Name: The AI pilot that makes it past approval
Uploaded: 2026-10-07T15:01:04.989Z
Duration: 1 h 25 min 28 s
Description: The AI pilot that makes it past approval

Transcript for "The AI pilot that makes it past approval": Yes. It'll be good. Right. I think we are live. So welcome, everyone, to our fireside chat with the collaboration of... Within the collaboration of Personio and Agile HR Community. My name is Riina Hellstrom. I am the founder of Agile HR Community, and we have Rebecca online as well. Rebecca, would you quickly introduce yourself? Yeah. So my name is Rebecca. I'm, working at Personio based in Dublin and, basically working in all things people, talent, and development. So I'm the senior lead for people and leadership development at Personio. Wonderful to have you here. I'm really excited to start talking with you about AI development. Let me just share the slides here first so we can get some visuals. So that's us. And so we set this up as as both of, Personio and we at Agile HR Community come across quite a lot of HR folks and they are overwhelmed with AI possibilities currently. Many of them are trying and experimenting in their own little areas of work. We can see HR people within their own work utilizing ChachiPT or co pilots or something which is kind of in a personal setting, improving their performance and helping them deliver what they need. Today, we're going to focus more on how are we going to move beyond that experimentation and that everyday use of chat GBT or LLM towards actually embedding that in HR's workflows, delivering something of value to the organization, employees, leaders, that is sustainable and maintained and continuously within. the workflow. So our agenda looks like this. We're going to start the Firestack chat very soon. We are going to first kind of set the scene what we're talking about and where we come from, both of us, and then going into, actually talking about a sequence of how we work to get get from from a kind of experimentation space towards an ongoing maintained AI workflows, agentic workflows. We also will have some time at the end for potential questions. I think you can ask your questions in the chat and those will be moderated to us. We've got a wonderful Phoebe in the background helping us, so thank you very much Phoebe for that. And at about at about, quarter past, quarter past three, twenty past three, we will be closing and thanking everyone. So, Rebecca, a bit of background because we are here two different organizations, Personio and Agile HR Community. We've found a bit of, the same clientele. We've got the people thinking about the same questions, but we come at it from two different perspectives. Would you want to have a do a kind of a one minute introduction of what Personio is so that people understand who who is here? Yeah. Definitely. So I always like to say I think the interesting thing about my job is I'm an HR person that works for an HR company. So, definitely, yeah, lots of fun. And if you don't know Personio, we are the intelligent HR platform. So we really focus on connecting everything around core HR talent and also payroll. Our main goal is really to help HR teams, especially now, adapt to changing priorities and really focus on what matters most, which is the people. So that's kind of our main, mission statement and something that I think we can all agree is probably more important than ever right now as we go through AI Am I numb? Yes. I'm unmuted. transformation. For sure. There's so much going on on techy side as well, and I think that many of the tech platform providers are already generating AI agentic workflows within the platforms itself and themselves. So it's really cool to see what's going on in that space. But, let me do a quick intro of agile HR community because we are not at all in the techy space. We are, an enabler for HR teams to move towards the future. That's what I usually say. So agile HR community has been around for ten years. We are one of the leading or actually the leading global community and consultancy delivering new ways of working. So the how HR team needs to be working to be able to move towards being future ready. We do training, we do consulting, transformation work, and we have a community as well. So this kind of a shared community event. This one, we have a community where we bring in meet... Where we have meetups and bring in people who've done amazing things with what we taught them. Our job is working incremental stepwise development. Many of them now utilizing that for AI, development as well. So that is us and, maybe a few words about my... Myself. Twenty five years in HR strategy or design. I've been looking into HR like my own pocket, so I know all of the... All about that. I'm fifteen years with bringing agility into HR. So why am I speaking with with, Rebecca about AI is because I'm very intrigued of that myself. So I've been tinkering around for four years and supporting both on executive level and team level with what this actually means. So today, I'm really glad to be speaking with Riina. You've been... You you come with very recent experience from from building, from developing, from innovating, and, what we're going to talk about today. So let me set up a little bit of the scene, around that. Let me get that up. So you will get this picture, by the way. So you will be sent this picture. You don't have to worry about starting to write it down or document or to do screenshots. You will get that. So just focus on being present and listening. When we're working with any kind of usually digital development, but specifically AI development, there are a few cycles that we have to look at to actually get from just having great ideas, which most of the teams right now have. Like, let's just solve with this with AI. Well, there's a long journey to actually make it work sustainably, ethically, and delivering the value it should. And we're gonna talk about this this kind of journey today, starting with defining the right problem and understanding what the actual problem is that we want to start solving or what the end to end value delivery is where we want to start embedding agentic workflows in. We're going to move into talking about the prototyping part. Prototyping, learning what really works because we can't just go from idea to implement or build and implement. We really have to start bringing in that incremental way of working. Next, we're gonna talk about kind of the development part, developing that and validating that it actually works, releasing and stabilizing it, and then running and improving the workflow, whatever that is. All of these require certain things which we have on the bottom. So a certain end to end ownership, the architecture data and knowledge management in the background, some kind of rules, guidelines about governance, assurance, and ethics, and kind of staying within within the constraints that we need to have, compliance. And then, of course, making sure that the value is gained and performance is measured for the end to end workflow, whatever we're developing. So we're going to talk about this today. Anything that you want to say about this kind of... It looks like a tricky image, but Rebecca, what's your take when you see this? Yeah. I mean, I think it's really helpful framing... To me, the most important part here is just defining the problem, and I think so much employees are like, let's use AI or even here, you know, from execs if you go on LinkedIn. Like, how we're using AI, implement AI, but I think going back to the why and what problems you're trying to solve and thinking about the process end to end first before jumping to the AI solution is something that can be missed sometimes, but it's so important, especially after you've implemented things and you go further on down the road. So, honestly, I think it's it's a really helpful graphic, but the most important part is just defining that problem. I fully agree. So that's... I mean, anything else that you go is just waste. If you're not solving the right problem, you're just gonna waste time. I'm gonna stop sharing this, and let's let's go into speaking about this. So let's just first start with, Rebecca, where you are with your AI journey, where you started, where you are right now, just kind of to get... I mean, together kind of image of what you've done personally as a HR leader to get where you are. Yeah. Definitely. And I think I'm kind of coming at this from two lenses. So being someone who's on the HR team, I can speak to how we're piloting and using AI ourselves. But then also in an enablement role, it's my job to kind of think across the entire company. So how are we supporting the whole company and doing this and kind of bringing everyone up to the same level while also making sure we're doing this for ourselves at the same time, which I'm sure a lot of people in HR can relate to. So even if I start kind of more towards the beginning, one of the first things that we did across the company is think about, like, what do employees need when it comes to AI and how can we get them familiar... Familiar to experiment with these new tools that we're bringing in. So honestly, this happened a bit even before the problems were clearly defined because I think you're kind of bringing people up to speed with new skills, new technology at the same time that you're building. Like, AI is moving so quickly. So we've done this, really through something called AI Surge Week at Personio. And this was, a company wide week where we really encourage people to cancel as many meetings as you can, dedicate to learning sessions, but most importantly, hackathons, at a department level. This is where people could experiment, could be... Bring the problems that they wanted to solve, and just get in groups and think about how they might solve those problems with AI and to get people talking. So I think, Riina, that was honestly one of the most helpful kickoffs, to also get that confidence level up. But the most important thing after that was then, okay, what do we do with this? So we continued, training sessions for the whole company. But on the people team, specifically, we created, from these conversations in AI Surge Week and beyond, a tracker of different problems that we thought we could solve with AI. And we really prioritize those in terms of, like, what's the simplest, lowest lift to what will have the biggest impact. And we're still working through that tracker today, updating it, but I think it really helps giving up... Everyone that kind of single source of truth and vision to, like, what are we trying to do? What problems are we trying to solve? So that's kind of where where we are today. Like, we're still building and iterating, but, yeah, have this tracker, have created more in terms of guidelines and governance, which I can speak about for hours. It's a huge thing that employees are asking for, even in our survey data. But really, yeah, building building the plane as we're flying it as well. I'm hearing so many different levels. Let me start. unpacking what you just was. that. was a lot, so I can. dive right. into any areas. So let me dive into a few things here, Yeah. first of all. So I've been working with quite many HR teams as well where we're looking at just kind of if we take a few step back was... Is where do we start? Which team should start? Like, all in? Where should we start more with experimentation, learning? Where should we actually hold back a bit because there's something bigger going on such as an infrastructure change, an architectural change where it doesn't make sense to start building anything before the back end is correctly built. So I'm I'm hearing that your organization made a decision like, let's just get people start becoming familiar with with these tools, practices. You offer them certain technology that they can utilize at their own work, and then you have the hackathon where they can start bringing kind of ideas together and thinking about what could we start solving with this technology. That sounds like a very nice place to build that energizing inspiration and to start seeing the opportunities. When when you do... When you did this, did you did you notice that you get... Like, the amount of ideas that you get is just massive. What's what's your take on that? It's actually... It's interesting. So we've done two different AI search weeks. We've been doing it for two years. I think the first time we did it, we were expecting an influx of ideas, but actually they were quite limited because people didn't even know it was possible. So it's a bit difficult to first say, okay, give us all your ideas of how we can use AI. What people actually needed and asked for first was, okay, what are some examples? Like, what are other people doing? What's even possible in these tools? And so I think that's where the experimentation piece and the peer sharing in our first hackathon was really important because people could see, like, oh, this person's doing this. How can we raise that a level and maybe build on that and kind of get that inspiration from peers? So, yeah, it was a bit counterintuitive. I think we were expecting a lot, but, again, I think with any new tool and any new way of working, people need some examples of what's possible. But then by the time we got to our second hackathon, it was much more natural and the ideas were pouring in and this is where we had to take a bit of a step back and think about, okay, what problem is this really trying to solve, what's the impact, and how much of a lift is this, how realistic is this as well. I think that's very intriguing as there's, I recognize a bias in myself now when you say that. My background is I've been working now for four years. Immediately when I got access to GPT, I was hacking around with it, learning how to prompt, learning how to utilize it. Now I'm building agents. Now I'm looking into knowledge graphs, etcetera, second brains. So when when you're experimenting yourself that much, you you might see those opportunities and you might rethink even full workflows is like, okay, onboarding. You don't go back to think about how onboarding has been done forever as a process. You go back and totally rethink how would how would I approach this from an AI native perspective. And that kind of clears the table quite a lot. And to get to that point, you already have those ideas on how it works. Does that make sense? Definitely. And I think when you actually take a step back, and we've done this with a few of our processes, so much of the processes that we have on the HR team are workarounds for tooling issues, not with Personio, but let's say, you know, with our single sign on issues or how we're using different tools or the kind of, like, the legal process to get somebody onboarded. And so if we take a step back and think, what are we actually doing that doesn't make sense anymore? It's a it's a larger kind of mindset shift, but it's so important. It is. And that... That's such an important thing, what you just said there that... So what I'll... I'm learning from you is when I go in and help teams understand these or HR leaders get their... Wrap their heads around it, I have to think about always where is the organization right now. So having some kind of maturity assessment, understanding, of course, where they are and working with that a lot. Definitely. So... Can I say something else to that too, Riina? Because, I led, we called it the AI Momentum Program. So it was basically a company wide AI upscaling development program that had a lot of different elements. But one of the things we learned that didn't go as well as that program is we made an assumption, about where the org was at as a whole. But in reality, people are at every end of the spectrum. And so when we think of the organization or even a department as a whole united group, it left a lot of gaps where people were, a lot less familiar, a lot less confident, a lot less skilled in AI than we thought. And then there are people who could have been teaching these courses who were a bit bored. So I think, yes, knowing where the org is at, but really thinking about how can you know where individuals are at and kind of maybe what your different adoption groups look like. Yeah. Or even build with AI native thinking about learning so that they would actually choose their path to continuously. go in. Some of that challenges them Yeah. with that. One thing that I noticed is also, if we go into... So let's think about, let's let's imagine that we've done a bit of upskilling. People are already there. They want to start experimenting, exploring. We have all those ideas. Now let's talk about the prioritization. That is... That's one thing that I quite often help with, whether it is AI or something else. We set up ways of prioritizing and quarterly thinking at what what we are going to use capacity on. So calling... We call that portfolio management. But, when you start prioritizing around AI, development, one of the key things that comes to me is, let's not even talk about AI development as a separate prioritization issue. AI development should be part of solving something bigger, something which is on your agenda already, which is high value, where you can utilize AI, agenda workflows, AI assistance, whatever that is, and prioritizing according to the biggest value that you can get either transaction wise. There are a lot of transactions out there that you can make so much more higher performing or quicker or efficient, or then there's just a strategic... Some kind of a direction where you need to move towards. And this agentic workflows or AI will help you move the organization, employees, or leaders towards that. Talk about the prioritization, Rebecca. I mean, that you would like to do everything, but you really have to be careful when you decide on what to invest in and where do you have the realistic capabilities to start building as well. Yeah. Great question. So I'll talk more, just about the people team now, since I'm sure a lot of people here are on an HR people team and thinking about that for your functions. But the approach is very similar across every department at Personio. We really... Again, after this experimentation phase, surge week phase, people got really excited. They were vibe coding lots of different things. They were building these different automations. Some of them, we decided to scale and kind of build out for the team. Some of them were just practice, like, really cool use cases, but maybe not something that we wanted to focus more on now. But the first thing we did was just compile a long list of ideas that people had and crowdsource that. Even when I think about buy in from AI, what's interesting is oftentimes in HR teams, we think about how do we get buy in from the top. But with AI, it's like that buy in is there, and we're being asked and encouraged to use AI in any way that we can. So now it's more how do we bring our employees along on the journey and get buy in from that way. So one of the ways that we did that was crowdsourcing ideas from the people on the ground doing the work. Right? If we have a coordinator who's managing a heavy process, they should be at the table to think about some ideas to automate that. And then I think the second piece around prioritization is exactly what you said. Like, it's one thing to say, like, how will we use AI? It's another thing to say, okay. How will we make our processes more efficient both from the HR perspective but also from the employee experience? How will we make sure we can save people's time to give them areas that they can grow in and more time to focus on more strategic things. So that was kind of the lens that we took even when we crowdsource ideas. But then when we got this long Google Doc of ideas, we went through it and sub teams, on the HR team. So, again, we've got, a decent sized HR team, about 80 people, and so there's different sub teams within there. So we had someone who was designated, kind of our HR champ or our AI champion for the larger people team and some more people who were very, very interested early adapters. So they went and talked with each sub team to think about, okay, what are your priorities on this list? What problems do you wanna solve? From there, they also went away and thought about where there were connections between different sub teams, right, who had similar problems that we could focus on that would make a larger impact. And then also what's the low hanging fruit and what takes more time. So really looking at prioritization through impact, but also through, yeah, what's what's the difficulty level, what's the time investment. And we really started with some of those simpler automations, and AI uses first to kind of build the skill, and now we're focusing more on kind of the larger things that take more time as well. I don't know if that answers your question. Excellent. There are a few things I want to point out to the listeners here. First of all, you mentioned you already had buy in from the top. And, with my clientele, I have very, very different types of buy in from the top. So some say there is a strategic must. You know, we have to go. We have the investment. We got the budgets. We might not have the capabilities and skills yet, but that's going to be sorted out as we move. But they they really have a go at, like, go and do it. Others are struggling with, that the main investment of AI goes into... Or not struggling. It's a prioritization on high level. The main investment of of AI development goes towards r and d product, software, towards supply chain, towards the business, and the support functions are kind of a bit back behind. And they would need some some, support from cyber, smart IT, tooling, etcetera, but that's not on the agenda because they need to prioritize. So they're kind of waiting in line for getting there. Some teams are saying, they are, of course, as I said, starting on with with their own work, but there were... There are a few HR leaders who are saying that we literally are almost blocked from starting because our legal compliance and cyber think that because we work with people data, it is so scary and so, sensitive that we shouldn't even use AI, and I am fully not of that opinion myself. So what do you say about these kind of stakeholder management and getting the permission and the investment to actually move to... Move forward? What comes to. your mind? So I think I can speak probably the most to that last point, because I think it's probably something we can all relate to being in a people function where we are dealing with highly sensitive people data. You can't just throw that all into an LLM and automate it. Right? If if we could, there's so much more that we could do. But one of the first things we did, even in the first AI surge week was we started a risk assessment with the legal team, which is still ongoing. But I think bringing in our legal team and our security team at the beginning, submitting some of the use cases to them, like not just saying how can we use AI, but giving examples of different ideas that we've crowdsourced and getting their feedback, and having them review it from their expertise against EU law, and different compliance is really, really important, but they're an ongoing stakeholder. It's not just a one and done thing, but I think if you need to get buy in, like, going straight to the legal and security is is really, really important. And from there, we built in guardrails for... I'll just speak to our people team, but it's happening across the company. So this is something... Again, it's not one and done. We're continuously updating these as things change, but just being able to think about things like data minimization. So when we use AI at Personio, we still do not put any sensitive employee data in our AI tools, which can hold us back from doing a lot of great things, but I think it's really important to start where we are. And so there's also a lot that we can do without putting sensitive AI data into the tools, and maybe that's something that I would focus more on. So one of the first things that we built as a people team after AI surge week, and this was just a few people on our people ops team, but they identified one of the biggest problems that they're spending so much time on is helping employees calculate their leave balances, gosh. when it comes to things like maternity leave, sick leave, child sick leave, like there's lots of different laws in these different countries. It's different in every country and so they built a calculator that any employee can go into that's embedded as HTML into our intranet. And there's no, like, there's no sensitive data, but employees can go into that calculator and figure out exactly what their leave balance is without submitting a ticket to the people team. So So I think that's maybe one example where even if you start small and, like, build the skills with those use cases, think about, yeah, what problems are we solving, what are we spending time on that's maybe we could fix with an AI solution, to kind of prove your use case as well. A 100%. Those are so quick and easy, kind of easy wins. And also releasing a lot of time from people to actually then think about those more more difficult, use cases. Then the other thing that you said, was that you had kind of AI architects. So somebody who is responsible for doing the thinking on a deeper level and sourcing information from the other teams trying to figure out what's possible, how does it fit together, where could we find modularized approaches or or synergies. So, clearly, somebody who is a bit further along their journey who can then start road mapping. Is that what I understood? Definitely. And I think that's so important. So one of the first things we did, even at a company level, was identify, an AI champion for each department. And so their role was to kinda represent their departments. They meet centrally as a group. But then from there, we actually did that more on a sub department level. Right? So what I spoke to is we had our AI champion, go to each of our sub teams within the people team and kind of identify the people who were really passionate about, you know, improving problem solving with AI. And so I'm also an AI champion now for our people team. I am not a technical expert by any means, but I can kinda support the larger group when it comes to topics that I know and the... What I'm building. And then I meet with the AI group, on a certain cadence as well. So I think just thinking about who are your champions, and it doesn't always have to be someone really technical. I was just talking the other day, and it's like one of the things that's most important in these AI champions is someone who's a problem solver and can collaborate with other people. So can you figure things out? Can you help define the problem and bring that to the larger group? But, yeah, that's really important when it comes to adaptation. I've noticed the same. I work a little bit around the kind of as a as a AI value architect for for HR teams and just being the person who can speak both with HR and know exactly what they're working with and what... Where they're aiming towards. And then I can have a conversation with the IT leads, cyber and the tech and the architects around, you know, the guardrails, the governance, who is owning what, how do we how do we get, what kind of capabilities do we have in HR? What do we lack? How can we build that collaboration? We also have set up, and this was quite an interesting case. We supported one larger organization. They really wanted to move fast with AI development in HR, and they need to set up an HR IT team for doing that. We've got some background noise. Oh, sorry. That's me. Okay. Is it done? Okay. Yeah. So we we set up an IT HR team, and we really needed to work a bit to make them work well together as the IT people, even if they've done AI development, they were used to this kind of a ticketing system that if anybody needs something from them, we're just gonna send a ticket to them. And you don't develop AI that way in HR specifically, So we really need to bring them together as a coherent team and and set up the agile ways of working, the frequent planning cadence, the review cadence, test validate cadence. And only when they could start speaking the same language and working together in the same way, we could start seeing, okay, now now there is production going. You know, we we get the production working and they start really developing something that works. So, so just having... What I want to come back to is just having those connecting people who can speak different... Speak with the different parties here, including finance, by the way. This... Very few people that I've been working with have yet established the collaboration with finance to discuss... This goes a bit higher, but workforce planning and budgeting of workforce, the Gentic versus people, token burn. All of these things need to be sorted because it doesn't make sense just to say, oh, we're gonna, you know, re... Redesign this workflow. We're gonna we're gonna have a 100 people be finding some other jobs because we got agents doing that. But what about the agent costs? Right? So we need to be able to have some kind of a discussion with finance as well. With this... These different stakeholders, anything before we move forward to the development space that you wanna say, Rebecca, when when speaking about kind of the key high level stakeholders to to make this work well? Yeah. No. I mean, I agree with everything that you said. And I think one of the things, especially from the finance piece, is it's it's a new world. Right? When we think about workforce planning and headcount, we're always thinking about a person. What's their salary? What's their bonus? What does that look like? And now when we bring agents to kind of work alongside humans in different ways, yeah, it's a new language that we're speaking. And so one of the things that we're doing at Personio, it's very, very baseline. It's like, where are you tracking how much you're spending now and how are you tracking that? So we've just built that into our tool so we can go in and see how many tokens, individuals are using on a team level, on a larger level, and kind of can dig into how they're using them. And then that's one of the first things that we can do to help inform that from the finance side as well. That is so helpful. That is so helpful to hear that you're doing that. Right. Okay. Let's move into the development stage. Imagine that... Let me bring that, visual. Alright. So let's imagine that we've done defining the right problem. We understand which end to end workflow we're going to start working with, bringing in AI agenda workflows or assistance or augmentation there. So we have the worthwhile problem to solve. Now we need to start... We can't just move into, okay, let's create a plan, and then we go according to the plan and deliver. So this is where... Maybe if I may say, this is where agile comes in. So with the previous project management methodologies that we have been teaching for twenty something years, thirty years in schools and and organizations, we usually come in with have a plan, have a perfect plan, then you have a time deadlines and and timelines and tasks and deliverables and gates, and then you move with that plan and out comes a perfect solution. Well, this is where, it falls down. This doesn't work with AI development at all. You will need to have a vision. You will need to have a goal, but you need to establish, but you cannot deliver this with a so called waterfall plan. This is where you need this incremental, iterative, customer centric user validation, continuous improvement loops of learning. That's called agile ways of working. I've recently had quite a few connections to some HR research companies that say now there's findings from HR research where HR teams are saying agile ways of working is the enabler for actually getting this done. So that's what you see in front of you, people is the prototype and learn that the develop for everyday use, the cyclic way of getting things done and not the waterfall way. Let's go into the prototype and learn to get the evidence to actually proceed and invest in something like properly. Rebecca, what are your experiences with getting something that you can see, touch, try, get feedback on very quickly without starting to build the whole thing, so that prototyping phase. Yeah. I think it it goes back again to just starting really small, with with smaller apps that we can test. And maybe it is a full solution right away, but it hasn't taken a lot of time. So I think the example is, again, going back to this leave calculator, it sounds so simple, but we were able to launch that really quickly after our hackathon in surge week, and then get that out directly to employees to test. So when we had them test this and get into the app, they weren't just testing that specific app, but we were also testing the larger, I guess, workflow of what... What's it like if we embed an HTML app within our intranet? Will employees go to it? Can they search? Can they find it? Are they using it at all? Are they still going to us for tickets even though we have this app? So I think, again, this is a really small thing, but I think you don't have to do large testing for every single project, but think more about how how you want to create things, how you wanna present things, and see if you can get people in testing it right away. And especially if it's really low stakes. So I think that was a really good example. And, like, we found that it did cut down on tickets and people were using it. So then from there, we've actually built a lot more, like, larger apps within HTML within Confluence as well. That's so good. What I heard here is product thinking. Mhmm. A product manager or product owner thinking is how can we validate some modular thinking? Like, you were... You're speaking about modularity there. Mhmm. We... It's not an HTML thing that solves something which validated not just that thing, the app, but also that we that people can utilize it this way. So I I just I just wanted to highlight that that's product thinking for us and starting to help us modularize and think about reuse of certain parts of what we have created. Also, they're not... You probably will set. When you prototype, you will start learning what you need to do with your input and how you need to work with your output and evaluating of if the LLM is go doing the right job. You learn how to build around it in a small scale. What works, what doesn't, what falls down, and what success succeeding without taking a massive risk. And I think this prototyping is something that every HR team should be doing, not just with AI, with anything. I mean, if we take Rebecca, let's take a talent acquisition and how we do interviews, for example. We just talked about hiring junior talent. How might we change the hiring process for junior talent? We can test and try and prototype, test different things in three different processes to see which one works better. We don't have to have the right answer. We can try things out and see what works, and that's the prototyping mindset, which I think is so important. Definitely. And we're actually doing that quite a bit even with our people team, itself. So I think sometimes, again, depending on the process, how complex it is, it can be quite complex to go to different groups in the business and say we're gonna try this thing with this group, this thing with this group. But we use ourselves as test subjects because it's much easier and it's lower stakes. So I actually just launched, our quarterly check ins, which is part of our performance process where leaders meet with their team members every quarter, talk about goal attainment, and give feedback. And so I wanted to incorporate some different AI, tools within that, including a cloud skill, they could actually... Employees could pull, you know, their data from their last check-in if they chose to to kind of look at that. And so that's one example. And instead of rolling that out to the whole company, we're just testing that with the people team first, as well, and it's completely opt in. So I think that's one example where, like, if it doesn't work great, I know everyone on the people team. They can Slack me. We can fix it, and it's a lot lower risk than testing it with, let's say, like, our GTM team, which is a lot larger. Wonderful. One of our clients, Experian, I can mention them because they they are... It's it's public information. They were speaking about this in our meetup, just recently. It was really intriguing. So we we taught them how to work this way, and they... In less than two weeks, they they had people come together for what they call kind of a sprint, development sprint or prototyping sprint. During that week, I think that that was one week. Each day, they built... So they were building a performance coach that were supporting people in performance management process, setting goals, which are better aligned with the strategy than before. And when building this performance coach, I remember them saying that they did really intensive days developing testing, already having booked feedback sessions with people. And every day, they went back to their tech partner and saying, here's what you're gonna build towards tomorrow. Tech partner took over, build it for the next day, and they continued the next day from continuous improvements, iterations, additional feedback, and again went back to their tech partner saying we need this for tomorrow. So they had also aligned their tech partner to this way of sprinting. And in just one week's time, they got to a point where they had a prototype ready. In two weeks time, they got to a point where they had all the other structures ready as well around this performance coach, performance management, including, including goal setting, including the communications materials, including training materials. So they did all of these parallel in two weeks to the point where in two weeks they were ready. It went into a, I think, longer process of approvals in the organization. And in in few months, it kind of came out and said, you know, with small iterations, it's now live. But in just two weeks, they were able to, when they utilize this way of working, get something out, which is highly valuable and utilized now by all of their kind of... I think 90 of the staff is utilizing this coach. So I think it's just remarkable when you shift how you're working to what you can get done. Instead of, Rebecca, what's usual is HR team sitting across different time zones, having Teams meetings, one meeting at the... In a week, Somebody's done maybe something. It takes ages to get anywhere. Yeah. Could you reflect on these two ways of working, the traditional, more project based, one project manager coordinating things, wrapping things together versus the iterative way that you are working in. You've been working with both. Right? Yeah. So I think what's what's nice about Personio is I think even before this larger AI transformation, we were definitely working in more of an agile way very quickly. I'm sure people in a tech a tech scale up can relate, and so lots of people were used to kind of shipping things quickly, testing, iterating, pulling the right people in. But I've definitely worked at jobs in the past too that were much slower and project based. And first, you have to do a project team, do a needs analysis. Like, it was a very long process, especially in learning design. And so I think it is a mindset shift and change management to think about how could we try something quickly. And some of it is kind of throwing perfection out the window. I think in last jobs, there was maybe a bit of fear of, like, reputational damage or not doing something perfectly. And so we built in all of these guardrails and steps, like, just to design a training, for instance. And so even when I work with people, I'm like, how how could we do this and get it to a place where it's 60% works, right, and then test that out. So even just giving people permission that things don't have to be perfect, that we're trying something new, that it can be, you know, 60% right, and then we'll we'll go from there. And I think I always use this. It's... People get very sick of me saying this at work but, like, being comfortable with the swirly face. So lots of things can be quite swirly and spending another six months on something and going through every process, sometimes it doesn't get any clearer until you actually try it and do it. So I think that's another... I don't know if that fully answers your question, but just... It's more of a mindset shift than anything and you can't... I think people oftentimes are like, okay, how can we create a new process to work more agile? And it's like, the point of agile is, like, you don't have as many processes in place. Right? Or, like, it might be simpler. So just being come up... Comfortable with that ambiguity. But the other piece then that I hear a lot and and that I've experienced is, like, the time piece. So, yes, we want to build more, we want to move quickly, ship things, but we just don't have the time in the day to take. the time to experiment. Right? So that's kind of another thing I hear, and so it's very valid. So that's the other question of, like, how are you building in time blocks to to try this new way of working and to focus on these things too? Brilliant. One of the things that came up to me when you just spoke was also that we need to be able to kill ideas. I mean, we need to be able to give up on ideas that we had and thought were good if the evidence shows us that it won't fly. And this is sometimes very, very hard for us HR professionals who are experts in our own area of expertise and we really know the answer. But if our users are not appreciative of a perfect solution, but they say, I'd only need 20% of that actually, I only... I just don't have the time to do this in an hour. Can we get something done which is good enough in fifteen minutes? That would be what I want or need. We need to listen to those things as well when we do design and develop. Yeah. I actually just heard the concept of de innovation, which is basically what can you subtract. And I think with AI as well and all of these processes that we have in the people team, it's oftentimes we just want to add more things to innovate, but it's actually like, okay, why are we doing this? What would happen if we just didn't do this? And we ask that question a lot now to kind of free up time, but also to simplify things so we can build more workflows and embed AI. Great. Let's talk about the development of the AI stuff, AI agents or workflows. One of the key things is starting to define roles and responsibilities. Who is owning the knowledge and the data and making sure that that's working well? Who's owning the process or the end to end? Who is owning the... How do we, you know, start building the testing and make sure that we got a testing parameters in place so that we know that we can continuously test whatever is improved. We can test that it actually works and delivers what we want. Who is responsible for the final KPIs? Who is looking at What kind of agentic workflows do we have across the whole HR? So talk about this a bit, the roles responsibilities. Yeah. And I think it's definitely a bit more fluid than it is in other projects too. I think the main... Two of the main roles that are a non negotiable is, like, the government's data security legal roles. So, like, what's that checkpoint? Who's owning making sure that we're being compliant and how we're using data and what we're proposing, especially to your point earlier when we're talking about people data. And the second piece is the data that we're pulling from. If that's not clean and we don't have someone looking after that and governing that, like, that can ruin a whole project, right? Especially if we're building automations that are pulling data from different places. So we've actually just, hired a new role on the people team, which is people analytics and data, to really focus on that people data piece and make sure that we're being consistent in how we're storing, using that data, we have that single source of truth. But even before we had that role, we had someone carve out, a larger part of their role to really focus on that because we knew that that would be a blocker. And I think what's interesting is in the building role at the beginning, a couple years ago when we started this AI journey, it was definitely, like, more of the IT team, the technical experts who were building. But now as people grow their skills, and also as Claude, which we use for our LLM, becomes so much easier and has so much more capability. Like, you don't necessarily have to be as technical anymore to be a builder. So that's something that we've kind of democratized a little bit. Like, if you see a problem and you get, you know, the first check that it's okay legally, that, we've got the data in place, you get a thumbs up from our AI champion, like, you can kind of try it and build it on your own, share it back. Like, we're not blocking people from building. And then with the testing piece, we often use our people business partners, so who are kind of like the partners already on the people team who are working really closely with the business to help us identify test groups as well. So I think again, like the roles are pretty fluid, like some of the things that I have built. I've kind of done almost all of them, not the data piece, but basically everything else on my own, but obviously with the blessing and with the visibility from our AI champion. Excellent. Thank you for sharing your your perspective into this. I've got a few... If I if I kind of go back to some of the cases that I've worked with and what we are looking at there, then it might be relevant for people here as well. I I think remember we had... Most of the people online are from smaller organizations, you know, less than 500 people. So you might not have a massive HR team. Other, like, larger global organizations might already have a proper HR AI teams working with this. But, Rebecca McGuire, you're also from an organization where you in the HR team needs to start owning this and building this yourselves. So a few things that I might just tip you off with who are listening here. First of all, when you start developing AI, tools... Let's just call them AI tools or part... Agentic workflows that you want to be live, out there, maintained, you already have to start thinking about governance around that when it's live. So who is owning it? The people who is the product owner, who knows that workflow in and out, who knows where the data comes from, how the knowledge sits where, how the LLMs are utilizing the knowledge, the access of rights to different knowledge needs to be audited, or it can't be a black box. It has to be... We have to be able to say, here is how and here is how this LM, if it makes decisions, how it makes decisions based on these rules. So somebody needs to own the product and actually know what it has eaten. The product owner is also what we call kind of making decisions of of what we prioritize to build or improve in the product, when we are going to release, when it's good enough to release, what is the level of quality that we can live with, for different types of products is different things. The the person who gets the user feedback and really looks at the user feedback, making decisions on improvements also when it's out there released and maintained. So, that's one thing that I would add to the discussion here. And then, this might require a bit of conversations with legal and cyber is what level of an LLM are we using and how much autonomy are we giving it? Many HR teams aren't in this mode yet because most of the LLMs and AI agentic workflows are still just automating rules based work. Like, if a, then b, if c, then d. But because the LLMs are growing so much stronger, we can say to it, like, we can move towards, hey. Onboard this salesperson with the knowledge data that we have in in here. You to, like, per purse... Personalize the onboarding according to a first conversation with them, analyze where they are, and assess their maturity, and then build an onboarding, journey to them. They are... They're starting to be so good already. So that makes the LLM make more decisions and gives autonomy to the LLM. So who makes that call of how much autonomy or autonomy are we going to give to an LLM? Where do we put a human in the loop to check on it or check on the results? Really intriguing conversations. What comes to you, Rebecca, when I talk about these things? Yeah. Definitely. And I think I've seen, you know, different posts and people talking about, like, from my space, like, work in the talent and performance space. So, you know, I'm seeing even fear around, like, oh, no. Like, could an AI LLM determine my performance rating and my comp. Right? And so that's something that we also need to be clear with employees of how we're using AI and how we're not on the people team, and we know that from the legal side, we actually can't, because of EU law, use AI to make autonomous decisions about employees. Like, it can help inform right, it can bring us the data, but as humans, we need to make, you know, final decisions around performance, talent, compensation. So I think part of it, this is kind of going off on a tangent a bit, but is being clear with employees about how we are using AI and being transparent. And then for things like onboarding, like, we do have different AI onboarding paths for employees, but the employee is also kind of in control and they can give feedback and adjust if they want to as well. So, like, one example is, we've created an onboarding path where employees can kind of test out of different topics, right, if they if they already know about it or if they... Depending on what job they're in. But they can also say, like, hey, I actually don't wanna test out of this. I wanna go back to that content. So we kind of give them an overall view of what's available for onboarding. We tell them what their path is, but they still, as humans, can make decisions about what they see at the end of the day. So more guidance, I guess, than direction. Excellent comment to add add. to that. Yeah. Super. Hey. When we develop, AI agenda workflows, how do we test? How do we build that testing loop, in your opinion? What's important there? We really need to be quite... It it requires a certain level of security and safety and and quality to be able to push on that button and say, we were gonna release this now. What do you need to do to get to that point? Yeah. I mean, I think we've already talked about the the security review, so that's something that we'd always do. And now we're kind of at a point with security where we've showed enough use cases to them that they have kind of given us permission to make decisions about certain things in these use cases without going to them, but we definitely continue to document and have a list of how we're using AI. We've got the guardrails, in place. The second thing, like I mentioned, is we always try to test with a smaller group. Oftentimes, it's people team or for managers. I have have a manager advisory group that will come in and test different things and give, you know, their perspective of... As managers, but also what they think their teams would say. So even just having the scrappy people that you can ping and say, hey. Like, by the end of the week, do you take a look at this and let me know your thoughts? But then the second thing there is, like, we're building now AI feedback loops within the AI tools that we're using, if that makes sense. So, like, this again is a really small example, but I built a cloud skill for quarterly check ins that helps people kind of pull from a different sources to see what they've done this quarter, what their achievements were, and helps them write their check-in. At the end of that skill, I have a quick feedback question like how helpful was this? What feedback do you have to improve it, and that Slacks the answer directly to my Slack channel that I've set up just for this feedback for me. So I think we know that survey fatigue is real. Sometimes it's hard to do, like, larger surveys and get feedback, but I would say, how are we building in smaller feedback opportunities within the tools that we're building themselves. Oh, that's so smart, by the way. Anybody who's listening, take this idea, build in right just in time feedback. Because when somebody's in the tool trying the process, that's when they are annoyed or that's when they. are happy, and that's when you get the exact right feedback to the point of what works and what doesn't. So it's really good. And now with AI, we can we can manage how... You know, as many data points as possible. One of the things that comes to my mind is... So when we develop AI right before we are launching, we have to test that it's given the... It's probabilistic. So AI is always probabilistic. It gives you the highest probable answer, and most of it is right if if there is, like, from a to b and there is no... Not not many different options in between. But we have to... If we change something. So say that we have a chatbot that is answering tier zero questions about our policies to the employees. We have... In the back... Back end, we've got 20 policies, two different countries, 20 policies being documented in a certain way, and we've tested that this AI chatbot is giving the right answers to a 100 questions, for example, top 100 questions about the policies. We can test, okay, it's giving the right answers to different kind of employee groups. Now we're gonna change something in the back end. Say there are three policies changing because of EU law, regulation, or something, and we need to change the documentation. We upload the documentation, the new documentation. We can't just trust that it's going to give the right answers now because we've changed something in the back end in the knowledge, repositories. Now we need to do what... What's called regression testing. So going back either with an agentic workflow that tests this or then with people that test, is it still giving the right answers? Because otherwise, we can't be sure. And if we're letting this go sour, if we're not testing continuously, we end up with the messy place where we've been in with in intra... Intranet work. Different types of documents in different places, nobody owning them. The rights... The one source of truth isn't there. We haven't tested. People aren't finding, and people stop trusting the agent. So Rebecca comes to you with my my my... No. It's it's a it's a battle wound that I have. I I mean, I think you said it better than I could. I'm sure everyone probably has experience with the fun knowledge management topic, especially with HR policies and things that are quite serious, and we have incorrect information that can have larger repercussions. So what we've done this year in tandem with our AI work is we've done a massive knowledge management cleanup because our AI bots are pulling directly from our Confluence. So, like, the AI bots are as only as good as the info that we have on Confluence. And part of it, honestly, was just getting rid of, like, 50% of our Confluence pages because we knew that we weren't gonna be updating them. We knew people weren't really asking questions about them. Like, in my opinion, it's better to not have any information than to have information that might be outdated and that doesn't have an owner. So we got rid of a ton of them, and the ones that were left, we assigned a clear owner to. We've also set up now, a Slack bot that uses AI to pull from Confluence, every month, and it goes to each page owner. So we've designated clear page owners, in Confluence. So So I'm a page owner for, I don't know, 30 pages. Every month, I get a recap of the 30 pages. It flags which ones might be at risk for outdated content based on other pages, based on Slack messages that are coming out. And so, you know, I don't have to look at every page, but at least gives me a reminder and then it's like this is read. This... You know, the content in this policy actually now conflicts with what this other page is saying. Please take a look. I'm it makes it a lot easier. hear this. I'm. so happy to hear this because some of the HR leaders that I work with go like, we don't even know where to start, and we can't even... I I just say, start with your documentation. That's. something do even if you haven't started with AI development at all. You need your own house in order and cleaning up and doing that. It's tedious work, takes time. But if I would be a HR leader right now, I would make that happen in a month. I would say anybody who owns anything, you make sure that your documentation is in order and you've got a month to do that. Yeah. And here's the... Also also what's really helpful is having a format that is shared. Here's the format for policies. Here's how you write exceptions. Here is how you write your country blah blah blah specifics. Here is how you blah blah blah. So we have established a format for those knowledge documents. You're so far ahead already if you do this because then it's so much easier to tap into that knowledge space with any AI development. Definitely. Yeah. I think... And it's not fun. Like, when you think about AI, like, it's not all the exciting work. I think if you take anything from this call, so much of it is just the cleanup and really removing things. Like, you can't automate complexity, I think is a phrase, like, we always say. And so if your data is complex, if it's not right, if your confluence is a mess, like, it's not, AI isn't gonna work the way you need it to. So just, yeah, clean up a bit first. Great. When we spoke about having this conversation, I really liked one thing that you said, and I'm gonna kind of give it back to you, Rebecca, and and have you talk around it. You mentioned that, there might be this kind of difficulty in starting and getting something out there and being happy with it because, you know, when you're developing it over a few months maybe, the tech has already changed. There's something better, faster, more fancy around in the AI world or already. And the thing that's out there is kind of outdated in a few months already if you're in a really high-tech environment. Can you share can you share some thoughts around that, that we still have to move. Right? Yeah. And I think that's gonna be the case no matter what, but I think AI just exasperates that because it's... The technology is moving so quickly. So, yeah, like you said, Riina, oftentimes I'll build something and by the time I launch it, it could have been done better or a different way, especially with Claude, which we use internally. They're constantly having new releases. But I think you have to work with what you have now, but that goes back to the point of, like, just like your knowledge management, what are your checkpoints to review the things that you've built with AI to see could they be simplified with the new technology? Could they be improved? And a lot of the improvements are actually amazing. So I don't know if anyone here is using innate end for automation, but I, again, am not a technical person, but there's a lot of things that I wanted to automate. So Claude, at first, would help me kinda describe what I needed to do in n eight n, and I would go do it. But now we have the technology to connect Claude directly to n eight n. So I tell Claude what I want, you know, in a simple prompt, and it creates it completely on its own in n eight n. Right? So I still have the n eight n workflows that I created myself. They still work, but now I'm just doing it a lot more efficiently. So I think that's another example of, like, it doesn't make what you did before obsolete, but, you know, you can probably count on whatever you're doing now this month will be a lot easier to do in six months, I think, which is exciting. One thing that comes to me when when you say that... So I haven't done automation with n a 10, and what comes to me is if I'm a CHRO and I I know that if I'm starting to hire people who've done this, they are going to start automating the heck out of bad processes. Right? Because it's it's how you learn. It's how you think. It's how you've done that. It's it's like people coming in and just knowing how these things work compared to me, for example, who have to learn this. Right? I have to really sit down, take time, block calendar for a day, and start hacking away with call because I got so much other other things to do as well, which you said is kind of time time there. Now if I'm a HR leader, do I need to understand this on a deep level myself, or can I trust my team to actually do this well? What's your view? So it's very interesting. We actually just ran a larger AI assessment survey within Personio and got a lot of great insights. And one of the insights, I've just analyzed the data so it's top of mind, is that our senior leaders, compared to our frontline managers are less comfortable with AI. They're less likely to share if they don't know anything. That was kind of a question of, like, psychological safety to ask for help and share what you don't know. They're less excited. They also see less value in AI. So I think that's already a disconnect. Part of it is because they're not expected to use it in their day to day works. Right? Like, you need to be able to think strategically and long term vision. You're not necessarily on the ground automating a process. But I think this is illuminating some gaps, especially in strategic planning and visioning. So we actually had a quote from one leader, one senior leader who said, like, I see the value in AI. I'm just not sure how to translate that into my strategic planning and vision because they didn't know what was possible. So I think, again, like, there's different job levels. You have different role expectations. But I would encourage any senior leader to at least spend time tinkering in it to know what's possible so you can use that for planning. And so you can also, like, have conversations with your team and support them because they are moving a lot quicker. Here's where where a few things come to my mind is I'm going... In my mind, I often going into, okay, what... What's the kind of executive level, the HR leadership team going to do with this? And I have a few things that come to my mind. First of all, if you have that AI architect or AI champions, they should be able to translate things between the exec level, the strategic level, the planning and prioritization level versus what's going on versus what's going on on the... On on more of the... On the feature and brand level. And, second of all, if I... I I just kind of said that if I'm a CHRO, I have a task workforce of, say, 50 people or 200 HR people or 500 HR people, I want to also know where the agent sits and what they do and who's built them, who's owning them, and how much they are ticking away of the token costs. So I'd like to have a dashboard of my agentic workforce as well and understanding because at some point we will come to a point where we need to start, I'm sorry, but crossing over some agentic workflows because of cost or because of prioritization or because of architecture, and we need that visibility into what we're handling. What comes to mind when when I say this? I mean, I don't probably need to know all the details about automation on on the kind of process level, but I have to have an overview of of this agentic workforce. Yeah. A 100%. And I think that's exactly, yeah, what we've started doing in Senior Lead, People & Leadership Development @ Personio even with this AI usage dashboard. So any senior leader the frontline manager can go in and see their reporting line, what the AI usage is like. And we just, it's it's a bit of a, I think, spicy topic, but we've just implemented token limitations for Senior Lead, People & Leadership Development @ Personio, with a really, really open easy way to request more tokens if you need to. But in order to request more tokens, you have to document and share what your use case is. So I say that because it gives our senior leaders a way to see, you know, token usage. We've got the dashboard. You can see cost spend. But then we also have this data that's coming in from this AI automation where people are requesting more tokens to see how they're actually using it. And you can see, you know, is there crossover? Are people doing the exact same thing? You know, how can you connect them? And I think there's just lots of different things that you can look out there as well. But I think, yeah, getting that data on how people are using it is really important. Just yesterday, I had a conversation with a with a very senior c level CHRO with large organizations, and she said what comes across to her in conversations with executive HR people is that people come to them and say, oh, we automated this and we saved this much capacity from our our team and blah blah blah. But they are not telling what do they do with that extra capacity. Where is that going? Is it going towards more strategic items that we need to get delivered or what's happening with the capacity? So I think that was a really good call out as in if you're going to be more efficient, you need to think about what you do with that efficiency gain as well. That is a leader conversation. That's somebody who sits on the resourcing budgets on the people's capacity, that has to have an answer to that, or at least I would demand an answer to that. Okay. Great. We invested 50 ks in developing an AI agentic workforce. You are saying it saves you 30% of 10 people's work. What do we do with that capacity? I mean, it has to be a logical flow of of capacity management. Rebecca, what comes to you? Yeah. No. Completely agree. And I think there's no shortage of work that needs to be done. So I think it's not an issue of, you know, we're automating things and now we don't know what to do. But to your point, it's more of being able to prioritize now. I think clear priorities, and short term goals are more important than ever When you're able to kind of get rid of that business as usual work and automate that away, that's like now what do you do? So that's why we're focusing even more even in our performance process and these quarterly check ins and identifying goals. We also have a question on there of like how are you using AI and kind of what are you doing with that time saved? And then I think the second piece is it's a great opportunity to focus more on those growth and development conversations. So even. in our quarterly check-in, that I just launched yesterday, we added in a short question for every employee asking, like, what energizes you and your role and, like, what are your thoughts on where you'd like your career to go? Because we know roles are gonna continue to look different, right, as as things change, like we can't escape that. But what we can do is look at this time that employees are saving and think about how we can help them invest it in the areas that energize them and help them get to where they need to go. Because there's no shortage of work to be done, but it's really, yeah, that strategic level of how do we know where people wanna go, how do we engage them, and how do we help them get there. I agree with you. We are coming closer to a kind of ongoing talent management or talent enablement, way of working as Mhmm. if you're leasing capacity somewhere, you need to look at the skills, you need to look at what's needed to be done. We're probably moving towards a more project based or talent market type of a thing where people can release... They can say, oh, I'm in this project for two months, then I'm gonna go do this other work releasing capacity from there. I've got 50% to put on something else. We're probably in three, five years going to be more of a fluid fluid workforce where people can be bringing in their knowledge expertise into different projects according to the prioritization going on. One of the things that comes to me is, I think I love the IKEA's example Mhmm. where all the customer... When they started bringing in agentic workflows to their customer service, where agents are now answering customer's questions and taking care of most of them, they actually turn their customer service agents into interior designer, kind of... So bringing that human interior design expertise into that. So they repurposed, the workforce or... And upscale them, reskill them so that people can actually find new careers when they have to leave something behind because of this automation. I find that really a beautiful thought. What is your thought. there? I agree. I love that example too, and I think that always comes to mind. And, like, we're still strategically figuring out what this looks like at Personio. I think part of this will tie to our 2027 strategy and plans, which is being discussed, this week. But, part of this right now is, like, the first thing that I could get ahead with somebody on the people team is knowing where these... Where each employee wants to develop and, like, having that documented and knowing where they want to go. And I think the second piece there is when we see what are our strategic priorities and bets for 2027, Workforce planning, you know, where are the gaps? Where can we invest more? And then from there, we can see, you know, is this person actually pretty well equipped to do this other task? How can we get them there? So I think it's, yeah, it's kind of holistic knowing your employees, knowing the strategy, and then doing some some matching there, which is, easier said than done, but I think it's something we'll all start facing even more. Excellent. Let's wrap up with talking about the place where we are releasing, stabilizing, and when it's out there running and improving things. So, if we have a workflow now that we have prototype, got validation of it's going to work, we got the investment, we develop it, we got the tech part sorted, we got the automation, we got the infrastructure architecture, the integrations, we got the testing done, we're clicking on the button to say release, it's out there to be used by people, we are getting the feedback from the continuous feedback loops, we are improving that, stabilizing that it works. If something... We already came to the point if something changes in the back end, we have to make sure that we test that it works as it's live. What about the part of where it's out there? Who is going to own it? Who is going to make sure that it's still valid and doesn't become something that sits on the shelf and nobody uses and it's not valuable at all? So the kind of run and improve part. Yeah. I I think you mentioned this earlier, but having that product mindset. And so we already tried to do this on the people team of, you know, everything we do is a product. So who's the ultimate owner, person accountable for that product, whether it's onboarding. or a people operations compliance task? Like, this is no different. So oftentimes, the things that we're building, whether it's an automation or AI agent or app, already fits under one of these larger products. Right? Or it should. We shouldn't just be building things that don't map to what we're already doing. So that person, let's say, like, I own onboarding, I know also responsible for this onboarding agent. And I might not be the technical expert. Right? So I know our onboarding leader created a really great onboarding agent with IT that's quite technical. She knows the basics, but she's responsible for monitoring usage, for flagging issues back to the IT team. So I think it's not it's not overcomplicating it, but these AI products now become part of our larger suite of products that we own. Fully agree. I think this shouldn't. I'm on the fence with people saying we need an AI in HR strategy. You know, we need to have an AI road road map. What what AI things are we going to develop? I'm kind of in... On on on the fence with that. I think, yes, in the start, we will need that, but it needs to move more towards what you said. It needs to be part of the work that we do anyway. Like, I wouldn't say we need an email strategy. It's just. part of... Say we need know, calendar strategy. It's just part of work, and this will become as a natural part of work as anything else. One thing that came to my mind is... So the people that I speak with, who knows about architecture, they are worried. And why are they worried? Because quite many organizations are choosing the path of, yes, here are LLM tools, start automating, start building your agents in your own work, and it becomes very siloed. So each silo is building their agents or automating some work, but nobody's really looking at the whole thing. What does that lead to? It leads to architecture where you have a 100 people owning 500 agents, all burning tokens. Nobody has looked into dependencies and kind of the whole thing. It will lead to a mess which takes a long time to untangle because people fall in love with their agents. And I have had to say that from own experience, it's really difficult to kill something that you've built and spent, you know, many days on and improve and improve and improve, and then somebody tells you this isn't needed. We don't want to invest in that. You have to do this, you know, in another way or we need to connect it to another workflow. So the migration and sunsetting of these things, we still are to learn how that works. Anything that comes to your mind? Any tips? Any tools? Anything that comes to your... I mean, I think it goes back to just something as simple as having those AI champions at the company level with the departments and down to the teams where it's really their job. Like, we're documenting all of our use cases now. They're seeing the connections. Like, when I go to create something, now it's second nature that I look at this Google Sheet to see, like, oh, is there anyone else who's created something similar so I can go to them? So... Yeah. I mean, I think it's a lot easier to do that at the beginning than go down the path of creating something and then have to change it. But, I mean, I'm sure this is still happening, like, in isolation. Right? So I think it'll be a larger topic, but really trying to think about who is kinda owning that central overall view and what are your checks right now as people are creating things. Like, do they have visibility into what else is out there that they can use? Because I'm sure people would rather use something else than create something new themselves at the beginning anyways. Oh, for sure. And that that visibility and that access is is quite. an important decision because why keep this away from people is the other question. I mean, Mhmm. there's these two... There's a tug of war between go and do, but then, oh my god. I I don't wanna end up in a messy architecture. Yeah. 100%. And I think we're still figuring it out. Right? We are. So we're all are. we're we're all kind of in the same boat, and, I think that's why it's so good also just to learn from other people and to hear from other HR teams because I think, yeah, we're all we're all learning as we go, and we know that things will only keep changing and, improving when it comes to AI too. Questions. I got something in my throat. I'm sorry about that. Oh, no. Could, could we have, Phoebe help us? Do we have any questions in the chat or can't see? Phoebe, can you, write a message to us? Do we have any questions, coming in? If you have any questions or any comments, we'd love to hear that with Rebecca. Not quite sure with it. Sorry about... This is the first time I'm using this, gold cast technology and being in back back end. So maybe we don't. So what we're going to start doing is, wrapping up. Where... I'm going to wrap up with a difficult question, Rebecca. Alright. I'll try. What are you most intrigued, enthusiastic about regarding AI and HR, and what are you most worried about? I think it's the same answer for both. So I think we talk and we think a lot about the technology piece. Right? Like, we just spend time talking about piloting, building AI from that product mindset. But there's a whole another layer around people in HR, especially for me in talent and performance, and that's how are we ensuring our job architecture, our role expectations, our workforce planning, like, how does that adapt when people's roles are changing so quickly as well? And, like, what does that strategy look like? Even our performance process. Right? So I don't have a clear answer, but I do know that the static role expectations that are over a year old have already changed, and we can't just keep updating them as AI changes. So what does that look like? And we're having lots of discussions about that. And so I'm excited about that because I think if it's done in a way that's thoughtful and employee centered, it can really create great opportunities for people to develop and grow and create that clarity throughout a time of a lot of change. But I also think it's a lot to do, and it hasn't been done before. Or if it... If you have done it, you can send me a message. I'd love to hear it. But I think, yeah, it makes me nervous and excited. What about you, Riina? So I'm really intrigued. I'm so thankful to be able to be in where I am right now. So I've got twenty five years of experience of organizations and people systems and the organizational design and the architecture of how work is done. And everything's changing right now. So it... It's such a complex nut to crack for my brain that I'm really enthusiastic about that. And each organization is different depending on where they are on the journey, what their ambition level is, what their maturity is, what their talent market is, how... So what I'm talking about is the organization decide and how will the organization set up to work, the value streams, the dependencies, roles, responsibilities. What do we do with leadership going forward? I mean, leadership changes totally in an AI native, agentic, kind of work, setting. It's so intriguing that everything we worked with for so long is now changing. At the same time, I feel grounded because there's so much nonsense out there, so much hype, and so much, let's just fix this with a... With AI. And you go like, well, no. You know what? I know what is stable, what's this... So we'd be sustained, what's human, where the ethics are. So that that whole thing inspires me quite a lot, the kind of the system changes. What what is worrying me? What's worrying me is... I can't put my finger on how to say this. People become more stupid. If I can say that, me included, if I'm not, you know, continuously awake. So how can we make sure that we're not giving up on what makes us so unique with our brains and our cognition and our understanding, our humanity, and just start trusting whatever AI will produce? How can we make sure we've got humans in the loop where they need to be? How do we make sure that we establish rules that you are responsible for the results? Whether it is you or your agent having developed that, you are responsible. So that that kind of going where... And we can see this already in certain schools. My children is like, they're not learning because they're just putting something in the chattyPT and like, no, that's not good enough. So that's my worry. What's happening with our cognition in a broader scale, if that makes any sense? Yeah. No. Great answer. I think completely agree. Like, even myself, I'm, like, I'm not reading this document. I'm asking Claude to summarize it for me when actually I'm capable of reading the document. Right? So I think you can even feel your own brain changing sometimes. Exactly. When you see yourself making very easy mistakes, I work on something, I create a report. I think I've read it through, but the next day I read it through again, be like, this is not good enough. What did I think yesterday that it was good? So you probably need to start understanding your brain differently as well and giving it a bit of space to go back to your productions and look at them after a while. Because when you're in that moment of working with AI, you just think everything is so great. Wow. Yeah. What it produces is fantastic, and it's actually not always fantastic. So also learning how to work in new ways by... When you... With your own work. I think we are wrapping up for today. Thank you so much for the collaboration between Personio and, us, Agile HR community. It's been a wonderful discussion, Rebecca. Yeah. Thanks so much, Riina. And, yeah, really good questions and good to hear from your perspective too. And I think, yeah, it sounds cheesy, but we really... We're all on this journey together and figuring it out. So thankful to have communities of other people doing the same thing. And sharing is caring. So let's share share what we learn and where also where we fail. So everybody might end up learning a bit faster. Thank you everyone who spent time with us today. Hope to see you around and be in touch with Rebecca or me if there are anything that you want to follow-up with. Thank you. Yeah. Thanks. Bye. Bye.