Personas You Can Talk To: Turning Research Into Persona Agents

Boris Divjak at AI x design

Setting the Stage: Design Starts with the Customer

Speaker H (Boris) opens by describing his design team's mission at ANZ to keep customers close to decision-makers, and poses the challenge of maintaining that customer presence in a large organization where most employees don't interact with customers directly.

The Persona as a Focusing Device

Boris traces the history of personas back to Alan Cooper's late-90s concept of designing for a specific archetype rather than an abstract demographic. He explains how personas help teams avoid designing for a non-existent 'average user' by putting a vivid, specific person into design conversations.

The Limits of Static Personas

Boris discusses the core weakness of traditional personas: they capture headlines but not the underlying research detail, forcing teams to fall back on assumptions when tacit knowledge from the original researchers is unavailable. He notes that valuable research often sits buried and hard to find in tools like Confluence or SharePoint.

Introducing Priya: A Persona Example

Boris introduces Priya, a fictional sole-trader dog groomer persona built for the conference, to illustrate a realistic but static persona poster. He explains how even a detailed poster leaves designers guessing about specifics, such as how Priya handles invoicing, because the real texture of research interviews remains hidden and hard to retrieve later.

Why Research Insights Stay Hidden in Organizations

Boris summarizes the core problem: organizations sit on a treasure trove of deep customer research that stays largely inaccessible, and asks how AI might help turn static persona artifacts into an ongoing conversation with research insights.

Live Demo: Talking to Priya, the AI Persona Agent

Boris presents a video demo of an AI-powered version of Priya that answers questions in first-person, conversational style about invoicing, tools like Xero and Square, and banking needs. He shows how follow-up questions can probe deeper, revealing specific, actionable, emotionally grounded insights that go beyond what a static poster could provide.

Grounding the Agent in Real Research Data

Boris emphasizes that unlike a generic chatbot, the Priya agent's responses are traceable back to source research narratives and interviews, ensuring trustworthiness. He shows side-by-side examples proving that her answers about tools like Xero, Square, and PayID come directly from underlying research material rather than being invented.

Building the Pipeline: From Research to Persona Agent

Boris details the scalable pipeline behind the persona agent: conducting and anonymizing interviews, extracting and tagging insights, crafting a narrative covering the persona's full lifecycle, and finally building the conversational agent with carefully framed instructions grounded in curated source material.

Design Patterns: Grounding Rules and Fallback Responses

Boris highlights two critical design patterns for persona agents: strictly grounding responses in source knowledge and building in an explicit fallback rule allowing the agent to say 'I don't know' rather than fabricating answers, since AI models tend to want to please users with invented responses.

The Temptation to Automate Everything

Boris explores the possibility of automating the entire research-to-persona pipeline using AI, including transcription, anonymization, tagging, and narrative crafting, which could keep personas continuously refreshed with new interview data, while acknowledging this raises new challenges.

Scaling Challenges: Tools, Curation, and Instructions

Boris walks through practical lessons for scaling persona agents responsibly: using already-approved organizational tools like Copilot to speed adoption, the critical importance of simple and well-curated knowledge sources over dumping in excessive data, and giving agents clear, narrow instructions with fallback permissions.

Keeping Humans in the Loop: Don't Stop Talking to Customers

Boris stresses the most important caution: AI personas cannot replace ongoing customer research or researcher judgment, and warns against letting AI create additional distance between teams and real customers. He advocates building persona agents alongside, not instead of, direct customer research.

Closing Thoughts: Good AI Gets You Closer to Customers

Boris concludes by reframing the talk's opening principle: good AI should get design teams closer to their customers, just as good design starts with the customer. He summarizes how persona agents can be a more adaptive, conversational tool than static personas while cautioning that AI must never quietly replace real customer research.

Q&A: Updating Personas, Reliability, and Reconnecting with Users

During Q&A, Boris admits he hasn't yet solved how to seamlessly update persona agents with new research and discusses challenges around AI model drift and reliability testing. Audience members share reflections, including one comment praising how the AI persona helps combat researcher desensitization and re-humanizes empathy for users, especially for younger designers lacking field research experience.

Thank you all. Hope you all enjoyed your lunch and are ready for a little bit more AI action. So, yeah, let's get on with it. Good design starts with the customer. Now at ANZ, we try to live up to that principle every day. Certainly in the design team, I look after a small design team, team of designers focusing on experiences of small business owners specifically.

Now a lot of our work involves keeping customers close to the people that are building and delivering experiences for them. So close enough, let's say, that a product owner and a designer arguing about the feature on a Tuesday afternoon, you know they can really picture that person in front of them, So that's our goal at the other end of a feature that they are designing for.

But you know that's quite hard in a big organization, right? Like we're an organization of thousands of people and so how do you keep the customers present really when decisions are being made every day, multiple times a day about design? When you know we know that a lot of our stuff, you know they don't really interact with customers directly every single day, right?

So it's quite important to keep the customers present in a different way. But how do you do that, right? Now for years the tool we've used for that is the persona which hopefully most of you have used at some point or another. You can see me on the screen there a few years ago, the way we used to do design.

I'll just let that run-in the background as an example of how we use personas sometimes in design to focus. And this idea goes back to Alan Cooper in the late 90s when in his book Inmates Surrounding the Asylum he was arguing that designing for a specific person can ultimately lead to better product design decisions. So a person that's basically like an archetype with a name, a job, goals, frustrations, specific enough that you can really picture them when you're thinking about what kind of feature to design and how to go about it.

So not designing for demographic, designing for the average, not trying to please everyone, right? That ultimately leads to better products when you're thinking about that specific person. And when this is done well, know, teams really do stop thinking about users in the abstract and they start thinking about okay, what does this specific type of user need?

And so that does help. So persona is a focusing device, right? It puts the person in the room when decisions are being made effectively, right? It helps teams avoid designing for the average that doesn't really exist and that's why we still use them a few decades later, right? Hopefully most of you have used them at some point.

Yes, there was a cat. Well in my experience and you know hopefully in yours as well, perhaps the persona doesn't always have the answers, right? That's one of the challenges. It is part of the conversation but when the decision needs to get made, it doesn't really have the detail we sometimes need.

And so teams fall back on assumptions because the persona just doesn't hold that detail. Like if you are actually part of the interview or the research, you may have that tacit knowledge and you may remember what the real need behind it is. But the persona really is built to focus, it's not meant to provide all the details.

And the frustrating part is that a lot of that information usually does exist, right? It does sit in perhaps like in an interview transcript somewhere or maybe in a research report. Very often it's somewhere on Confluence, in a SharePoint folder, maybe in a repository tool of some sort that people don't always remember where it is and how to open it and where to find it.

So the question I keep coming back to is what would it take to keep the research and the customer present when those decisions are being made? Now let me frame this with an example. So throughout this talk I'm going to be using Priya here as an example for the persona. And so this is a persona I built specifically for this conference.

I'll talk about that a bit more later. Now she is a sole trader. She owns a mobile dog grooming business. She kind of operates in the inner suburbs. She's digitally savvy. She works, she banks with ANZ, of course. Yeah, she was built for this talk, so what do you expect, right? And I'll keep using her as an example.

You know, she is realistic. She's basically what we would expect based on the real research we do with small business customers. It's a fairly standard profile. I'm guessing if you've ever worked with personas, if you've developed one yourself, you've probably done something similar, right? So you can see there's an illustration, there's maybe a quote, there's some goals, banking needs, pain points, fairly standard, right?

We always kind of do very similar things there. There's actually quite a lot of detail on this one but it still may leave you guessing, right? Because everything here is frozen, it's a static page, like everything is basically a headline. Now behind every headline there's an interview, right? And so what exactly did that person say when they were talking about chasing invoices? You know, how did they describe a quiet month?

Maybe how did that feel like? Right, so there's a lot of that behind that we don't really see. Like the poster is a summary but the texture stays buried. And, you know, when the team is still present, the team that was doing the research, I mean, that's a little bit easier to manage because they do have the tacit knowledge.

But perhaps the project moves on, right? And maybe six months later someone else is like, okay, need to design something for a very similar type of customer, so let's use Priya. And so maybe they have a question, how does Priya actually handle invoicing? So what works for her, what doesn't? And the poster kind of gives you a hint.

There's a mention there that says there's a goal, get paid faster and chase fewer invoices. But that's not really giving you all the detail, right, that as a designer maybe sometimes you need. And so getting to that answer, I mean it's sometimes possible, right? But it does mean having to track where the research is hidden, you know, trying to find the right project, talk to the right people, have the right search tags, the keywords, you know, spending an afternoon perhaps trying to dig up what the research is.

And so very often again teams fall back on assumptions, they guess. And so Priya is a poster pinned up on the wall but maybe of little use in some way. And so we're sitting on this treasure trove of research, lots of research we do in big organizations, lots of really deep interviews. A lot of that is very difficult to really get our hands on, right? So it stays hidden.

Repositories keep growing. And that's really the problem I'm trying to solve here, right? Now we're at an AI conference, right, so how can AI help? I think that's actually quite exciting because AI has lots of potential in this case. I think AI can enable us to have an ongoing conversation with research insights. I think this is really quite powerful when you think about it. So when you apply AI to a persona, it doesn't have to be a static artifact.

Basically it becomes something you can talk to. You know, personas maybe used to be documents we handed over, but now they become basically this ongoing conversation. And so that's something that AI is really good at, right? It's making sense of like big amount of unstructured text and then kind of summarize it, send it back to us.

I believe we've all kind of used that feature very often when we get along, know, you know, summarize it. It's pretty good at that, right? So in that sense it's a very powerful tool, right, for research and I think that there's a real shift there when we can use that with personas for example. But rather than just talk about it, right, I

mean

it's a talk, right, but still we can show some things. It's not a demo but it's gonna be a video. This is what Priya looks like in practice, and again disclaimer here, I built this just for the conference. We do have different personas we use with ANZ. They're built with our own research data.

This one was built just on publicly available data. But basically works exactly the same way, The mechanism is exactly the same And so it kind of works like this example. And so yeah, think let's ask her a few questions. Let's see how she does, let's see how she responds. So the first question I wanna ask her, how do you handle invoicing? What works, what doesn't?

And you can see she's going to respond very much in her own voice. She's going to talk in first person, very conversational. She's going to be specific about for her needs specifically, what works, what doesn't. She talks about being paid on the spot when she can. She works in a van, moving around a lot. She talks about using Xero when she has to invoice commercial clients.

And she talks about an emotional angle. Think that's quite interesting. She gives an emotional angle about how chasing late payments feels awkward and uncomfortable, and she doesn't want to spend too much time doing that. So quite a lot of information already, right? Now I do want to dig a bit deeper, right? And so I say, okay, what tools exactly do you use for invoicing and payments? So probing for a bit more detail, the answer gets a little bit more specific.

She doesn't just list the tools. She actually talks about why each tool fits, what works, what doesn't. She gives the detail that you normally kind of have to dig through research to find rather than just look at the poster. But here you can do it on demand, and I think that's the real benefit. Now third question. When do you need help from your bank?

Now I've kind of asked a question here that's maybe a little bit too high level. Right? I'm like, okay. Maybe I'm not happy with this answer. I wanna be a bit more specific. But that's okay. I can follow-up because this is a conversation, so it allows me to do that. And so I follow-up trying to better understand, you know, what how she handles payment issues, for example.

And, you know, she talks about really quite a bit detail that I can, you know, very comfortably apply to some kind of a design design flow. She talks about not wanting to call the bank all the time because she doesn't really like to stay on long phone calls. I'm sure we've all experienced that at some point. But when you're working out of a van, in Priya's case, it's very applicable. And so this gives us quite an actionable insight there.

It's a very tangible picture that we can use to design specifically for that type of customer. Right? So trying to really help her out. Now, yeah, two things to notice there. I think there's couple of patterns, right? So first, this is like a persona, right? So she's talking about very specifically in her voice about her needs.

So it's behaving in a very similar way to that poster. It's just a specific customer. And second, the conversation keeps going, and I think this is really important here. It's not just a poster. It's not tangible artifact, a static one. It's basically an interaction, so we can keep going back and forth. I've asked four questions here.

You know, I could have asked 40 or more, and I'm sure I have in my actual work. I think that's the real shift. I think, you know, we talk about AI a lot, like, yeah, it's there's dangers coming with it, right? We're all a little bit scared of it. Big big change happening. But I think there's also things to be excited about, and I think that's, you know, something that excites me is what's possible now that really wasn't possible before, right?

So, you know, this kind of example is exactly that. It's like we couldn't really do that before we had AI, right? You can't have that conversation with the customer but also not the customer there I suppose. So yeah, can follow-up, you can probe specific scenarios, you can ask her to prioritize what's more important to her. I think an important element that I still need to talk about is that this is all grounded in research.

So this is not just a model making stuff up. It's not just a chatbot in a costume, if you will. So it's really a conversational layer over the research that's already been done, right? So it's sitting on all the data that we already have. So yeah, it's grounded. So just to show you what I mean by that.

So on the left side here, you know, you can see the answer Priya gave. Then on the right side, there's the narrative that's being supplied to her as the source material, right? And you can see that, you know, she talks a little bit about invoicing using zero, and that's taken straight from that narrative, right? You can see how she talks about using Square in the van, again taken straight from the narrative, you know, using pay ID, same thing, right? So we're trying to really develop this in a way that's quite traceable, right?

So we can trace everything that she says back to the narrative and then back, you know, if we can to the original interviews where that happened. And that's I think quite important, right, because traceability turns something that's a clever answer into something that you can trust a little bit more, right? So we're not asking designers to trust into AI, but rather we were just using AI to surface research the designers already trust. And so I think that's quite important, right, because how do you know can I actually make a decision based on these outputs?

So yeah, not magic, right? I mean, we know AI is not magic. But yeah, taking the persona, putting that into some carefully framed instructions prompt, and then having some really well curated source material. I think we talked a little bit about curation earlier today. Think yeah, that's definitely very important in this case as well.

And yeah, so the real work really happens upstream, So it's not so much even in the agent itself. It's more like this. So it's like this is the real work. This is a scalable pipeline. So how do we develop that pipeline from the source material? So from doing research, from doing interviews with our customers to then extracting those insights, making sure obviously that's all anonymized, extracting the insights, tagging, categorizing them. So this is all stuff maybe you already do, right? And it's important to keep that process in a way so that we can have some rigor in our research.

And the third step there that really is where the persona starts becoming a little bit more like a persona is turning those insights into some kind of a narrative. And so that's what I was showing on the screen earlier. So the source material I'm using here is basically a narrative that goes through the entire life cycle of a of a persona, trying to understand what are all the different kind of things people might ask her, what are all the jobs to be done, if you will, and trying to provide some sort of a narrative for each of those. And yes, using obviously the insights from the research to inform that. So yeah, that's the fun part in a way.

And you know, the most fun part I suppose turning all of that into an agent and that's where, you know, it really starts feeling a little bit like magic in a way. And, yeah, the agent itself, I think it's really important here to provide it with some limitations in the sense that we do want it to refer to that source material.

We don't want it to, like, grab everything from the Internet and try and just make stuff up. So yes, some carefully framed instructions there for sure. And obviously carefully tagged insights and all that. And so most of that careful design lives there in the agent. There are two patterns I'd like to point out.

So first talked about the grounding, right? So there's instructions that you can only use the grounding from the source knowledge. And the second one, which is also quite important, is to have a fallback rule. So what does Priya say when, you know, that knowledge isn't there? Right? So if you ask her something that's not in the source so it's quite important to actually provide that to give her the freedom to say, don't know.

Because, yes, we've heard also today, you know, AI is really, you know, the way it's developed is in a way that's quite happy to please us. Right? So it's trying to make stuff up if they don't know the answer. They're just trying to provide something to us. And so if we give them that freedom to actually say I don't know when that happens, if you explicitly frame the instructions in that way, that definitely helps a lot.

I think it's something to think about and experiment with. Now if we don't have that, try to trust and quickly disappear and, you know, we can have made up answers. If there's a made up answer, people start questioning it and then the whole value goes down the drain. One more thing I kind of want to point out here, there's a temptation and I think definitely a possibility.

Right? When we have this sort of a pipeline, you could say, why not just automate the whole pipeline? And, you know, it's certainly something we're experimenting with a little bit. You know, you could automate maybe not the interview itself. I think it's good to have customers to talk to. But, you know, transcription, anonymization for sure, tagging the insights, extracting the insights, you can do that with AI, right?

Then crafting that narrative, you can do that with AI. It can actually summarize all that even with the parameters of a persona. And so that's potentially exciting, and one of the exciting things there is that it could allow us to have a persona that keeps being most recent, that kind of keeps being alive. So let's say when we do new interviews, maybe that can feed automatically into this pipeline and the persona keeps being refreshed all the time. So that's pretty exciting.

But there's also some challenges potentially that come with that, so I wanna talk about that a little bit next. So the question really isn't whether, you know, we can do this. We know we can, but it's more like how do we do it in a way that makes it easier for teams to adopt. So, yes, scale brings bigger rewards.

I think with AI that's very tempting. It's tempting to automate everything. Very tempting to think about, okay, how do we feed in as much information as we can? Let's just feed it all. Let's feed the monster. But, you know, Skelt brings bigger rewards but also new challenges. So I've got a few points there to just talk through about how to go about those challenges and what they might be.

It's not comprehensive. I'm not talking about managing the system itself, managing users, all that. It's basically an IT system of its own, right? So there's all sorts of challenges like that to manage. But in terms of it being an AI persona, right? Like some of the things I've learned, think it's great to use the tools you already have, right, in the organization you work in.

Now at ANZ, we use Copilot. So this was built using Copilot. That just makes things a lot easier because it's already an approved tool. I don't need to request for 10,000 licenses or something like that, or I don't need to go through procurement processes. A lot of that has already been done. So if you're trying to onboard a new tool, obviously that comes with a lot of those things, so procurement, security issues, whatnot.

And that can eat up with the benefit of trying to do something like that because the goal is to give this to the people that need it, you know, as soon as we can to get it into as many hands as we can in a way. So how can we help it to get adopted faster? Right? This point is super important.

I was talking about that a little bit. That's all the curation. So simple and well structured knowledge source. I think definitely when I started out with this, one of the biggest challenges was that I just tried, okay, how about if we just feed it all these random interviews that we've done on slightly tangential topics? Or how about if we tried feeding it, I don't know, some NPS verbatims, more data you feed it, you feel like, okay, maybe it's going to provide deeper, better answers.

But actually the curation is really, really important. And so that's one of the biggest challenges for me in a way. It's like how do you cut down the data, the right data so that the model can give you the best possible answers. Now the challenge there is also that if you feed it too much data, you know, it's like what I've seen happen, it just cuts down to a slice of it.

It doesn't look at the whole thing. So the pattern recognition kind of, you know, doesn't really happen. So, yeah, curation is super important. Clear explicit instructions with fallback. You know, we talked about it a little bit. Yeah. Don't try to get it to do too many things at once. We do live in the agentic era, so yes, if if you do want to have more, individual jobs, I guess having separate agents for that is probably a better idea.

So try to be quite specific, give it one job, you know, and then also give it that permission to say I don't know, right? I think that's worked really well in this case for me. So yeah, explicit instructions and a fallback. Now the most important thing of all, obviously we are at an AI conference, do not stop talking to customers.

Yeah, that's really I can't emphasize that enough, right, because the AI can't really tell you what we haven't asked someone yet, right, it's only relying on the data we already have. It can't replace researcher judgment if we're trying to develop something new. It can only reuse insights, right, it can't really generate anything new. So, you know, if your researchers start getting fewer slots, then you're probably going about it the wrong way, I'd say.

Right? It's really important we keep that real connection alive. And there is a real risk there, right, because you know once you have something like this available to people then, you know, it's very tempting to say, oh, we already have that, why why would we need to talk to customers, right? Like it looks like a customer, That's it.

So yeah, think that's again something that's I'm sorry, I'm still trying to really figure out, right, what the impact of this will be and how do we navigate that question. But yeah, I think it's a very important one to think about. Because the whole point of personas, right, was to keep teams closer to the customers, right, to keep teams customer centric.

And so we don't want to create another layer between our teams and our customers. Right? There already usually are a number of layers. So if we can kind of bring them closer, that's the goal. We need to really remember that what are we after. So yeah, honest framing is that we want to extend access to research rather than replace it.

So we do not want to replace research, researchers, or just sitting across from customers. If you've ever had a senior stakeholder sit in a customer interview, I think that's always so powerful, right? So I really don't want to lose that moment where they really kind of internalize that experience. So, yeah, build it alongside, not not instead of research.

Now I wanna end kind of where I started. Right? So bringing it back to, you know, that good design starts with the customer. I think I would say good AI gets you closer to your customers. I think that's the test that's probably applicable almost to any AI you may be developing. So think about it maybe tomorrow if you're working on something.

I think this is the test, right, I wanna leave you with. And so if it really doesn't do that, then it's kind of a distraction. It can be impressive. You know, you can have a great demo. You can have a clever model. You can, you know, have vendors trying to sell you all sorts of AI stuff. But if it doesn't do that, it's kind of not doing the right job.

Now we've had static personas in the past, you know, we use personas still today, trying to help us focus on the customer, trying to get that product owner and the designers talk about the customer. We know it can't hold all the details sometimes, so it's not perfect as a tool. I think the persona agents, maybe they're a better tool, right?

They can definitely be a little bit more adaptive, they can have a conversation. So it's not a handover, it's more of an interaction with the research. They are grounded in all the richness of the data we have, right? We have so much data in the organization. And it can surface all that detail, right? And it can put someone who's never maybe talked to a customer in person really in that room with them.

But it also has the potential to kind of quietly take the customer out of the room in a way when maybe sometimes it replaces the research, right? It shouldn't be doing that. So we need to be careful about it. Now to close off, I do believe that AI can really emphasize that human in human centered design.

So let's try and humanize things with AI wherever we can. And good AI should get us closer to customers. Now that's a test, but it's up to us to make sure that it does. My name is Boris. Thank you for listening.

Thank you, Boris. I have a question before I ask anyone else if they have, but that idea of sort of adding new research, adding new insights into the personas and and making them available to the agents, have you settled yet into a cycle of how frequently you're doing that? Have you tried to update the personas yet with new research?

I've tried to do that as a kind of proof of concept. I'm still struggling to make it work in a way that would be seamless. So we don't have like an operationalized product in that sense yet. Yeah. I think that's it's a pretty challenging thing to tackle.

I think like that's a challenging thing with personas, full stop. And that's why I was curious whether or you've made that attempt here. I'm working on a project at the moment where the personas were created by somebody else and and made available to our team, but they were developed five years ago.

They

haven't been updated in that time.

Exactly. And I

think that's probably too long. Even in that particular context, I think it's too long. But, yeah, I was curious. Does anyone have a question? Yes. Danny, I'll give you actually, I'll just give you this one.

Thanks for that. It was really good. It sounds like you're covering your bases a lot with, you know, the the fallback rule, and, you know, you showed the example of it having that traceability. I'm just wondering in terms of, you know, soak testing it, like, how have you noticed it drifting at all?

Or, you know, kind of finding you know, if things can go wrong, they will go wrong. Right? So, like, if if there is a gap to be had, AI will usually plug that gap and will sometimes make it up. Have you found that it's reliable or too early days yet?

Yeah, I mean it changes all the time a little bit, right? I think the challenge is what kind of tools you're using again. I think currently pushing this through to a pilot in a way that we can then say okay, this is now good enough to scale up. So we're still learning a lot of those things. I would say even with tools like Copilot, because they do change the model behind it, you can then every now and then see that oh, actually today it's behaving a little bit different to how it used to, right?

So there's definitely that challenge. One thing that I also need to still properly research is does it behave differently to different people? I'm not entirely sure there. Like does it behave differently on my particular setup with my personal kind of context? So yeah, I don't have all the answers to that yet.

I've got more of a comment which is I think, you know, the purpose of personas is for us to be able to feel something for the people we're designing for. And probably a lot of us are a bit desensitized. I think I've met five or six Priyas in the last few years. Circle photo, a lot of lines around motivations, interests.

And what this is this is really interesting because I feel like what you've done is sort of break that the desensitization that we have to go, oh, I can feel again. Oh, that must be really annoying when you're waiting for that information. Oh, is that how long it takes? So I think there's something here. I've time back to Steve's point, think, you know, making sure that they're updated and that we can continue to actually have the real research there, but really helpful, especially for some of the younger designers who have come through COVID and maybe don't have that same contextual research background where you're out in the field talking to people.

Really helpful to get into the mind of the user. So thank you.

Yeah, thank you. Thanks for the comment. Yeah, I would second that that like, one of the things I like most about doing research is when the person answers something you're not expecting, right? So that's the element where, ah, okay. So yeah, totally.

Personas You Can Talk To

Turning Research Into Persona Agents

Boris Divjak, Strategic & Service Design Lead

3 June 2026

The ANZ logo is displayed in the bottom right corner.

Personas You Can Talk To

Turning Research Into Persona Agents

Boris Divjak, Strategic & Service Design Lead

3 June 2026

ANZ logo.

Good design starts with the customer.

A small blue abstract flower or star-like logo is in the bottom right corner.

Good design starts with the customer.

A small blue stylized four-petal flower icon is in the bottom right corner.

Good design starts with the customer.

A small blue logo featuring a stylized person is in the bottom right corner of the slide.

Good design starts with the customer.

A stylized blue icon of two interlocking shapes, resembling leaves or petals, is located in the bottom right corner of the slide.

Good design starts with the customer.

A small blue icon resembling a stylized human figure is in the bottom right corner of the slide.

Good design starts with the customer.

A small blue logo resembling a stylized person or plant is in the bottom right corner.

Good design starts with the customer.

A small blue icon resembling a stylized plant sprout or human figure is in the bottom right corner of the slide.

Good design starts with the customer.

A small blue icon depicting a stylized person is in the bottom right corner.

Good design starts with the customer.

A small, stylized blue icon, possibly representing a person or a leaf/flower motif, is in the bottom right corner.

Good design starts with the customer.

An abstract blue icon resembling a person or leaf is in the bottom right corner.

Good design starts with the customer.

Designing with personas in 2014, personal archive

On one of the documents pinned to the wall, the text "BOOKS ↑" is visible.

A photo of a whiteboard displaying several cartoon-style persona cutouts. Above and around the personas are various printed documents and yellow sticky notes, suggesting a design workflow. A person is shown from behind, seated at a desk and working on a laptop in front of the whiteboard.