need to write in simple language. But then, you start solving problems very differently. So, I would say lift up the AI-literacy inside the HR team first. That's one. Helping build the capabilities for the business. If you can make sure that there is no siloed organisation on the side, which is usually called the Transformation Officer or leader driving AI, please stop that immediately, and try and find that you have AI growing inside each of the teams. This is not a digital transformation being done from the outside. So, if you can help each leader really get comfortable with the learning loops, I think that's the only one thing I'd say. Create a culture of experimentation, innovation day, every eight weeks, whatever it looks like in your environment. If you can create that and get that engine going, I feel like we have done our jobs, because this upskilling and capability-building is not like training or creating more modules or courses.
Or even if you're thinking about any of that, we built an agent built by the HR team, which is like a personalised AI agent for every employee, and it is a very personalised learning. You go there, it's called Loomis, and you don't need to ask anything, it already knows what you do, and it helps you how to get there. But that to me is like tier three, tier four. The things that I said in the beginning in terms of really creating that learning loop, cultural experimentation, is how the capabilities will be built, because it is going to be evolving. It is changing at such a fast space that hence, creating that loop is more important rather than building a platform.
[54:15] David Green: Avani, really enjoyed the conversation. And the work that you and the team are doing at Atlassian is super, super-impressive. And thank you for sharing some of that in the Leading With Context research as well. Before we finish today, where can listeners follow your work, learn more about Atlassian's research, and maybe what should they be looking out for next from you and the team?
[54:38] Avani Prabvhakar: Oh, we are building more playbooks right now. So, we just released a forward-deployed engineering playbook, a feature of software development lifecycle. Where to find them? It's on our website because we believe in open source. So, we publish all of this up there. And I'm also happy to share this with you, David. A lot of the State of Teams research that our lab is doing, in terms of the ways of work and roles and everything that I spoke about, happy to share it with you and we can use your platform.
[55:02] David Green: Brilliant. Well, Avani, thank you very, very much. I know it's coming up later on a Friday night in Australia for you. So, thank you for taking the time to share the work that you're doing with our listeners. Thank you very much.
[55:15] Avani Prabvhakar: Thank you, David. Thanks for having me.
[55:18] David Green: Avani, thank you so much for joining me. I've really enjoyed hearing how Atlassian is approaching AI, and how openly you share what's working, what isn't, and where your thinking has changed. One point that particularly stayed with me was your advice that HR should build these capabilities itself. The People Insights teams experience is a great example of how experimenting with real work can develop confidence and help people rethink what their function could become. So, turning to our listeners, I'd love to hear where your HR or People Analytics team is experimenting with AI and what you've learned that has changed your approach. Come and find my post on this episode on LinkedIn and let me know in the comments. If you'd like to explore Avani's work further, Atlassian publishes its research and practical playbooks openly, including the Leading With Context report we discussed today. You'll find the links in the show notes. And if you enjoyed the episode, please share it with a colleague who would find it useful, and subscribe so you don't miss what's coming next.
One last thing before we go. If you'd like to keep up with what we're working on at Insight222, follow us on LinkedIn, or head to insight222.com. You can also sign up for our bi-weekly newsletter for the latest thinking on HR, people analytics, and workforce transformation. Right, that's us for today. Thanks for listening and we'll be back next week with another episode of the Digital our Leaders podcast. Until then, take care and stay well.

.png)