Vision to Workflow, Magnolia Agent in Daily Operations. So my name is Nora Novak. I'm a Senior Product Marketing Manager here at Magnolia. And standing right next to me is Chris Jennings, our Senior Solution Architect from the AI team. Hi, good morning. Yeah, I'm a Solution Architect within the AI team. My job in this part of the organization is to help people find the real value in the AI tools that we offer. And I hope we'll take you through some of that now. Yeah, so we're not talking about the theoretical AI stuff today. We are going to show you how I use the Magnolia Agent in daily operations. I know marketing teams these days struggle. I'm part of a small team as well. So we are bound to do more campaigns, create at higher quality, but with the same or even lower budget. So we are trying out all these different AIs and piling up on them. But what the real issue is that we have still the old manual processes that are holding us back. So I've got the data to back this up. Three quarters of leaders are testing agentic AI, but only a minority is ready to actually implement it. And half of them are experiencing actually agentic sprawl. So we have all these big MarTech stacks and expensive ones, but we actually only use the bare minimum. We are lacking orchestration maturity. So our goal is now to build a smarter foundation. So we already hear that the branded search has changed. We typically were looking to optimize for Google clicks, but we are now optimizing for AI answers. So our websites have to be pretty storefronts, but they now also need to be structured, machine accessible, sound sources of truth. So that the AI is ready to read, interpret and cite. Otherwise we are staying invisible. And how can we fix this? We have, think of the Magnolia agent as the workforce multiplayer across these four areas. So detect, create, scale and optimize. These are the ones we are running you through today. And I've got a real life example with me today. So I've got a new analyst report for my consumer banking campaign, talking about consumer banking trends. And I need to find the content gaps for this campaign. So Chris, can you kick us off and show me how I do that? Sure. So we'll jump straight into Magnolia and we'll start to use the agent. Jan showed us yesterday how we get this open. And we can put a query in straight away and say, I've got some new reports in here. Can you find the report for me that runs information about trends in IT? So we just speak to it through the chat interface here. It'll run through, it'll find our document for us and we'll just see some simple results over here that give us a rundown of what it's found. So if we just take that apart. We ask just a plain English question or any language of your choosing. What new analyst reports have I got? Sorry. And this is the kind of result set that we get now from this agentic search. It's able to give us a snippet of what it found. It's able to give us a clickable link and we can see exactly what we're dealing with. Is this web content? Is this damn content? But the interesting bit of that quick conversation that we saw was this little purple box that came up. They said, I'm going to execute a tool called context search. So tools are the key of how the agent works. These are extra bits of functionality that we plug into the agent and they bring a new skill to what otherwise is just a chat bot that you can talk away to. So some of these skills are very generic. We can search content, as we saw. They can understand the metadata of the content within Magnolia. We can find stale content. We can go and search for something that's using a particular template that perhaps we want to retire and replace with something else. But also these tasks can be very, very specific. We can go and fetch data from somewhere else. We can really dig down into our content. These can be built in two ways. They tie in with Magnolia's light development framework. So you can build something without having to write any code. If you just need to explain something in a prompt, you want to go and fetch something from a REST API. Or you can write custom code. You can get these to do something very, very specific to your business needs. It becomes available to the agent. The agent is now customized for you and everything you've done is on your side. It's okay. We'll go from here. It's fine. There we go. Crucially, all of these tools respect the access rights that your user has. We know people are concerned about giving an AI agent access to everything that they own. So I can't come in, start using the agent to pry on content that I don't have access to. When I use the agent, the agent gets my user roles. You can only see what I'm up to. So how did this particular tool work? The context search. We've given the agent access to a vector database. Bill has mentioned this already just now. This is a different way of indexing content for a language model to understand the content that you have. The crucial takeaway from this is that this is the difference between keyword searching that we've had in Magnolia for a long time, and it's how you might interact with a Google search, and the ability to come in and ask something like perplexity or speak to chat GPT. Your content is understood and has context to it now as well. So now that we've got a document in mind, we can start to do something with it. So let's ask the agent, can you find three key trends in here that are relevant? As Nora said, we're running a banking website. We see we get a plan, and we'll highlight this text in just a second. We see another tool has run, and out comes our summary. And then finally, the agent is going to show us at the bottom here what it thought was relevant about those trends for the industry that we've given it. So again, we just asked a very simple question. What are the three trends that are relevant to me in banking? And first off, the agent gave us a plan. It said what it was going to do, the steps it was going to follow. The agent doesn't run as a black box. We can always see what it's thinking, what it's going to do. We can interrupt at any point and say, no, you're going in the wrong direction here. These are one of the highlights that we had come out. So it's taken the content from within the document. It's taken its general knowledge, its understanding of the world. And it brings that sense to us through combining the two things. And then finally, as we saw this non-black box approach, it told us why it thought that was relevant. So if we're not sure if the agent has understood, we can see what it's thinking. We can argue with it. We can get it to start over again. And the tool that we saw running here was this one called Summarize Document. So this is another generic tool that we've bundled into Magnolia. And it will take a piece of content and it will pass it into that agent. So it understands what are we talking about. It's how we start our discussion by loading it into that context. So this, we found the document. We found the relevant bits about it. But the question that Nora originally had was, where is the gap in here? Where is my content falling behind? So for this, we'll go back to the agent and we'll ask it to take a deeper look. Thanks. So we'll ask it, right, of those three topics, it understands the content. It's working as a conversation. Where do I already have some content? And where is there a gap that I'm going to fall behind? So again, we get a plan. And we see now, it'll run three searches. It understands that there are three topics. It's going to find content that matches all of them. What we see now is occasionally, these agents, they can lose track of what they're doing. But because we have the context in here, we can just nudge it. So did you find that gap I was after? And it gets back on track. So what we see now is it's going to generate an awful lot of text. Again, you don't have to read all of this. But we can see the agent is very, very verbose. But it's able to make suggestions here. And it understands. I've given you a lot to read there. So it says, would you like a tighter matrix for this to take away the key points of what I'm telling you? So we say, right, create that matrix for me. And we run. We get our table comes up through here. So it knows how it's going to be more helpful to you. And it knows how to give you the most value. So I asked it, of those three topics, which ones are the most valuable to me in banking. It showed me its plan. And then finally, what I end up with is this is my matrix to take away. So it's given us three key topics on the end there. It's given us evidence of where we already have some content. We could click on these links. We can go and check on what it's thinking. And finally, it's given me this list of priorities down the side here. It's found that last topic. We don't have much coverage at all. So, Nora, that's your priority. And that's your gap. Thank you very much, Chris. So how great it is that I don't have to click through the documents myself, finding the right report, clicking the right app. So with an only agent conversation in less than five minutes, I was able to identify the gap and create a new content idea. And from this content idea, I went to my copywriter and got my new content piece. But now the next issue is how do I get this content live? This is usually where my laptop dies because, yes, I'm that person with the 40 open browser windows. And, yeah, finding the right document, the right asset, getting the right designer image, or even creating an image generation model and being on brand. So how, Chris, how can the agent help me here? So, as Jan explained to you yesterday, our latest product research shows us people are drafting their content away from Magnolia. Perhaps they're passing around a Google Doc or they're putting things into a wiki page. And then some poor soul has to cut and paste that content into Magnolia. So another tool that we have is the ability to access the agent through different UIs to give it more information rather than just typing everything into that box. So what we're going to do here is we're going to go into the Pages app. We're going to say, right, I need a new page in here. We give it a title. We've got some preset layouts. We'll pick one of those out. And then Nora's draft content, we just paste it straight in. It'll understand the content we've given it. And we can just put in here, we can put in little hints, little placeholders. So I'm going to need an image to go in here. We'll run this through. We've just elapsed some of the timeout. We'll open this up and we'll see we've got our new page in here. And if we open this up, here's the content that we pasted in. It's understood our placeholder is in there. And if we click through into this, we can see it's turned everything into Magnolia components. We've not vibe-coded a page up that is not going to match things. It's not going to work properly. This is all now a standard Magnolia page. Nora, you can take this away. You can fine-tune it. Awesome. My page is live, but I wouldn't be a product marketer if I don't have running a different campaign simultaneously. So I need this page for my DAG region, my FBS region, what now usually takes days, if not weeks, getting the right word document, getting the translation agency done, maybe even internal colleagues to get it translated, and then translating this via copy and paste back to the layout. So this is where my campaign loses momentum. So can we not use the agent to do that exactly where the content lives, Chris? Sure. So our earlier work with the accelerator, we've already introduced automatic translations. You've already seen this before. But what we can do now through the toolchain is we can just simply issue commands to the agent and ask it to do things on our behalf. So if we see how that looks in this short clip here, we'll find our new page. And rather than open it and run through any actions down the side here, we'll simply tell the agent, I want you to translate this for me. Again, we've just taken out some time here. But we can see as well the agent can understand complex tasks. It knows it'll have to run the translate command twice in order to do this. And so it does that. It runs it for us twice. We can open the page up. We can see here's our preview. If we switch this over to German, we've got our German preview in there. If we switch this over to French, we'll see that the French is in there as well. And again, all of the content is within Magnolia. So we open this up, and now we can fine-tune anything that we got wrong about those translations. We can continue to work on it as standard Magnolia content. So although the action that we completed here was to create an automatic translation, what's important is that this is a tool that can simply go and execute bits of code that already exist within Magnolia. So anything that you have buttons for at the minute can be run. So the agent can take care of publishing things for you. It can take care of generating any missing search data. Everything that I can do, I can tell the agent to go and do it on my behalf, and I can tell it to chain these things together as well. So this is now where we would show traditional case studies. But I think you just heard the best proof points in the last two sessions. So where Union Investment and Prodiner showed their process optimization that you get with better prompting context and Bill showed us how we use that content for AI touchpoint flows. So this is exactly what the Magnolia agent provides. It acts as the centralized operational orchestrator. It scales patterns across all active workspaces and keeps your strategy human while execution is super fast. So I've got my page updated. But if we remember from slide four and Bill's talk, if I now publish, I publish for the old search model. And we now want to be discoverable not only for those, but also for the new AI search models. So Chris, can you show them what I do now? Sure. So we've heard a mention earlier about JSON-LD, and then perhaps you're thinking now, well, what is this? How do I get it into my page? So we've taught the agent to understand what you need. And again, it's just a very simple case of telling the agent, can you generate this GEO metadata for me? So again, we've just done a time lapse as this tool runs. It'll go away. It'll analyze your page. It understands what you need in there. And it'll explain to you exactly what it's generated for you. And if we just open up the page now, we can see the structured data is in there as well. So we've moved on from simply generating metadata and pulling keywords out of pages. We can generate this structured data as well. So we can bring a different schema into there. And we can generate microformats in there as well. So by simply saying, generate that metadata for me, we got a summary of what had been produced. And this was all through the one tool that can be adapted to run any kind of schema that you want to insert into that page. And it can handle those machine-readable formats. So Nora, your page should be ready for you now and ready to go. Awesome. So what we've just seen is our path to autonomy. My marketer role has shifted. I don't have to do these individual repetitive tasks anymore. I can focus on the intelligent, autonomous agent. No manual work for me anymore. I can highlight on the value strategy and the final review while the agent does the heavy lifting. It all started with a clear market signal. I had a content analyst report that we turned into structured pages while we scaled locally and ensured that the page is optimized not only for traditional search but also for air answers. Thank you, everyone, for joining us. And Chris and I will be around the rest of the afternoon to talk with you about other use cases or how you can deploy the agent workflow for your operations. Thank you very much. Thank you. We have a couple of minutes for questions. So let's see if there's one, two, three questions about the Magnolia agent. We can take them now. There is one. Thank you for that. That was really helpful. Your doc to page functionality, does it work for existing pages? So you have an existing page. And then in the document in Google Sheet, you change layout. Can you re-upload it and we'll update it in Magnolia? I think at the minute it would create another page. But you can... I chose that pre-built template of two columns. You can point that at an existing page and say, I want the same structure as that. So the result currently would be you'd have two pages. But I think we could easily improve that to say, no, I want to overwrite that one with this one. So yes, there would be that extra step at the moment. But I think that's quite easily solved. Thank you. Thank you for the talk. I would like to know something about the translation. So when it comes to translations, you normally have to consider blacklists, keywords, brand market names, etc. So words that you really have to be careful when doing the translation. Is it adaptable? In the skill set or anywhere else that these can be considered? So all the agent was doing was clicking the button on my behalf. And that button then runs through our AI accelerator, which contains a kind of a switchboard, a registry of different services that I have. So I can configure the prompt there that I'm going to send off and say, do my translation, but don't do this and make sure of that. Or we just rely on the service that is behind it. So because of the switching around, this could be DeepL, this could be ChatGPT. Just Magnolia would route it wherever you want it. So yes, you can put it in the prompt or you can set it as a hard rule within the provider of that service. And all the agent is doing is triggering that customized action. Thank you for your presentation. Very insightful. You mentioned that in the top right corner, I think in the creation pillar, that you also have image generation. Is it already in that sense that you can connect potentially your favorite image separation API token and you can generate images based on the text that already have been generated for the page? Is it like this? Yes, we saw the placeholder there. I can put anything in there. So yes, you could ask the agent, you know, based on this page. We didn't go into the page editor, but you know how I clicked on the page and it understood the page I was talking about. You go into the page, you can click on an area, on a component, and it understands that the context you're working in. So you could say, yes, based on this component, generate an image for me. That would be saved down into the dam. Then you can plug it in and you can use it in that component. Yes. Thank you. Yes. This may be a little bit more of a basic question, but the GEO properties that were in the demo there, are those available out of the box or is that something you have to custom configure? I know what you saw there is the default set. But all you would do is you would change the prompt if you said, no, actually, I don't want references or I do want this. We've been fiddling with this. We had FAQs. We were told that was bad practice. To take that out, all we did was change the prompt and then they would go on forever. In terms of the document to page capability, currently, is it only restricted to work? Let's say if we have embedded video on a page, for example, can we embed code in it, in the document? Or let's say if someone has an image that they would like to update on, add onto the page, could they support that? I think in the placeholder bit there, you can tell it to go and search for a piece of content to go in there. So, if you know that video, if you can describe that video, the agent will go and find it within your DAM. Because it can read all the descriptions. All that metadata is in the VectorDB as well. So, it would do its best to find that video. If it didn't, as you saw, it's all Magnolia content. So it's designed for the expectation the agent is not going to be infallible. You'll have to potentially go in and clean up after it. So that might have to be a cleanup task if it doesn't get an exact match. There's one more question in the back. Yes, maybe I have two questions. The first question is regarding the translation. We saw you asked the AI to translate from English to French. But is it also possible to translate directly in multiple languages? With the tool at the minute and the APIs that we use, we've kept it open to just do one at a time. Because we found services that are dedicated translation engines, they were only able to say from X to Y. We weren't able to give those multiple ones. We've tried to keep it open. So it has to run twice. But as you saw, we just let the agent run the tool twice. I didn't have to type anything twice. We can translate this into all of the languages that I support. Our demo supports about six at the minute. And the video is a little bit unwieldy. So I just picked out two that I wanted. Okay, thank you. And the other question is, before you publish, you do the step that you put in the JSON-LD. And is this action also included in the doc to page? That when you give the document that it's already doing the JSON -LD for you without thinking about to do the step before to publish? Yeah, that could be done. We can chain these things together. There's observers in Magnolia that would recognize, oh, you've made a new page. I'll generate that for you. I wanted to be able to show it as an action to show what's happening. But yes, you can automate everything you want. The tradeoff becomes, how frequently do you update those pages? Do you want to generate fresh JSON-LD and spend tokens? Every time you correct a typo, do you want to go and have to regenerate everything? But yes, everything can be automated or it can be added in so at creation we generate some metadata and then perhaps you run it again when you know you've made all of your changes to it. The benefit also of currently having it separate is we have one of our use cases is to look for old pages that might not have JSON-LD or might not be optimized for geo. So I can actually talk to the agent and let me show all my old pages that are lacking that information and then prompt for those pages to optimize it for geo. So, okay, that makes sense. Thank you very much. Awesome. So, yeah, as you've seen, Nora and the marketing team, we currently have it. It's in beta. Actually, you can also experiment with the Magnolia agent. It's in beta. And end of July, beginning of August, we're going to release it for public availability. Thank you very much. Thank you, Nora. Thank you, Chris. Thank you. Thank you.