Agentic Experience Design at Singularity — Fireside Chat with Loni Stark & Sandy Diao

Julia Nimchinski:
Welcome to the show, Cindy Dow, leading growth expert from Meta, Pinterest, and Descript, among many others. And welcome, Lonnie Stark, VP of Strategy and Product at Adobe. Stoked to have you here. How are you doing? Welcome to the show.
Sandy D:
Thanks so much, this is gonna be a fun one.
Loni Stark:
Yeah, thank you. All right, Julia, and great to meet you, Santi.
Sandy D:
Great to meet you.
Julia Nimchinski:
One question, Cindy, and the stage is yours. We’re asking this every session, what is your top GTM and AI prediction for 2026?
Sandy D:
Oh, is that a question for me, or for, for Liz?
Julia Nimchinski:
Both of you. Both of you, yeah.
Sandy D:
Top GTM prediction. My GTM prediction has to do with talent, and that’s that teams are going to become much more generalized in their skill sets, meaning that we’re all going to be forced to move out of the comfort zone of our own specialization. So if we’re, you know, salespeople, if we’re marketers, if we’re content folks, we’re just going to force ourselves to stretch across various disciplines, whether it’s coding and data, or someone who’s not in content will have to think about content, and that’s because the the powers of AI allow for us to do all of that and more.
Loni Stark:
And my top question ties to what Sandy noted, which is, I think with all the AI tools, we’re just beginning to explore the possibility, and marketing is all about differentiation. So I think the bar is gonna raise on personalized experiences and what consumers will expect from brands, and so learning to work with AI and come up with new forms of creative ways to reach your audience is gonna be… it’s gonna happen in 2026.
Julia Nimchinski:
Fingers crossed. Cindy, back to you. The stage is yours. Take it away.
Sandy D:
Wonderful! Well, that was a really great, warm, warm intro into all of this here. So, for folks who are listening in, welcome to our session. This is going to be all about agentic experience design. Just as a brief intro, as Julia mentioned, I’m Sandy Diao, former growth executive, leading teams at companies like Meta, Pinterest, Descript, and these days, I run my own growth advisory practice. And for the next 30 minutes, we’re going to explore one of the most important shifts happening right now in customer experience design. that’s the usage of AI agents. Just as a bit of a primer before we dive into our conversation here, you’ve probably all encountered some kind of agent in the wild in your own experience as a customer already, whether it’s an agent phone caller, or chat support, or a personalized landing page that was autonomously personalized and generated just for you, or even a custom onboarding flow that’s been guided by some kind of voice agent. So. We’re clearly entering this world where AI agents are becoming far more commonplace, and it’s making a big impact on our experience as customers. For organizations, though, it can really feel hard to keep up. I know I feel this way. Whenever I pull up in LinkedIn, or I pull up in the news, it feels like every other day a new model provider is offering some kind of new agent platform, or there is a new agent platform that has cropped up, from GPT to Manis, and even individual standalone tools like Intercom Spin for chat support, or Braze AI for autonomous email generation, and agents for pretty much every customer experience. you can imagine. So, actually, according to CB Insights, as of recent, there are over 250 startups now that are building agent infrastructure or standalone applications. And when it comes to adoption, a Gartner report shows that 70% of enterprise leaders want to deploy some kind of agent across customer experience and marketing this year. So, there’s a lot of interest and a lot that’s happening right now. So, in today’s session, we’re fortunate to have Loni Stark join us. Loni is Adobe’s VP of Strategy and Product, and she’s led Adobe’s experience platforms through multiple waves for products like Adobe Commerce. and concierge, and now LLM Optimizer, just to name a few of those. I’m so excited that we get to learn from her today, so without further ado, Loni, I’d love to dive in. I want to actually just start with a macro picture. You’ve led Adobe’s experience platforms through multiple platform shifts. I’d love to understand how you think about this transition to AI agents, and what makes it different.
Loni Stark:
I think if we were to look at the other platform shifts, we’ve been through an era of mobile, cloud, personalization. I think one of the key differences is that this is an era where the technology around AI allows for agents to do things a lot more autonomously. And also, to interact with, customers brands who have consumers or business buyers through conversational. And so, when you think about, the transformation that needs to happen in a company, as you were saying, it’s quite complex, and people are trying to figure it out. It’s not just about the technology, it’s about how do people work in relation to agents. And how do you design experiences for your customers that are personalized and use experiences… use conversational AI in a trusted manner? So, I think that requires a major shift in how companies think about this transformation compared to in the past, where I would say it was more of a technology play.
Sandy D:
Got it, and just as a follow-up there, from what you’ve seen, do you feel like the energy around the adoption is similar to some of these other platform shifts to mobile, to cloud, and some of these others that you mentioned? Or is there more of a reluctance, or just how does that spirit compare, essentially?
Loni Stark:
Yeah, I would break it down into a couple areas. One is from consumer adoption, like you and I. We’re in ChatGPT, there’s 800 million users there, we’ve seen all the other, and that’s on top of all the other LLM platforms. So there, compared to when mobile came into play, or when cloud, the actual consumer has adopted much faster than any other technology out there, which has caused pressure, I think, for enterprises and companies to figure out, how do I meet that demand? So… on that front, I think that companies are looking at it in two fronts. One has been, how do I, adjust and transform my customer experiences, so I’ve seen an increase in the need for customer-facing interactions like mobile phones, or mobile apps or websites to have a conversational experience. So that’s why we really innovated around Adobe Brand Concierge, which is about how do you launch trusted consumer and business buyer experiences, taking information from your website, product catalogs, for doing that. So really reimagining that part of it. The other part is around internally. How do marketers use these AI agents and tools? And I think there, there’s been various studies where there’s been discussions around prototype, is happening, but how do you bring it to production? And I think there, the reluctance is around trust, and also, a lot of these technologies ask you to Change the way you work. And for most, marketers, they’re busy in their jobs, they have their, processes and flows, and so… the way I think, technology needs to be adopted there is really… trying to minimize the change management and change behavior, build up trust incrementally, and then look at transformative applications that maybe change the way that people work. So introducing AI agents in a way that, Really fit into the ways that people are already working is one way to drive scale of adoption in an organization.
Sandy D:
That’s really helpful, and a really great perspective. I want to dig a little bit deeper into this topic of trust, actually. So you mentioned that, you know, the people internally who are using this technology need to sort of be able to trust it, and there’s also this layer of trust between a customer coming to your website, or interacting with your brand, and needing to trust that the sort of technology, or the AI or the agent is offering, you know, trustworthy experiences For both types of trust, actually, and maybe specifically more so on the first… on the former. What do you see as the ideal model of interaction between humans and agents? You know, in your view, in your experience, you know, what have you seen to, you know, really work? And do you have any suggestions to people who are starting to adopt agents on how we should ideally set these up?
Loni Stark:
Yeah, I think a lot of the adoption models or the way that people think about it is more of a technology adoption curve. So it’s like, hey, someone adopts a few sort of capabilities that the agent can do, and then it kind of levels up to more autonomous, sort of, capabilities. I think that’s one way to look at it, but I actually think a more important way to look at it, if you think about that the one key barrier to adoption is change management, and trust is a big factor in that. So, designing it so that these AI agents fit seamlessly into someone’s existing workflow. And thinking about the risk and reward factor for a marketer or for a user. a human user here, to actually call on that agent, and if it turns out something that is of poor quality, what happens? If it turns out something that’s of good quality, what are the… what’s the reward there? So, an example of that is if you are looking to, for example, and a pattern that we’ve used in, experienced production agent for experienced manager is if you are looking to update a, a page or its content, right, and you want this agent to do it, as opposed to having it clog up the Your inbox of tasks to do. instead of just giving it to the person and saying, okay, now you’re going to somehow assign it magically to this AI agent that you may not trust, and if they screw up, it’s all on you, because they’ll say, well, you’re the one that called on this AI agent. Instead, be able to have, the AI agent do some invisible pre-work ahead of time, so that when you get the task, when you as a person get the task, you see the task, and you can choose to either start from ground zero and do the updates, or the work or the task that needs to be done, or start from where the Agent has already done the pre-work, right? Has already done the suggested updates. And that is a way of having low-risk, convenience. no real change in the way that a person works on it. And when you do a couple of these, and you see that the agent is turning out quality work. That allows you to maybe do a couple of edits or things, but speeds up, and removes things from your inbox, right? then that builds up trust to then perhaps say, okay, how do I start to now assign certain things to agents directly, and then maybe I do bulk approves or things like that. So that model of trust. going from low to then, at some point, looking at more transformative things, meaning now that we have these agents, what new experiences could I create? What are things that otherwise I wouldn’t have been able to do before. And, allow for that… those individuals, people to, trust the agents and the technology to get to that point. But the other part of this that’s, I think, really important is the agents are learning from the context of the company, of, things that are changing as well, because there’s a lot of, I think. training on AI, but every company is different. Their brand voice, their processes, and so that… there’s no way to, to really… get to that quality without some of those iterations as well. So that’s where, both the agents are getting smarter, but people are trusting it more through no need to change, get some benefit, to then, okay, because I see the value, I’m willing to now change the way I work. you get to a better human-AI collaboration.

Sandy D:
That makes a lot of sense. In addition to the trust, I think the trust actually is also bundled in with this idea of, you know, identifying and really having your finger on the pulse around what agents are really good at. You know, for example, some of the retail-related use cases that have been released, we find that… they find that with some of the retail agent chat support, for instance, agents are really good at certain types of, you know. Scenarios, like a lot of the transactional scenarios, billing, payments, they can… deal with refunds, you know, 10x faster and reduce the amount of time and increase the satisfaction dramatically. But then there are more highly complex conversations, like more complex disputes or situations that require a little bit more empathy, that can be a little trickier for agents to tackle sometimes. So, kind of your suggestion of having this iterative approach and continuing to build that data set in that context, so that these agents can continue to outperform even in some of these other use cases and scenarios over time. And actually, sort of speaking of the retail use case, and I’d love to understand, you know, what are some of the agent-led customer experiences that you’ve seen, and what are some of the ones that you’re personally excited about?
Loni Stark:
Yeah, so I was actually about to, when you mentioned the retail case, Sandy, I was thinking, we’ve seen that with Adobe Brand Concierge. Same idea. So, really, the focus area for Brand Concierge, which is a tool for companies to be able to more easily design these, consumer-facing experiences, is the area around product discovery. And recommendation is usually the first place where companies start, because there’s a lot of content on the website, there’s product catalogs, there’s all of these, parts that, can be pulled together, and getting that out to consumers. the conversational analytics that come back. You learn a lot, so there’s a… when we work with customers. there’s quite a number of assumptions or thoughts or ideas of what do their customers want to actually do in these conversations. Like, they know there’s a pain. They know there’s a pain, like, when there is a high consideration product. I don’t want to look through all these search filters, or I don’t want to understand your whole website hierarchy to get to the information, so it wouldn’t be nice If I could just have a, you know, conversation. So the pain is understood, but then what kind of specific questions are people asking? What information and content do I have to actually serve for conversation? Because having a conversation is very different than a mobile app experience, or website experience, and the kind of information that you have. So… Being able to, like you said, launch something where it provides, great information in a conversational way, and then, as we observe other, behaviors in terms of what people are asking, that kind of conversation analytics can then drive the next phase and rollout of the brand’s conversational experience for customers. So just an example, Brand Concierge, Adobe likes to use our own products, so it is live on adobe.com. And it… the first use case we went out there was, there’s a lot of creative apps, customers. have trouble sometimes figuring out what do I want to use? And so, we wanted this, conversational, app to really be a creative source, to say, hey, if you want to build a poster for your son’s birthday, here are the sort of things you want to use, and here are some tutorials, and things like that, right? And done it in a really conversational way. What was really interesting was when we got out there. Some of the first conversational that live came in was people coming to the, the chat box and wanting to generate images and generate actual artifacts. Right? So that’s an example where, rolling it out with a customer, you might find, with a retailer, they’re asking for certain, comparisons, or certain, information, like, is the cotton organic, right? And that might have not been something you had thought about putting it on your website, or even that was a part of your product mix, and that kind of information is important. So there’s an opportunity to deliver a greater conversational experience, but also immense learning, too, from these new experiences with agents.
Sandy D:
That’s a really great use case, and kind of you brought to life one of these areas in which agents are already making a big impact and have opportunity to further improve. In that scenario, with your experience with either brand concierge, or more broadly speaking as well. how does an organization know if that conversational agent design has worked, right? We’d love to understand what are some of the metrics or signals that teams are looking at in order to know whether or not agents are worth deploying, and whether they’re actually successfully impacting customers.
Loni Stark:
Yeah, that’s a really great question. So, when we work with, customers, what we see are clients, what we see are… there’s usually, like, the demand team, and the demand team wants to measure the agent on, are they booking the meetings? are… is it leading to a conversion funnel increase, right? Like, if I’m using this concierge for advising on vacations or things like that, is it leading to a higher, you know, purchase rate or larger basket size, things like that? And then we have the, a group in the company, usually, who’s interested in that engagement. And it’s like, well, there’s… it’s almost like the brand, right? It’s, not just about the funnel, but, can I actually get them to engage and maybe build a brand love, loyalty, that sort of thing? So, usually it’s, really interesting from a design perspective because you’re trying to balance both of those needs, and where it can sometimes become a friction point is, okay, if I’m having this conversation, is it being a distraction to someone? you know, going and purchasing something. So, some of the metrics, that and whether it’s a good design, I think is, one, benchmarking on your existing you know, site conversions or your content engagements to start off with. So, while this is a new technology with agents, it’s a shiny new toy unless it can drive, these sort of metrics, right? And they are still the same sort of customer experience metrics. That you’re looking at. And so it’s, one, looking at thumbs ups, thumbs down, so that’s an important thing, to put a simple way to give feedback, right? And that’s in a lot of different conversational sort of agents. A second is, the types of questions they’re asking, and whether they lead to an answer that is useful, delightful, that’s accurate. So that’s another piece of it, where even before it gets launched, there’s annotations, there’s testing, evals, but then when it gets launched, that’s something that needs to be monitored as well. And then, third is the… is the conversion. So, usually there is another step, right? And that other step could be, I want more content from the cus… from the… from the brand. Or now, because I’ve gone through this process. I understand what vacation package makes the most sense for me, and I’m not calling in to have to go through that whole process. I feel the confidence in being able to… to book. So those are some of the key metrics, and then I would say qualitative metrics are important, too, so it’s not just quantitative, but getting, information on what are the things that your customers are asking about. And one thing that has changed is, because of our conversations with ChatGPT, etc. There is a bit of a shift in… we’re seeing in consumer questions that we’re becoming a lot more specific in terms of the kind of questions we ask, versus when we were trained in just the keyword era, we’re becoming more demanding in terms of how personalized Could the responses be to a question?
Sandy D:
That’s actually really helpful. I love that cascade of metrics that you just described there, the benchmarking, the model evals, and then, additionally the conversions. I think that framing’s really important, which is that even though we’re using agents and this is a new technology and experience, it doesn’t necessarily mean that we have to craft a whole new set of metrics, because at the end of the day, we’re still trying to use these tools in service of growing our own North Star metrics and the things that we care about growing in the business. So I think that’s a really important point. Something else that is important in the topic of data here is this idea of how much customer data and context that we have, and how we should appropriately use it. In the example that you mentioned, where maybe a brand wants to be able to offer personalized recommendations, you can totally imagine a multi-agent orchestration workflow where one of the agents is actually going out and enriching the traffic lead in real time, right? So let’s say I visit a website. And the website knows that I’ve worked at Meta, or that I’ve worked at these companies, and that my title has marketing in it, and all of a sudden recommends me a marketing.
Loni Stark:
Yes, keep going.
Sandy D:
Yeah, exactly, and that’s kind of creepy, to be honest with you, right? Because it’s my first time interacting with the brand. At the same time, the output is truly what I need, right? So, it’s this really interesting balance that brands have to make. Around understanding how much customer context and data to use. And here, I’d love to understand, in this world of basically infinite personalization, or increasingly infinite personalization, driven by that customer data and context, how do we… how do you think about preserving that privacy and that identity integrity?
Loni Stark:
Yeah, I think there’s a few comments here, and it was interesting as you were bringing it up. One is, similar to all the tools we already have, you know, third-party cookies, first-party data, etc. Just because we have an agent, doesn’t change, kind of, the respect of privacy. It’s really interesting, because I remember, I’m old enough to remember this when, we’re… you know, when digital first came out and, you know, people got online personas, it took a while before people, understood or got to a point where they had certain etiquette for online interactions. It almost felt like people were, like, I don’t know if you’ve had this phenomenon, but when digital first came out, it was like, there were people that were the super nicest people in person, but online, they were, like, totally different personas. So, so I think that, just because we have these agents does not change, kind of, the privacy dynamics of respect. And so, the way we look at it is, it is about, unauthenticated and authenticated, users, and being able to make sure if someone’s unauthenticated, then you are privacy first, and you are looking at, potentially, their website click behavior, but just because there’s a way to triage Some other, you know. way through, sort of, third-party cookies or things of who they are. As a brand, I… it’s… I generally think it’s not a good way to act on it, not just whether it’s possible or not from a legal perspective, but more from a trust and relationship perspective with your… with your customers. So we look at personalization for agents after they’ve authenticated or they logged in. And be able to, personalize based on that. And the main thing that we see with brands that want to do that is around their loyalty programs. So, as more consumers are going to ChatGPT, etc, one of the main concerns that brands have is, with my existing loyal customers, am I going to get disintermediated, right? Is someone going to come in directly? And so, brands are really thinking about, hey, what are the, you know, what are the reasons why a consumer, my customer will want to come directly to me? And a big part of that is connecting. How do I offer a premium experience? How do I offer, you know, premium content. Experiences, to… customers who are loyal, and how do I tap into, understanding where they are in my loyalty programs? So connecting those two has been a large sort of value. I think the other thing that is going to be important Is, with all the content being generated out there, brands will need to face how do they like. make it clear what content came from actually their brand, right? What’s legit, from a brand? Like, so if you are a clothing company, or you are a retailer, how do they know that this is something, you know, how do your customers know That the information or the content you’re putting out there is actually from you, and it’s not just generated by someone else. So that, area of trust is where, we’ve invested, and a bunch of folks in the community have invested in, content credentials, right, and being able to, to designate that, as, as well.

Sandy D:
I think that’s actually a really helpful playbook for people who are still kind of on the fence, who are kind of figuring out their own policies. I love that approach of authenticated versus non-authenticated, because an authenticated user is essentially opting in, right? You have their consent, essentially, and you can let them know what types of data and context you’re going to pull to create a better experience, whereas the new cold visitor, the first time, you know, user who’s coming to engage with you, they don’t necessarily have all of that context yet around, you know, how your brand processes that data. So I love that approach, and honestly, it’s one I would opt into as a consumer, so love that. And maybe to just kind of tie things up here, I’d love to actually touch on some of the other work that you do in addition to your product leadership at Adobe. I had a chance to take a quick look, but you’ve been up to some amazing work, including some of your previous self-identity research at Stanford, co-founding Stark Identity, a media platform that’s generated millions of views, to actually founding a fine arts studio, Atelier Stark. I’d love to know, Loni, just, how has your research and some of your creative successes influenced how you think about AI and agents more broadly?
Loni Stark:
Yeah, I think that one of the aspects, that’s so important is, to really, as, technology leaders, as creative leaders, to break down silos of knowledge. And one thing about AI that’s interesting is When I got into computer science way back, it was less human, right? It was like, here’s how you code it, and, you know, you gotta put into structured information, and I think with AI, there’s a possibility of collecting… connecting more directly with AI and how it works. It’s… Machines have become more human. Right? And so, I think that, my explorations into identity shifts applies both to how do people relate to AI as AI becomes able to do certain things that we, as humans, have thought were inherently ours, right? Including anything from coming up with a marketing brief as a first draft. To generating really, you know, pretty amazing. images. And so, this study of, of identity, I think, is important as we continue to ensure we assert, sort of, what is human and our intentions and our will. And I think the other part of it is, how we interact with AI. I think, a lot of time is spent on benchmarks of AI, where we try to play through tests and say, well, it hallucinates, or can it write symphony, or could it do all those? And I think we can learn a lot as well. If we think of it more as, I… being collaborators, as opposed to putting them through tests that we would never put a human through. I mean, I don’t think I could pass any of the tests that are out there, that we put AI through. So I think those are some of the things I think about in terms of how we make the collaboration between AI and humans work really well.
Sandy D:
I couldn’t have said it better myself, and that’s actually really inspiring. I’d love to continue to follow some of your both creative work as well as your product leadership at Adobe as well. So, thank you so much, Loni. That was really inspiring.
Loni Stark:
Thank you, Sandy, for the great questions.
Julia Nimchinski:
Thank you so much for the phenomenal fireside chat, Loni and Cindy. Just one last question. Loni, I know that you’re organizing your own conference, which is really, really special and amazing concept. Could you tell us more, a little bit about that?
Loni Stark:
Sure, it’s called the Third Mind Summit. And, it’s actually one where, I’m doing this experimentation, Where, 6 agents And two humans, myself and Clint. my thought partner, I would say, are going to host a summit where the AI is going to present as well, and it is going to be AIs that have, over the last 6 months. gotten a lot of context from working on projects together. So, they’re actually generating, presentations, and we’re also reflecting on how we provide the space for them to potentially have agency. And… It’s one of those things where it’s an experiment. We’ll see whether something new comes out of it, or it’s just a mirror of what we’re feeding into these agents.
Julia Nimchinski:
Love it. And what’s next for Adobe?
Loni Stark:
What’s next for Adobe? I think that, we, see that brand visibility is going to be important with AI agents, and how do companies continue to create great experiences, not only for humans, but now for machines as well.
Julia Nimchinski:
Thank you so much again, and Cindy, what’s the best way to support you? I know you’re writing quite a bit on Substack. What else?
Sandy D:
Yeah, writing, sandydow.substack.com, write a lot about growth and go-to-market, as well as AI adoption for companies. Also, that conference sounds super interesting, would love to be a part of that and watching as well. And generally speaking, if you have growth questions or, you know, want to chat about how agents and AI can help further growth goals, chat with me, because that’s also something I’m experimenting with.