Name: From Rules-Based Workflows to Autonomous Decision-Making
Uploaded: 2026-04-10T07:16:04.826Z
Duration: 1 h 12 s
Description: From Rules-Based Workflows to Autonomous Decision-Making

Transcript for "From Rules-Based Workflows to Autonomous Decision-Making": Hello, everyone. I'm Theo Sharp from Henry Stuart events, and I'm delighted to welcome you to today's webinar sponsored by Oprimo. During the session, please send through your questions at any time using the questions box, which you can open by clicking on the question mark at the top of your go to webinar window. Also, reminder that a recording of this session will be sent to you tomorrow so you can view it again at your own pace. And now it's my great pleasure to pass you over to Satarupa Chatterjee, senior director product marketing at Aprimo. Thank you. Thank you, Theo. And thanks to HSAM for organizing this webinar. Hello, everyone, and welcome to today's webinar on agentic DAM from rules based workforce to autonomous decision making. We are very excited to be here and very happy that you could join us. Today, I'm also thrilled to be joined by our Primo's head of product technology, Tarun Chawla. Tarun, welcome. Do you wanna take a couple of minutes to introduce yourself? Sure. Yeah. Thanks, Satrupa. So, like Satrupa said, I'm the head of product technology over at Aprimo, which means that my job is all about looking at emerging technology, especially things like AI and agentic technology, and figure out how we can put features in our product, to unlock value for you and guide our customers in understanding how to take advantage of these things to really solve problems as quickly and effectively as possible and especially with agents move forward towards more and more autonomous operations in, the content operation space. Awesome. Thanks, Darren. So like you just heard, Darren and I are gonna cover today how critical it is for DAM and content ops to adapt to this new world of AI powered content. We are calling this future the agent exam. AI, as you can imagine, is now being actively embedded in every part of the marketing ecosystem, including in dam platforms. So it is our goal in this session to share some insights on where the dam industry is heading, what an agentic dam should do, how agents need to be governed, and also the operational shifts needed to be successful. So with that, let's dive right in. If you can go to the next slide. Great. Okay. So, let me start by throwing a question out there for all of us to think of. Who are we creating content and digital experiences for right now? Is it still for humans exclusively? I think it's really important that we challenge this core assumption that many marketing teams are still making. A lot of us are still primarily creating content only for people, be it images or videos or text. And this is based on the assumption that humans would see the marketing message on a channel like Google Ads or email or social media, and then they will click through to the website, go through several product pages, and then make a purchase decision. Or in case of a b two b company, they will fill out a get in touch form. It's time, however, that we start challenging this reality. AI is fast becoming as important a target audience for our content as humans. And let me share a personal story here. My husband and I traveled to Singapore last year. We had some very specific requirements. We wanted a hotel with a sea view. We wanted adjacent rooms for some family members who are also joining us. So I gave AI our budget and all of these requirements, and we were able to find, the top three best options to choose from. And the interesting thing is I never visited the individual hotel websites till I was actually ready to make that purchase. So I think we can all go on and on about how AI is disrupting the buying decision in every industry, not just, you know, hotel bookings or groceries. Even people in b to b tech, buying decisions are consulting AI for shortlisting and comparing vendors. So here the reality is that content now serves two audiences. Humans do continue to consume it, but AI agents are increasingly discovering it. They are interpreting it, and they're acting on it on behalf of humans. And if you see the stats here, this shift is already underway. Accenture reports that 72% of consumers regularly use AI tools during purchase decisions. This means that AI is reading your content and paraphrasing your customer's buying decision. So you need to influence how your content is getting paraphrased by AI. Also, if you look at McKinsey, McKinsey estimates that agent ecommerce could drive 3 to $5,000,000,000,000 in global retail revenue. So if you're in industries like manufacturing or retail or travel and hospitality or any kind of b to c commerce, it's going to get majorly disrupted by agent ecommerce. And this is also true for b to b brands who are seeing significant rise in AI queries, that are relevant to them, which is to say that agents are going to be making trillions of dollars worth of purchasing decisions very soon. So industries have to make it easier for agents to understand their offerings, to make recommendations for humans, and even buy their products online. And thirdly, AI is not just a consumer of content, but also agents are now digital workers on our team. Gartner actually predicts that nearly 40% of enterprise software applications will embed AI agents by the end of this year. So brand and marketing teams have to think carefully about how these discrete agents will be orchestrated and content will be governed all the way to the final delivery. And let me give you an example here. Say, you're in the health care industry, and your marketing team uses a marketing automation platform for email campaigns. And the AI agent in the platform takes approved content from the DAM. But while creating the email, the AI disregards the disclaimer content that was inserted during the compliance review. And this can happen if the email agent is not fully trained on regulations and not in sync with compliance agents in your DAM system. This as you can imagine, it can be a big problem unless companies figure out how to tie their agents together across all of these processes. So summing up, what these three key trends are telling us is that content is no longer just being consumed or created for humans and by humans exclusively. It is constantly evaluated, ranked, selected, edited, and activated by machines. So if your content is not structured, governed, and intelligence ready at any point in the content ecosystem, it could affect growth and increase risk for your business. So our content ecosystem has to evolve to be prepared for this agentic future. So the next question you may have is, okay. Sure. You know, I get it. All technology needs to evolve to this new reality, but how are dam systems evolving? So let's talk about how the dam industry has evolved over time. So if you think about it, DAM systems started as basic file storage. They were primarily shared drives that acted as informal systems of record with very little governance. People would just use folder structures to keep content semi organized for later retrieval and use. We went from there to the next phase where we created this global single source of truth in the DAM system. We added critical features like version control, improved retrieval to solve some of these problems like discoverability. But even here, workflows were still very human driven. Someone had to, you know, click, click, click in order to complete one step in a 10 step process to upload a document properly. From here, the focus shifted downstream to how the content would be used. So dams then became distribution hubs supporting CMS systems, portals, omnichannel delivery, and also providing more brand control in a single location. And then over the last few years, we have moved into this AI powered dam phase. So vendors have introduced cool new AI features such as auto tagging, semantic search, and some level of workflow automation. But even here, the automation was a bit localized, for example, at the point of ingestion. And it did not necessarily cover the entire content ops life cycle. So the next evolution that's happening right now is agent exam. Here, it's not about individual AI features that involve a human interacting with AI or a chatbot. Here, agents become first class participants in content ops alongside the humans of the marketing team. So content is no longer just stored or distributed. It is acted upon autonomously by agents that are governed by humans. So now next, let's get deeper into how AI is changing the damn industry. If you go to the next slide. Yes. So first, I think we must accept this new role of AI agents. They can now ingest and enrich content automatically, not because a human searched for it, but because a workflow, or an AI model or a decision engine triggered it. They can also transform content. They can generate variance. They can determine where it goes next. So content is now a primary input into enterprise behavior. As a result, we need to treat AI agents as first class content consumers. So what does it mean when I say first class content consumer? Look at your damn today. Who gets to pick assets, modify them, or use them from your system? It could be a marketing channel like a website, a mobile app, or a marketplace. Or it could be a human on your marketing team who uses or or add some content. So what we are now seeing is that you also have to treat AI as a content as a constant consumer. So this means designing your DAM so agents can access, enrich, and modify DAM content in your home. And they can also follow the same permissions, brand governance, and rules as humans. And as these AI agents work on all parts of content ops, we must also extend policy rights and compliance from the DAM all the way upstream to where content is being created or downstream to the point of consumption. So it's not enough to store assets securely in the dam. We must ensure that when content is created or when it is deployed, we are ensuring compliance, we are checking usage rights, and applying brand controls on it. This here also underscores the importance of human oversight. So teams must train their agents, they must monitor their outputs, and also provide continuous feedback. And this in general help enhance the accuracy and also improve performance over time. So the next result of all of these changes is that in an agentic DAM, content is now transformed from this passive file storage into a system of governance and execution. So DAM now has a direct influence on content creation, on marketing campaigns, and even on customer experiences. So next, you must be wondering, okay, if your DAM is truly agentic or what the parameter is for a DAM system to be truly agentic. So if you go to the next slide, the first difference that I wanted to highlight here is the role of AI. A lot of DAM vendors today offer AI assistance on discrete tasks such as, you know, tagging assets or searching faster or even getting suggestions. Oftentimes, these features can also be human or event triggered. However, in an agent exam, AI agents execute work continuously across the value chain to achieve these desired outcomes. So for example, agents could continuously enrich asset metadata automatically as they get added to the dam, or it can enrich only the file types that are recognized as product shots. And, you know, no one needs to tell the agent to do it. You can just set up the rules and let the agents do the rest. Next, coming to the operating model, in legacy DAM systems, workflows are still led by humans and enhanced by AI. However, in an agent exam, agents lead the workflows and humans maintain monitoring and control over the operations. So for example, agents automatically figure out figure out which content assets need branching reviews. They can annotate it with the right comments and suggested edits, and then the humans can then approve or cancel the agent's feedback. Now coming to the next difference, that would be on the scope of automation. So if you think about it, in a traditional DAM, some specific user tasks are automated such as adding tags or searching for assets. However, in an agentic DAM, automation is an end to end orchestration across the content ops life cycle all the way from content creation to content use. So an example here could be you can automate agents to work in parallel or one after the other using intelligent workflows. So for example, one agent can annotate the asset, and then as soon as a human approves it, another agent could create all the necessary variants without the need for human intervention. Now next, coming to, governance. Governance here is also enforced differently. So in a legacy dam, governance is applied before or after certain actions can happen in a dam system. But when governance is truly agentic, it is actually applied continuously at run time. So let's give an example here. Let's say an image is created in a creative tool. It can already tell you that some of the primary colors you used are off brand. Or upon ingestion, it can catch AI generated portions and send it for human reviews. Or when it has been deployed on a website, an agent can tell you, this particular image is not usable in Europe because models only gave you usage rates in The US. And so finally, as a result of all of this, the last piece here is around scalability. So it's not just about handling more assets, but also about scaling automation alongside them. So whether you're managing millions of assets or launching thousands of campaigns, you don't need proportional increases in time, cost, or effort to make it happen. A lot of BAM vendors, again, are limited here because the systems, still rely on very manual human driven actions, so very lightweight, agent use cases. They could also lack a strong content model and infrastructure that is built for sustained automation. So as a result, these systems may not scale as when compared to an agent exam that is built on a foundation of very rich content modeling, enterprise grade infrastructure, and continuous innovation. Now we next dive a little bit deeper into exactly how an agentic DAM operates. So if you go to the next slide, it is, if you think about it, it is built on three core principles. So firstly, agents are persistent, which means that they are always on. They never sleep or wait for someone to trigger them with a click. They can act continuously as content moves through the ecosystem. So this eliminates these manual bottlenecks and also moves content faster through the process. Secondly, agents are also orchestrated, meaning multiple agents are coordinated within the DAM system. So the orchestration rules or workflows can be customized and tailored for each business. As a result, automation works end to end to expedite marketing campaigns and not in pockets or in isolated ways. And then, finally, agents are also governed with human in the loop. Policies are enforced continuously while exceptions are still routed to humans. And most importantly, humans still continue to have visibility and control over how agents work and also the quality of their output. So with that, I'm now gonna hand it over to Tarun to walk us through each step of the agentive content operations and what agents should achieve for you at each step. Tarun, over to you. Thanks, Jitroopa. So, yeah, at this stage in in the presentation, I'll basically try to go through a little bit more of a grounded look at, you know, what actually is happening that we see, in an in an agentic dam. Right? How does it affect your kind of step by step content operations? And we'll look at this at kind of different lenses of detail, so you can get a better grasp of, like, what what is it actually that's going to end up happening here. So if we really think about the content operations process, right, and we think about how we can frame up it in, an organization of how we start bringing agents in. Really, it's about taking each part of that content operations process and starting to figure out where agents can further automation of the tasks that people are doing today. Right? So if we start the very beginning of a content operations process, typically, that's gonna start with something like planning, where I'm creating campaign or content plans or briefs or things like that. Right? So instead of a human having to go through and create that, an agent can go through and look at, say, all the objectives of what your, you know, marketing is supposed to be able to achieve, start planning out different campaigns, start proposing different campaigns and content briefs, take into account the different rules that you have around all of that, and ultimately create plans for what your marketing might end up looking like. It might even go through and search the DAM automatically to find either inspirational content that maybe came from last year in the same kind of theme, or content that already exists that could be reused for certain campaigns or for certain content needs already just at that particular process. Right? But, eventually, what we expect is that, you're gonna end up having content get created as a part of these different campaigns and, and, you know, content cycles that you're running. Right? And so, ultimately, that content needs a home. It needs to go into a system of record where it can power, users to go find that content as well as how Satrapa mentioned power agents to go help get that content. And so, it becomes of utmost importance that when that content ends up in the DAM, it is cataloged incredibly well. Right? So a huge thing that we see here is librarian agents coming in and enriching content automatically to a degree that we've never seen before. Like Satrupa mentioned, we've seen things like AI powered tagging and things like that, but now agents are so powerful, they can understand your taxonomy. They can understand, the complexity of a content model that you might have and go in and fill out fields that historically only a user would be able to fill out. Things like abstract this particular piece of content for me or generate an alt text for this or figure out what product category and product is depicted here. Figure out what kind of content it is. Right? Is it a photograph? Is it a lifestyle shot? Is it a park shot? It's a white paper. All those kinds of things can now be done through agents and populate an an extremely robust metadata model that powers further automations down the line or just helps humans find content a lot easier. Right? So if we look at the next stage, right, typically, what happens after you get content into a system, you don't wanna say this is available for everybody. Right? You it goes through some kind of review process. Right? Is this content actually compliant? Well, historically, the way that has worked is, sure, someone goes in and creates some content, and they do their own little assessment of it, but it goes to some subject matter experts that can actually look at this in a variety of different ways. Right? Folks in regulatory industries, especially, are are probably sitting up and taking notice. Right? Because that's where, the the process can slow down a lot and be expensive and time consuming to actually go through. Right? Even outside of that, things like general brand compliance. Right? Are the logos obscured? Are the colors correct? And then the more compliance and regulatory industry settings, the language being used, something that is approved language that can specifically be put there as opposed to something that's just a tone check that says, hey. It's the tone of voice in in line with our brand standards. Right? Both very valid checks that need to be run, but different agents might be out there to do those different kinds of checks. And so when we look at this automation review, we divide these things into what we call credit agents, which are about subjective quality, of the actual content itself and evaluation of how well it might meet a campaign's needs, things like the tone, things like does the message match the segment, things like that. And then we have compliance agents that say binary yes or no. Like, does this meet or violate some compliance standard that, prevents us from from actually using this particular piece of content? Right? The cool thing here is that you'll you'll kinda connect the dots on these pieces where, you know, historically, we've had to pass to a human to go do review and get that first round of feedback. But now as soon as you upload and the content gets enriched, we'll be able to have agents actually review that content on your behalf. And so even the uploader gets to see through preflighting, an ability to understand what issues might arise in that content right away before you go bother a reviewer to take a second pair of eyes on it. Right? After that, we start looking at the concept of search. Right? Once that content is, uploaded, enriched, and approved and, you know, we say it's all compliant, well, people have to find it and agents have to find it. Right? So, we'll have agents that come through that can perform search on people's behalf. We'll have things like the librarian agent that, upload content I'm sorry, enrich content on upload to make search easier. Right? So there's a lot of benefit to the enrichment early on to benefit search, but we'll see the search process itself change as we, put out more technology that allows agents to find content for different needs as well as empowers humans to do better searches of content as well. It's a really cool area. Then we've got this area of content delivery. Right? So, how do I take a piece of content that's come up and it might be, say, a Photoshop file, but in order to actually deliver that, I need different aspect ratios. I need different, different crops as a whole. I need different, file type variance to deliver as high res for print or for, you know, low res or or something that's very web friendly. Right? Things like that. So I need to be able to cater to all the different channels that I need that need this content represented in different ways to have optimized delivery. So this is where production agents come in and actually start automating more and more of that process. Right? A lot of people probably have things like AI auto cropping today and different creations of these kinds of different renditions of content that power different channels. But with agents, we're gonna see a much wider ability to do much more rich, prescriptive transformations on content as a whole, which leads to kind of the last stage, which everyone's probably thinking about, which is personalization. Right? So it's not just about saying I can take this content and I can, transform it to be the right format for this particular delivery channel. Right? And it fits in the right spot on my website or it's, you know, web optimized or, you know, the color is faded a little bit so that I can put text on top of it. Not just that, but how do I actually start going through personalization pieces where I can say, I can start creating, generated content that matches to a certain segment or to a certain person based on their history, and then composite it together with other approved pieces of content as well. Right? So it's a huge area around personalization. And then, of course, the whole thing is a cycle, right, as it always is, which is now I gotta go and analyze and optimize. And instead of having to have humans go through systems where they are looking at aggregated data across all of my different execution systems for campaigns and content and trying to figure out what worked and what didn't, well, now you can have agents go through and decide that as a whole and then ultimately use that as part of data input into your planning process. So the next time you run a campaign around the same theme or the same product or the same, audience target, you can learn from those things and have agents take that into account all the way at the top of the process. Right? And it's really cool looking at it like this way and looking at, different agents powering different parts of the process because what it enables you to do is figure out where's my biggest pain points. Right? Where do I go invest in first? And we'll talk a little bit about that too. So if I go a little bit beyond as well, I wanna talk about this concept of, where the agents actually live and how they interact. Right? And what you're gonna end up seeing is you're gonna have the ability to, have vendors that come to you, and they'll say, look. We've got stock agents in our system, and those are gonna be fantastic for you because those are gonna be things that these vendors who are working with a lot of different customers are going out and they're saying, we know the general needs of everybody, and we're gonna go ahead and build up a bespoke agent that you can just flip a switch, turn on, have all the controls you need for, and be able to execute very rapidly on turning things on, and starting to automate things using agents inside the system. Then you're gonna have things that require custom agents, things that are really business specific to you or things you really feel are worth the the investment in tailoring quite a bit. And so what's gonna happen is we're gonna see the scaling of agents across the content ecosystem. So for instance, if we look at the top left here, right, we think about that. You can kinda see that circle being that same, level of phases that I gave you on the previous slide. But maybe what you want is a third party agent going out there that ends up doing your briefing, something that you've built in in an agent orchestration platform. We're seeing a lot of customers license agent orchestration platforms or even home grow their own as well to be able to interact with this stuff. And those third party agents need to be able to do things like access content and information in the DAM to help power their processes even if those agents are running outside of a particular vendor's piece of software. Right? And what we basically can end up seeing is a system where maybe you've got third party agents doing planning, maybe you've got librarian agents that are stuck in your DAM that are executing, but can be augmented as well and extended, by your own, implementations of agents that you're building in house. You would see critic and compliance agents kind of being the same way as a whole here where, you can take advantage of maybe stock review agents that, are being shipped out there. But being able to look for vendors that support things like expanding into your own bespoke, review checks, and and agents doing their own reviews, right, the way that you need them to happen. Because your rules and how you do reviews, especially in things like heavily regulated and compliant industries, might be something very specific and bespoke that you wanna build. And so you wanna make sure that your your agents are able to plug in at the right points in the process and, and facilitate a great experience for your users as they're taking things through, a content operation cycle. Right? And then production agents is a huge area here. It's interesting with things like image generation models, how the general image generation models still struggle with, you know, creating things that are on brand, right, and and, making sure that it's all according to what their specs are. What we're actually seeing is more and more customers do custom trained models for things like image generation for very specific products, right, and and very specific brand colors and very specific messages. We're getting success with that, right, out of the loop really point specific things, which means today, the likelihood is a lot of those production agents do very specific builds can be done through third party agents and plugged in, but there's also a lot of general functionality for things like removing backgrounds, for instance, to, allow you to composite images with AI generated personalized backgrounds that can be kind of a commodity type service that, you end up seeing baked into a vendor. So, again, it becomes this kind of this, ecosystem of agents where you can take agents that you've built and plug them into the vendor systems that you have that are controlling the orchestration of the process. And, ideally, even, the next level of that is you should be able to take the the, the great agents that your vendors have built and run it on things that are content sources outside of what that particular vendor controls. So for instance, taking a review agent and shipping it upstream so that, instead of the first time a review gets executed is when a piece of content hits the dam, you should be able to run that review when someone's working on something in Creative Cloud or Microsoft Office or stores it on their OneDrive. Right? So the moment they're saving that stuff, they're getting some feedback out from that agent that says, hey. This is not brand compliant yet. Don't send it in for review. Don't even bother uploading it to the DAM yet. Let's kinda get it fixed up and brand compliant first, and then we can move forward with the process. So hugely, hugely powerful as we look forward to the future of agents being able to kind of loosen up boundaries and operate kind of across different platforms and and, you know, through your your marketing technology ecosystem. So I'm gonna go through one more very specific grounded example here of how we might see something flow in the system when we look at a particular piece of content that might be, have processes automated by a series of agents. Right? So if we go back to that concept of a librarian agent that's going in and enriching content as a whole. Right? This is not just a, hey. I'm going in and just, you know, applying some random values into the the metadata. Right? This is really intelligent. Right? What this is is is typically a sequence where you go in and the first thing that, the librarian agent does, it says, what kind of content is this? Right? Is it a product shot? Is it a lifestyle photo? Is it a white paper? Is it a technical spec? Like, what what is this content that's actually, coming in front of me? And the cool thing about that is that once it knows what the content is, it can start reasoning decisions off of the next steps in the process. So for instance, you can have something that comes in and based on whether it is a product shot, which might need the specific product or SKU number or color or whatever, all tagged as part of the the metadata, the very specific metadata, that might be very different from a lifestyle shot of what you might end up needing to categorize as metadata for, for lifestyle shot. Right? Things like that. That might even have multiple products in it. Right? So a huge, opportunity here is that these agents can come in and decide what this content actually is, and then it can dynamically decide what kind of additional metadata needs to be populated on here. And not only that, but it might come in and actually say, okay. Based on the actual product that's in here and based on the mood of the image, maybe my alt text that gets generated or my description of this asset that I wanna use on my website gets generated differently based on what this content actually is. And so the sky is the limit on how you can basically have, a system that can start with saying, okay, we're gonna go through and kinda figure out how to automate this for one content type. But see, now I have a 100 content types. I have 200 content types and 10 or 20 different brand or brand categories or brand voices, and you can imagine how big that tree can grow of automation on how you can have these agents come through and automating every part of that particular process. Right? And a huge value of this is that, again, it's not just tags that you're throwing in. It's actually saying, okay. You know, the the the, the business has a need for a bespoke metadata model, a content model around this stuff. I'm gonna fill out that model. I'm not just gonna throw random tags on this. So hugely, hugely powerful and hugely, hugely intelligent. And as we go beyond that, right, we start thinking about things like review. Right? So now after we've gone through a piece of content coming through and, having all of its content get enriched, right, all all the metadata get enriched. Now I can decide based on what this content is, what reviews do I even run. Right? I don't need to run, you know, 200, 300 reviews on every piece of content. That's one, time consuming, two, costly, three, creates noise for your users. Right? But based on what that content is, I might know the reviews that need to be run, or it might impact how that review is run based on what that content is. Right? And so if you get your agents to a point where you have, content automated, content metadata being automated and have a really highly effective agent doing that, well, now you can jump into the next step of the process of having agents automatically review this content with a very bespoke review checks that are specific to that kind of content as well. Right? And this is where agents can do things like add annotations on assets to help aid human reviewers, how humans can collaborate with agents and say, an agent found something. Is it a problem? Is it not a problem right? Let me talk to it about this stuff. Let me note something for, you know, the the creative to actually go and and, address this particular issue, all that kind of stuff. Right? And, again, it it covers this idea of objective, like, oh, man. Like, this is a total failure. We cannot have this content go through because it has this very, very strict and hard compliance failure versus subjective things. Like, hey. By the way, here's your brand color palette. I'm gonna put it up next to the image, and I'm gonna give you a general assessment of, hey. I see a color splash here, and it is within the brand palette, so it's good. But I see this other big color splash, and it's not quite within the brand palette. Right? That's ultimately up to a human decision to make potentially on, on if that that matches compliance or not. Right? So it's, it's it's very, very impactful for, a piece of content to go through and be able to have all these different reviews run against it that are really specific to what that content actually is and give information to humans as to whether things are, you know, hard yeses or nos or information that might just help them make a decision on whether a piece of content should be marked as approved or or it shouldn't be. Right? And then lastly, we go into this this area of production agents. Right? So this idea that, great. Now that I've got, my metadata enriched and I passed review, now I need to start scaling. I don't wanna do that before I pass review and incur a bunch of time and cost and noise and files to create, you know, a thousand different variants of a file if it's not gonna pass review. But now that I've gone through that and I know what the content is, I know what channels it might serve, and I know all the variants I might need based on the segments that I'm targeting and all that kind of stuff, I can come in and I can start scaling the production creation of asset variance on that stuff. Right? So things that are as simple as different crop sizes for different channels, things that are as simple as is it high res or low res, does it need to be a web p or an AVIF to go out to the web, all that kind of stuff, or it might even be stuff that ends up being really, really, segment specific where you're personalizing content to specific segments and doing things like doing model swaps or color swaps or product swaps or even just saying, look, I have a product shot. I just wanna place it in 30 different backgrounds that might appeal to different segments and audiences. Right? That's the kind of stuff that can happen as a result of this. So you can imagine if we go back to the start of that idea of one piece of content coming through and librarian agents enriching the metadata and deciding what reviews to run on it and helping human reviewers, figure out what matters in this image and what doesn't. And then ultimately scaling up the production of this, you could start piecing together this idea of as you gain a little bit more trust into agentic systems and how they work and build up your agents. Right? You can get this whole pipeline ended up ending up being more continuous automation as opposed to, oh, we're stopping for a human to do something. Oh, we're stopping for a human to do something. Right? So it's really, really powerful. So I know that's a lot. Right? A lot of people are probably sitting there thinking, how can I possibly get started with this? And I'll tell you the and the landscape changes every day. The amount of news that comes at you is just just nuts. Right? And so a lot of what our customers come to us for is saying, great. We have all these areas that we could automate. How do we get started? Right? And and our philosophy is generally crawl, walk, run. Right? What you're gonna end up doing is you're gonna end up saying, let's start automating bits and pieces of the process. Right? And, let's go through and figure out what is a certain pilot group that I can work with, which certain area that I can work with. Right? So, typically, we see a lot of customers start picking one or several areas to automate. They don't boil the ocean. They don't come in and say, we're gonna do it for every brand we have and across the entire, the entire organization at this point. We're gonna pick a pilot group to learn. We're gonna figure out how we do this well within one pilot area. You see a lot of our customers starting with things like automating metadata population. In fact, the vast majority of our customers have, metadata automations in place already, right, to actually enrich their content and start moving on that. Right? And then the second thing that we end up seeing is then graduating into this idea of, great. I started a process. I started automating. Right? How do I start, looking at whether that that output of the agent is good or not? Right? So then you start monitoring and optimizing output. And an example here is, you know, we offer things like dashboarding, right, that come in and let you understand what is the actual metadata population rate, how often is user changing it. Right? Those kinds of bird's eye view paradigms help you measure the system at a bird's eye view and see, are things actually moving the needle, or am I just creating noise? And if you're just creating noise, that might be okay at at the start. Right? There's always gonna be a little bit of noise. The important part is you go in and you can go tweak the system to get better and better and better and better one bit at a time. Right? It's not gonna happen overnight, but you'll start automating more and more and more, and you'll get better and smarter at understanding how to interpret things and how to shift the system to get more and more automated over time. And then lastly, the other big thing here is understanding when to keep humans in the loop. And at the start, humans are in the loop everywhere. Right? And they're gonna continue to be. Right? They're gonna be high priority things that you want a human to look at. You're and you're gonna go through. Right? You're not gonna end up having a marketing department that's one person sitting in a room with 20 monitors, seeing all the agents fly by and do their work and try to course correct them all at once. Right? There's still gonna be humans that come in and need to actually watch the process. They need to be flat. They're gonna be points where there's flagging on that stuff. But, eventually, what's gonna happen is you're gonna start trusting the agents more and more to do certain types of automations, and you're gonna let that loose. You're gonna say, okay. For this particular brand, for this particular type of content, I'm pretty comfortable that it's gonna populate all the metadata properly. So instead of having a human go and review and make sure the metadata is all good, let's just skip that part now. Let's go through and say, let's just jump straight to the automated reviews. Right? Because now I can go through and run the reviews, and the point that I really wanna stop now is saying, okay. Let's get the review agents to come out and give their output, and then I'll decide basically what to do next as a result of that. So when you start thinking about that concept of where humans are in the loop and where you build trust in the system, that's how you end up scaling from kind of point specific, do this workflow, do this workflow, do this workflow all the way through to more continuous automation cycles. Right? Again, you won't get there overnight. Right? But you will get there bit by bit by bit by bit by automating pieces of the process and building that up over time. So with that, I'm gonna pass it back to Satroupa for a little bit to talk a little bit about the benefits of an agentic DAM as a whole. Thanks, Darren. So, of course, the crawl walker approach, I think, is what we should all be adopting. And let's talk a little bit about if you orchestrate all these agents successfully, what are some of the outcomes you can expect? Firstly, campaigns can be launched faster, and it can drive higher ROI because agents automate critical steps such as enrichment, reviews, variant creation, and also remove inefficiencies. Secondly, brand and compliance risk is reduced through foolproof reviews ahead in the creation cycle in the dam and also during run time on the marketing channels. Thirdly, teams focus on high value activities because AI handles repetitive work like metadata creation or annotating assets or creating variants. This gives time time back to your teams to be more creative, to be more strategic, and also to do more human to human collaboration. Content also becomes more visible internally and externally through richer metadata, all tags, and SEO and AU optimization content. The entire content ecosystem is strengthened as assets are enriched and validated both upstream and downstream. And then finally, organizations can optimize agent performance over time by training agents to understand the business and also improve their output. So, if you go to the next slide, I'll share a little story about our customers. Our customers are already seeing significant benefits from deploying AGTIC DAM. And this story is from a Fortune 100 life sciences company operating in 118 countries. So as their content volumes increased, it became more and more difficult to maintain consistency in metadata classification and quality. Now when you're in the health care industry, this can create regulatory risk, which means their content ops was slowing down as they needed to validate claims and important safety information against source libraries. And these manual review processes were not just slow but also very prone to human error. So approval agent, DAMP, fundamentally changed this operating model. Our librarian agents now automatically populate predictive predictive metadata during ingestion, ensuring every asset is classified consistently from the start. Secondly, compliance agents identify claims, within the content and validate that those claims are correctly linked to approved reference materials. And then, also, a third party compliance agent analyzes content to detect safety language and confirms alignment with the most, current approved library. So even though this analysis happens outside the dam, results are surfaced back into the DAM, making the review and the approval process and workflow even faster. So this consistent automation of metadata classification and identification of preapproved claims really accelerates review cycles for the customer and also helps them stay on top of compliance requirements even when their content volumes go up. So this brings us to the end of the session. Before we break and take some questions, if you go to the next slide, I wanted to leave you with this QR code. Scanning it will take you to our 2026 dam trends report, which is written by some of our veteran dam experts. It's a really useful resource if you work in constant ops or you use the dam system regularly as a part of your job. So do scan the code and check out the asset. And with that, we are now gonna take some questions. I know we have a lot of questions coming in. So turn over to you for the q and a segment. Yeah. I've, these are great questions coming in. So I'll just read off the questions, and I'll just kinda give give answers. So the first question that came in was, is defining taxonomy and creating a metadata schema infrastructure for DAM more or less important now that the agents are auto attacking and ingesting data at scale? This is an awesome question. I love this question. The answer is it is way more important. It is way, way, way more important to have a really defined set of information. And the reason that is is because, you know, having structured data allows agents that are going to say search for content or leverage content, have a much easier time finding it. Right? And and even populating it. Right? So I'll I'll kinda put it this way. Let's say that you just had one metadata field that said, it's a open text field. Right? Throw in all the metadata in this field. Right? Whatever you think is relevant. Well, in that case, right, every time something might run to go and enrich that content, well, it's gonna kinda decide on the fly. Like, what should I put in here? Right? What shouldn't I put in here? All that kind of stuff. The key to having a really successful, scalable agent infrastructure is to have the right kind of input to the system that allows it to make decisions that work for you. And a huge part of that input is going to be the schema of the metadata that actually matters and what the point of those fields actually are. Right? So a good example of that is if I go in and I tell an agent, tell me what the metadata is for this agent. Right? It's gonna give me, again, a random blob of text. Right? And it may try to categorize some stuff. But if I tell it, I need these 10 or 15 pieces of metadata, right, based on what this content type for a a photo, a product shot. I need this metadata. Right? It's gonna have a much better idea of what you're looking for and be able to populate that. Then on the flip side, it makes it easier for users to find, makes it easier for you to curate content in collections that agents can end up using, and it makes it easier for agents to ultimately find content to the dam when you give requirements, like things like don't go use something that's expired or find something in this product category or make sure you're grabbing the, the variant of this particular product shot, that shows it in a grocery store, right, or whatever. Right? Things like that become much easier for agents down the line to perform if there's a strong metadata model behind it. Right? I I could talk about this particular question probably for an hour, but I'll throw out one more more thought before, we keep keep going. But the other aspect here as a whole is that, with with a much more defined metadata schema, you have this point of review, right, where a human can come in and say, yes. This is right or, yes, this is wrong. If you don't have strict metadata schemas, that becomes a lot harder to do. Right? In order to scale an agentic system, a lot of the kind of pattern that you're looking for is an agent comes in and it does something, and then it gives an output that writes it somewhere that a human being can go in and audit and review and understand. Because if the next agent comes in and picks up from where that agent left off, you need to understand whether it was the first agent that goofed up on something or the second agent that goofed up on something. And by saying I have a really robust metadata model and saying the first step of the process is to enrich that content and have all the output right to a robust metadata model, all of a sudden you've got this contract of how the other agents understand how to interact with your data. Right? So super, super important to keep going on a really robust metadata schema. The next question was around, saying thank you for breaking down the difference in having AI centric features in your DAM versus an established agentic DAM. My question is, what are the platforms available to create your own agents? I've attended several webinars about AI agents, but no one has really specified who to talk to in your company about building an agent or if there are self help platforms to create one yourself. This is a really great question. The term agent gets thrown around a lot. A lot. A lot. A lot. And there's different definitions for what that is based on whether you ask someone who's really, really technical or someone who's trying to sell you an agent platform. Right? And so, the way that I like to look at this is essentially, you've got categories of systems. Right? You've got this idea of systems that do have their own agents built in. Right? And they'll have certain configuration, and they've got essentially their own prebuilt logic and knowledge and understanding that you can just kinda use off the shelf. The second kind of system is typically things like an agent orchestration platform, something where you can build your own. You've probably seen, commercials for things like Agent Force or Microsoft Copilot Studio, or, n eight n is another popular one, and Gleam, we we hear quite a lot of bit of. And those are more platforms that are more akin to traditional middleware platforms out there that let you go in and kind of build drag and drop flows and connect up your data and all that kind of stuff. And the advantage of those platforms are you can go in and you can build whatever you want. The disadvantage of those platforms are they don't have any opinionated way to do things. So you're kinda starting from scratch and building things up every single time. Right? So there's kind of a spectrum of things, and I'll I'll throw it as well. We've seen some customers have their own IT departments build up their own agent orchestration platforms as totally custom builds as well. Right? We kinda see the spectrum of things out there that that exist. And I think what's really gonna happen in the future is, you're gonna have those agent orchestration platforms. You're gonna have custom built. You're gonna have vendors that come in and bring agents that are easy to plug and play that become smarter and smarter and their capabilities expand. And then you're also gonna have vendors out there that say, we're we're not built bring bring our own agents. What we're actually doing is we're gonna help build things called skills or tools that help agents actually expand what the possibility is of what they can do. Right? So hopefully, that kinda answers your question. I know, it's not, it's not a slam dunk on, like, there's one thing that I can answer here, but there's kind of a spectrum of things, and that's how I like to look at it. So let me hit the next question here. Regarding usage rates restrictions, what's the liability model when an agent surfaces or remixes an asset it shouldn't have? Oh, I love this one too. This is why this ties back to my first point about, metadata schema infrastructure being so, so, so important. Because the point here is that if your librarian agent comes in and enriches your content properly, the agent that fires and is told, go create all these variants of these assets. If the metadata model is set properly, those agents should go in and only touch the assets that they are permitted to actually touch. That's part of the value of building a scalable, trustworthy, agentic system. Right? I'll also say that accidents happen. People mistag stuff. Maybe an agent mistag stuff or something like that, and this is where a human in the loop becomes so, so, so important. And this concept of where you trust and where you don't. Right? If there's something that is very high liability, high risk for you, right, you wanna make sure there's a human in the loop at each one of those processes that if the agent goofs up and changes something and does something, that goof up exists inside the DAM. It doesn't exist out on your channels. Right? And so that's kinda where you have to decide where are the right controls for each type of content, for each type of process moving through depending on how high risk or low risk it is for you to execute on a you know, in an automated way. Next question. How do you see agentic AI different from how databases slash dams have auto populated metadata upon file upload or auto changing file sizes? Systems have done this since the nineties. The big thing is is reasoning about content and being able to understand content in a really, really big way. So, you know, if we think about stuff where, historically, right, you can think of the most baseline automation is something like, well, I looked at the file type and decided it was an image. Okay. Smart, but not super smart. Right? Next level up is I'm using cognitive AI that is a pretrained model that lets me understand this has a bunch of things in it like a ball or a chair or a person or things like that. Right? And then you've got stuff that ends up getting more and more sophisticated things to large language model based AI. And so with stuff like that, you end up getting a huge degree of sophistication that we never have been able to get before. So case in point, populating an actual bespoke metadata model and changing how it populates it based on what that content is, the same way a human might look at it, that's how agentic AI changes things. Right? That's that's one part of it. Right? But then the second part of it is saying, great. Now that I have that in a very prescriptive bespoke data model, I can now have in an an agentic DAM, the rest of the operations process be influenced by what that is. You only fire reviews on the right pieces of content. You only fire the right transformations on the right pieces of content and things like that. And it's only possible when you have a rigorous metadata model that's out there and smart enough AI to populate it to end up fueling the entire agentic system. Right? So next question was around, can agents reliably distinguish Gen AI versus human made provenance, for example, c two p a, and propagate that downstream when an agents get reused or, recomposed? Another excellent question. So the the short answer is, you know, that there's, there are probably a lot of solutions out there that attempt to look at whether something was AI generated or not, but those things are likely not slam dunks. A lot of those pieces of technology are based on, other AI models trying to assess whether something was created by AI. And as things get more sophisticated, then that's gonna be harder and harder to do. Now c two p a is a great call out. Right? So for instance, you know, what what happens in c two p a is, vendors will go and as things are AI generated, write metadata to the file saying, this is the provenance of this asset to sell things have actually changed. And the I think the long term promise is that the c two p a data sticks with and lives with the asset. Right? In a practical sense where we are today is that when you take something, let's say a Photoshop file that has the c two p a data in it, essentially, that lives as metadata on that particular asset, and you go pass it through an engine that says, I'm gonna make a thumbnail out of this. Right? That engine very well might strip the c two p a data. Right? And so that's kinda where things get really, really, complex. And in the long run, ideally, you're working with tools and platforms that have adopted c two PA properly and can keep that provenance record held. Right? And it does end up getting held on to as things go downstream. I think today, that's not quite what's happening because that is not a standard that has been adopted, and rolled out through the entire process flow. Right? I'll also just throw out in general. Right? There's other ways to start thinking about this with AI where, some of the experiments that we're doing and some of the technology actually that we're investing in experiments actually is probably the wrong word because we're rolling stuff like this out now. There's technology out there that allows us to say, let's take an asset that's out on your website and identify that that asset is a cropped grayscale version of this original asset back in your dam. Right? Without needing that c two p a piece. So I think it's lovely if the c two p a piece is there and it actually does end up giving you this, like, chain of custody, of content as it goes out, but it you can't assume it's always gonna be there, so you need another solution to trace back. Right? And make sure you understand how the original asset was even created if that's something that you you're concerned about. Oh, man. Lots of questions rolling in. I don't know if we'll be able to get to all of them. We just got a couple minutes left here. But let's see. What kind of labor do you envision dam managers needing to do prep, for the data for agentic AI to be able to use it? Content model creation and, really just shaping the AI. Right? A lot of what you have to do is is design the system. So, for instance, right, that idea of the flow of first identify what kind of content type it is and then figure out based on the content type what metadata needs to be applied. What are the parent child relationships between all those fields? What fields need to change based on what the content type is, what brand it belongs to? Right? What fields are needed to end up driving automation further down the line, like what reviews run, what transformations are created. That's gonna be the new role of damn librarians, right, which isn't actually that different from before. It's just your consumer isn't users. It's also going to be agents now. Right? And why it's so important to be so rigorous on your modeling and your content modeling specifically and making sure that you are figuring out what can AI actually assess, how many different content types do I actually need to fuel the overall agentic system. Right? Next question. If an agent can generate a compliant on brand asset on demand, do we need to store as many sets at all? Oh, great question as well. Right? I would say let's cross that bridge when we come to it. Right? Because right now, we really can't. Right? The cost to generate an on brand asset that a 100% meets your needs is, is very, very high to train models. The most successful customers I've talked to that do this are at a point where they created models that can generate one kind of asset and only if the right human being is prompting it who understands how the model is trained in order to get the results they want. Right? So you're right. In the long run, we might end up doing things more on demand, but generating assets on demand is expensive. Right? And creating things on demand is expensive. And it also reduces auditability. Right? Because you probably aren't super comfortable saying, AI, go create a personalized asset and have it show up on my website without someone looking at it and making sure that it's doing something okay. In the future world, maybe we're gonna end up seeing this whole thing be automated, and it'll all be magic and you'll trust it entirely, right, at some point, but we're not there yet today. So we've got to take advantage of the automations that we can today, build that trust, and build up those systems over time. Would DAM managers need to enter in clear schemas or documentation on metadata profiles, for example, librarians to be effective? Yes. Absolutely. Absolutely. Right? So you you need to be able to tell the agent here's how to populate these things. We call those prompt hints essentially, right, to explain and say, let's try to just tell the the agent to fill out this field, and if it isn't able to do that correctly, go ahead and shape how it fills it out based on the hint that you give it to shape what its output actually is. What is the recommended sequence of employing agents if you're starting from scratch? Start with the planning or production. What are the dependencies among them? You can start automating at any point in the process. I'll throw out that the vast majority of our customers have seen huge gains with, starting to automate metadata, which I I believe is kind of the start of the whole system. Right? You can automate metadata, then you can start deciding how the rest of the system behaves. But we've seen customers say review is really important to us, so let's go start automating reviews. Content transformation is really important to us. Let's go start automating production. Briefing is really important to us. A lot of customers send out multiple pilots running in multiple areas, right, thinking about segmenting them in those ways to to get moving forward. But my personal favorite that I recommend customers to start with is metadata automation because I think that powers a lot of the system and the technology is at a maturity state where you can accomplish a lot with that. So I think, fortunately, we are almost at time. So I'd like to hand it back to Theo. But I think, again, really great questions that came through, and hopefully, we were able to answer as many as we could. And we'll try to respond, offline as well if possible. Thanks, Darren. Theo, over to you. Thank you, Satorupa and Tarun, for a great presentation, and thank you everyone for joining us today. Just a reminder that you'll receive the recording of this session tomorrow so you can watch it again at any time, and we hope to see you again soon at another Henry Steward webinar. Goodbye, everyone.