Adobe Summit: Why Context Is the Missing Ingredient in Enterprise AI

[00:00:03] What happens when AI stops being a tool that simply responds to prompts and starts becoming something that can reason, decide and actually start taking action? Well live here at the Adobe Summit that question is everywhere. And over the last few years we've all become familiar with generative AI and asking questions, generating content, speeding up workflows. But this year the conversation has clearly shifted. Now it's all about agentic AI.

[00:00:33] And for many business leaders that phrase might feel a little abstract, maybe even overwhelming. So today I'm joined by Daniel Schleinberg from Adobe where he's focused on customer experience orchestration and helping shape the company's AI and agentic strategy across its portfolio. And he's someone that sits right at the center of one of the biggest conversations happening at the Adobe Summit this year.

[00:00:59] And that is how businesses move from AI that assists to AI that actively drives outcomes. So in this conversation today we'll break down what agentic AI actually means in plain English. What the real difference is between AI that generates outputs from AI that can interact, reason and take action across multiple workflows.

[00:01:21] And also understand how this changes the customer experience, internal operations and ultimately the way that organizations think about growth. We'll have some real world examples and for tech leaders listening we'll also get into the architecture behind it all. Data, permissions, governance, trust and what needs to happen behind the scenes for agentic AI to work safely at scale.

[00:01:45] So if you've been hearing about agentic AI everywhere but while still wondering what does this actually mean for your business, this episode hopefully will help connect the dots. But enough from me. Let's get Daniel onto the podcast now. So thank you for joining me here at the Adobe Summit, Daniel. Can you tell everyone listening a little about who you are and what you do? Daniel Scheinberg Yeah, absolutely. It's great to be here. Daniel Scheinberg So yes, my name is Daniel Scheinberg.

[00:02:11] Daniel Scheinberg I am part of the product team here at Adobe focused on our customer experience orchestration business. My team is really spending most of our time on our cross product portfolio efforts. And so as you'd imagine one of the most important things that we're all focused on and how we're how we're leveraging it across the portfolio is our AI and agentic strategy. And so that's where I'm spending a lot of my time these days.

[00:02:36] Daniel Scheinberg And from all the conversations you're having, all the keynotes you've seen and indeed yourself being on stage, what are the key things that excite you? Daniel Scheinberg What's everyone talking about this year? Daniel Scheinberg Yeah, I think there's a there's a lot of excitement around a couple of things. Daniel Scheinberg One, when we think about where everybody was a year ago coming to this conference, and it was very early on in how people were thinking about agents and what would they mean? Daniel Scheinberg And what could they what could they possibly do in the enterprise? Daniel Scheinberg And you fast forward a year and the technology has advanced so much, but not only has the technology advanced,

[00:03:05] what people are doing in their daily lives, all of us as consumers, you know, we're sort of expecting as these experiences that's changed a lot. Daniel Scheinberg And then also what we're starting to see across enterprises and across enterprise adoption of these technologies has changed. Daniel Scheinberg So that's kind of one big piece of it is the just general environment that we're all living in now and how how much AI and agenda capabilities are starting to become a big part of that.

[00:03:31] Daniel Scheinberg The second part of it that's been really exciting for me is what we're coming across as Adobe in our point of view. Daniel Scheinberg And, you know, it's a very fast moving space. Daniel Scheinberg And so we feel like, you know, even just in the last six months, we've been making big strides. Daniel Scheinberg And what we're here talking about with everybody and the vision that we're putting forward, we think it's really compelling, very exciting, builds on everything we've been doing before. But like, as I said, you know, even six months ago, it wouldn't have been what we were talking about now.

[00:04:01] So I think that's really exciting as well. Daniel Scheinberg It really is. Agenti KI is that big buzzword. Daniel Scheinberg I think many business leaders still find it a little bit overwhelming, a bit intimidating. So I'd love to take everyone on a journey and maybe demystify it with you today. I think, as you mentioned, in our own habits, things have changed where we maybe don't no longer Google as much and maybe go to a digital assistant to find information. And we have spent a lot of time over the last few years talking about generative AI, but now we're hearing more about agentic AI.

[00:04:28] So from your perspective, what's the real difference between AI that just generates outputs and AI that actually takes action? Because that is the shift, isn't it? Daniel Scheinberg Yeah, that's great. Great way to think about it. And, you know, when we were first really trying to help our customers understand this, we laid out, you know, what do we think of as an agent? What are the key components of an agent? And we kind of boiled it down to there are really three elements of this.

[00:04:53] An agent can interact, it can reason, and it can take action. And what's interesting is like, we've had agents for a long time. You know, if you think about the classic kind of. A customer service phone tree agent that we're all, unfortunately, a little too used to. Things are changing fast, so we're getting away from it. But, you know, you can all remember you call in. There's a menu of click one, press one for this, press two for this. And, you know, these are your options.

[00:05:22] You pick one, then it takes you down, you know, a deterministic path. And at some point you get to the end of that path. Either it can solve your problem or it can't solve your problem. Usually it can't. And you start clicking zero and saying, give me an operator. I want to talk to a person. But that idea of an agent is it's the same concept, right? Like in that pattern, it's interact. It's tell me which one of these. I'll tell you the menu of options. You tell me which one of these things you want. Then based on that, I'm going to take. I'm going to take you down a path in this deterministic tree. So that's kind of the reasoning layer.

[00:05:51] And then the take action is, oh, you know, I'll tell you what your bank balance is right now. So that that idea of an agent has been around for quite a while and we've all used them. What's changed really was generative AI was the unlock for what could agents do and how basically how smart they could be and how much better they could be. So and it works really across all three of those layers. So the first part is you think about that interaction layer.

[00:06:17] What generative AI is really good at, right, even from the beginning was if you gave it a prompt, if you ask it a question, it's really good at understanding, oh, this is what you want. Right. So it's not pick from this list of things that I understand. It's tell me what you want and I will understand you. So that interaction layer became very different with the advent of generative AI and that ability to understand your intent from that interaction, even if it's, you know, having to ask a clarifying question or kind of work down that path.

[00:06:47] But then that next layer of reasoning and how does the agent reason. So before I kind of described the phone tree, the deterministic path. Now the these LLMs, these generative models, they're able to come up with the approach to take to your problem and figure it out on the fly rather than it all having to, you know, someone have to plan ahead of time. These are all the possible things that can happen. So the reasoning layer got a lot smarter.

[00:07:14] And then the action, the taking action. This is something that you also hear people talking a lot about is these agents now can use tools. And so, you know, in the software world, a lot of that's through APIs or application programming interfaces, which is basically just a way of telling an underlying system, this is what I need from you and getting a response. And so these agents can now do that across a wide variety of tools in the software ecosystem.

[00:07:43] And so they can make all kinds of things happen. So you put those pieces together and an agentic agent is really incredibly, incredibly capable. They can end up becoming more sophisticated all the time and just a huge leap forward in what agents could have done in the past. And it's all driven by the generative AI advances. And of course, every tech project now is under close scrutiny for what value does it offer? What's the ROI?

[00:08:08] So if AI systems can now reason, decide and execute right across workflows, how does this fundamentally change the way that enterprises maybe even create value compared to the last wave of AI? It's a great question because there's a lot of ways that I think this changes the way work gets done, the way businesses think.

[00:08:26] One of the most straightforward is, and we've done a lot of this within our software, within our applications, is helping practitioners be able to scale up what they do. We talk a lot about driving personalization at scale, which is the value that a business can get by ensuring that every interaction that they're having with an individual or a business, if they're B2B enterprise, that every interaction is the right interaction.

[00:08:55] You know, that you're bringing the right information to bear, you're sending the right message, you're understanding the intent and responding to that. And historically, that's something that it takes a lot of work, a lot of coordination across people to get it done. And you can't always do it to the level of granularity that you want to. So just as an example, you know, if you think about one of the things our customers are doing is managing customer journeys. And we have obviously have software for that to build out these journeys.

[00:09:25] Think about which audiences you have and how do you orchestrate journeys? What's the right journey for them? And at human scale, you might be able to run, you know, 50 journeys, let's say, which is pretty good granularity. When you think about like, wow, you know, think about my customer base. You know, what are the audience segments I have and sort of the intents they have. And so that's a pretty good level of scale.

[00:09:48] But when you start to think about, sure, but then I've got within my audience base, I've got people thinking all across my product portfolio. I've got people in all different markets. I've got people in all different languages. I've got people with all different aesthetic senses. Like, how could I get to that level of granularity?

[00:10:05] And when you want to scale it up even further, that's where these agenda capabilities really make a huge difference because they can take what your practitioners are doing and they can you can help them scale that up, you know, 10x, 100x from what they're doing right now. And I think that's the piece that there's a lot of value to be created from that. And it's a little easier to wrap your head around sometimes and a little more incremental in terms of the way, you know, the way we think about things.

[00:10:31] And then there's this sort of more transformative side of like, how do things just evolve completely differently? And that gets back to, I'd say right now, a lot of it is really art of the possible. It's not happening at broad scale. You see it in small, small pieces here and there. But but that's where you start to think about like, OK, great. So now we have these agents that we're bringing into our ecosystem and they're starting to and this is some of the things that have just happened over the last, like I said, three, six months.

[00:11:01] Where they're getting to the point where they're persistent and they have memory and they can understand underlying context. And so as they start to become more autonomous, potentially autonomous or be able to do more on their own, we're getting to a place where rather than saying, hey, this is what I want to accomplish or this is this is the task that I need to do. Help me scale it up. You can get to, well, here's the goal I'm trying to achieve and help me to achieve that goal.

[00:11:29] And so, you know, when it starts to be, look, we're trying to, you know, increase increase sales volume by this percent or something like that. And you have this agent that has access to all the information across your enterprise or the information you get access to and and an understanding of kind of the broader context that you're working within. Some of it could be external market forces.

[00:11:53] Some of it could be internal, you know, goals and and restrictions and constraints that are in place. But it starts to understand that ecosystem that your business is operating in. Now it can really start to, as we've been saying here, you know, act as more of a teammate and potentially come up with new ideas for innovation on its own and, you know, recommend them, recommend them to your team, recommend them to your business.

[00:12:21] And that's really where I think things will start to transform, because the idea of being able to kind of have always on monitoring that's not predefined. This is what you have to look for, but always on monitoring that understands my business and says, oh, wait, that's something we should be paying attention to that. And then not only is it capable of doing that, but then also saying that's something we should be paying attention to. And here are some thoughts on what we might do about it.

[00:12:48] That really starts to transform the way we think about business. It really does. And I was speaking with Virgin Atlantic yesterday, and they've got the very nice problem of every customer is completely different. They have frequent business travelers, families, people doing a once in a lifetime trip. And they work in with Adobe, which is working out brilliantly. And it's not just solving problems and answering questions. One feature that they almost didn't expect to take off was a trip planner. And you can have your goal, but it can take off in different areas as well, can't it? Yeah.

[00:13:16] And that's, I think, one of the things that we're learning now and will only sort of accelerate as we go forward. A lot of the agentic capabilities I was talking about before are really aimed at the people within your organization and helping them be more successful. The brand concierge, we use all the same technology. But this is AI and agentic technology that a brand can turn outward facing to their customers.

[00:13:41] And so the brand concierge is an interactive agent that a brand like Virgin Atlantic might put on their website, might put in their mobile app, you know, put it into third party environments like a WhatsApp or other areas. But that technology is basically an agent that's interacting with someone on behalf of your brand.

[00:14:04] And, you know, back to what we were talking about is kind of the advances in agentic capabilities, like with the intelligence and the generative AI technologies that you have now, you can have this agent that's trained on everything about your brand that has access to your content, has access to the data to understand this individual, you know, the profile data when somebody comes to talk to it. And to your point, like it knows, is this person on a trip already right now?

[00:14:31] Or is this person, you know, maybe they have a trip that's planned three weeks out. You know, are they typically in business traveler mode or typically in vacation traveler? So it has that understanding of the individual as well as builds up context through the conversation and through multiple conversations, builds out more understanding of that person and what they're trying to achieve. And so, you know, it really does create this richness of that interaction layer and really getting to understand what does this person want to accomplish?

[00:15:00] And then that reasoning layer, you know, what you can bring together is here's all the things that I can do within Virgin Atlantic. Here's all the things that I understand about what the company can do for you. But it also has an understanding, you know, because it's based on this LLM that's trained on the entire Internet.

[00:15:20] It also has an understanding of like when you say I'm going on a trip to Tokyo, there's all this information available about Tokyo, you know, as well as whatever Virgin Atlantic has pulled together specifically about traveling there. And so that can really lead to the richer conversation as well. To your point, you know, when you start to think about like how can we help with your trip planning, it can bring together a lot of the sort of content marketing materials that the brand may have assembled.

[00:15:47] But it can also take into account what's out there in the broader world of things that people might get excited about in that environment. And for the business leaders listening, I think we've covered that focus on the front end experience. But of course, the real work happens behind the scenes. So for the techies listening, what needs to be in place from an architecture, data and systems perspective for AgenteVe to function at scale? Yeah. And there's kind of the general answer to that question. There's the specific answer to that question in our case.

[00:16:17] And I think the general part of this is it really needs to be to work well for your enterprise, work well for your business. It needs to be grounded in the fundamentals of your business. It needs to understand what you're trying to accomplish. It needs to have access to the underlying data. It has to be able to understand that underlying data. And then it needs to have access to the tools that are required to execute the job.

[00:16:47] So if those underlying tools are within your CRM system or within your ERP system, wherever they live, you've got to have some way of giving it access to those tools. And increasingly, software providers like ourselves are making those tools available. What you hear about the most common is the MCP servers or the model context protocol server. That's a common protocol for agents to interact with these underlying APIs.

[00:17:14] So you kind of have to build out that ecosystem. And then you have to spend the time on training your agents, getting them to behave the way that you want them to. It's not none of this happens just magically. And I would imagine for tech teams listening as well, trust in autonomous systems, governance, the fact that context is everything. There's a lot of big topics that tech teams might be concerned with. How do you address those issues? Absolutely. Yeah. And so that's part of this whole kind of idea of the training is, right, you've got to give it the constraints.

[00:17:44] It's got to have an understanding. And then you've got to test it out and make sure that it's delivering on what you expect. You know, so that was kind of in the general sense. And when I said, let me take it back down to Adobe because that's what I really know best, obviously. And yeah, you know, in our space, what that means is we've built out our orchestration layer. We are building with what's known as a with what's known as a harness, which kind of pairs with an underlying LLM and really operates the agentic loop.

[00:18:13] Basically, that's the way that this thing, that it plans, it executes, it understands the feedback, it learns, and it improves the cycle over and over. So that's kind of at that top level of how does it come together. And then what it's doing underneath is it's connecting into all these underlying systems.

[00:18:32] And so we, when I talk about kind of being grounded in the enterprise and the enterprise's data, you know, for us, what that means is really understanding your customer profile data, really understanding your all the content repositories, the content you have available. And that's not just imagery, but it's also, you know, can be like thought paper, you know, white papers and thought leadership.

[00:18:59] It can be product catalogs, you know, all these pieces, all this understanding of what's, what makes up the brand. It's an understanding of your brand. So, you know, we introduced the brand intelligence, Adobe brand intelligence component this week. And that's just a really in-depth way for, for the AI to understand your brand, not just kind of like, here's the list of brand guidelines that, that we publish.

[00:19:28] But it's also an understanding of looking at all the content that, that is approved. And, you know, what does that mean about your brand? It's gaining, you know, the ability to actually generate on brand content, all those pieces. Like, so those become some of the tools as well that are accessible. But all that sort of context underneath. And then we also try to bring in some of these elements of like, what is, what are you trying to accomplish? What are the, you know, what are, what are the creative briefs in the system?

[00:19:58] What are the, what is the strategy that you've captured? All those pieces become important context for this agentic system. And then obviously we're building in the, the trust, the governance, the permissioning, right? So these, these agents, they are typically have permission, you know, can only access some elements of, of the underlying data and content based on the permissions often of the user.

[00:20:25] So whoever's asking something of the agent or telling the agent what to do, it, it inherits that person's, that person's permissioning and entitlements across the stack. And then, you know, we're building out our full skills catalog. So the skills are basically like the recipes or the instructions for how to go use tools to accomplish something. So we might have a skill, for example, typically it's a text file written in Markdown.

[00:20:52] We would have a skill, for example, on creating an audience. And it basically says, well, this is what you need to create an audience. These are the, these are the API calls that you need to make the tools you'd need to use. And that agentic harness, then on the fly, you know, and it's having that conversation with somebody and it's like, oh, we want to create this journey and it's going to need these audiences. It's pulling in, it's reading the relevant skills, finding and reading the relevant skills, executing against those underlying tools in our system to make it happen.

[00:21:22] One of the, I feel like I've taken you on a long tour here, but one of the other things that's really essential to all of this is we live in, and this comes up a lot. We live in a world of unstructured data and, and everybody is looking at, well, how can we make sense out of this unstructured data? And one of the things that we do at Adobe, we have the Adobe experience platform, which really helps to apply structure to all this data.

[00:21:50] And what that does for the agentic systems is one, we can use them to get the data into this form, the data and the content into this form. So it's, you know, well understood, well maintained. Like you don't have, I'm sure many people have experienced this where, you know, you kind of go and use generative AI to get an answer about something. And it's kind of looking across all the information available and it doesn't know which information is current, which information is most relevant, right?

[00:22:16] Like when you actually dig into the sources, you're like, well, these three sources were great, but I don't know why you gave me information from that source. So that's where it's important to have this layer of bringing some structure to the underlying data and content, because then you actually know, like, here's, this is the most relevant. This is the most timely. This is what I should be relying on as current rather than having, asking the system to try to figure out what's important.

[00:22:41] And then once you have it in that structured way, then it's much more straightforward for the agents to go and access it and know, you know, where to go, what to look for, how to use the tools available to it. Well, thank you so much for sitting down with me today. I'll include links for everything to help anybody listening and any organization interested in safely and effectively deploying agentic AI at scale. But just thank you for stopping by today. Absolutely. This is great. Thank you.

[00:23:08] So a big thank you to Daniel for helping me turn one of the biggest buzzwords in tech right now into something far more practical and understandable. Because agentic AI is not simply about smarter chatbots or faster content generation. We're now talking about systems that can understand intent, reason through problems and take meaningful action across the business.

[00:23:31] And that changes everything from that customer experience to marketing, to operations, to product strategy and even decision making itself. But what stood out to me most here is we're not talking about some distant future conversation. It's already happening. And Daniel made it clear that none of this works, though, without some strong foundations. Data architecture, governance, permissions, trust and context. All these things matter more than ever.

[00:24:00] An AI without structure quickly becomes just noise. An AI without the right systems behind it becomes something much more powerful, though. So for anyone listening who wants to learn more about how Adobe is thinking about customer experience, orchestration and agentic AI, I'll add links to everything we discussed today so you can dig a little bit deeper. But as always, I'd love to hear your thoughts here. Are businesses ready for AI that acts as a teammate rather than just a tool? And where do you see the biggest opportunity?

[00:24:29] Customer experience, internal operations or maybe somewhere completely different? But let me know. As always, techtalksnetwork.com. I'll return again tomorrow with another conversation. But hopefully together we'll make sense of the technology that is shaping the future of business. But that is it for today. So thank you for listening. And I'll speak with you tomorrow. Bye for now. Bye for now. Bye.