Small group demo: Intro to Agent OS

All right. Well, again, welcome everyone. We got a really exciting topic today. If you didn't get a chance to check out our announcement at Content Con last month, like I said, we were in London and announced everything. If you didn't get a chance to check that out, we're going to go through a lot of those details today. I'm joined by the man who delivered that demo on stage, Tim, and my counterpart, Low. Hi everyone. Part of our technical market also delivered it but then holding hands with everybody you know on the first row. Definitely definitely everybody there. Yeah. All right. Well, we're going to go through we're going to give you a little bit of an intro and background on on Agent OS and kind of how we got here and then we're gonna kick it over to Tim. He's going to show us in app what Agent OS looks like, how it works, all that. We're gonna answer all of your questions. So, be sure and submit them. There's a place in the uh on screen to do so. We'll try and address them as uh as quickly as we can if it makes sense to do it in the moment, but we also have time at the end and we want to make sure you get get all your questions answered. So, let's dive in. I think we should just start by pointing out one thing. Like agents are really hot right now. It's a buzzword. You're seeing a lot of people come to market with their offerings. I think one thing to be clear about at content stack, we are not part of that crowd. We have been working on agents for years now, right? This started back in 2023 when we released automate. This was our no code workflow builder that allowed developers and business users to come in, create integrations or crossplatform events, things like that that would all play into their content operations. After that, we came out with an AI assistant that really brought generative AI into the authoring experience, allowing you to brainstorm or swap out, edit, make optimizations on text, things like that. Then we saw the opportunity to take that uh take those AI capabilities and and plug them into automation workflows. And so adding AI connectors and quickly realized one of the most important parts of this is that AI output can be very generic. And so making sure that those outputs are on brand is uh just highly critical. And so brand kit we released to keep control put some constraints and governance around how those out uh AI outputs um perform and how they're how they're how they're produced. And that that is how we arrived today. where we have agent OS we've had this foundation in place and allowing us to take all of the learnings that are required all the things that you have to uh overcome in order to uh to come up with a successful product that that delivers value and so what does an AI agent look like at content stack right this give you a quick just intro here but this really plays a role into how you build agents in content stack every agent needs a model and this is critical because at content stack we don't force you to use uh an out- of- the-box model. We do have one to make it easy for you. But if you want to bring your own model, your own credentials, you can do so. We provide it with instructions. We want to make sure that our agent has a goal. It knows exactly how to carry out and behave. Context here is critical. This is one of the most crucial parts of the a successful agentic experience because without context, you're just basically uh working in a a generic chat GPT type uh type environment. Giving it context of your content, your audience data and your brand allows like we talked about those outputs to be consistent and on brand and and then we need tools, right? An agent needs abilities. It needs to be able to carry out actions. And so we assign it a set of tools and actions that it can choose from to carry out uh tasks based on what its instructions or goals are. And again, we're going to go into this a little more. Tim's going to show you here as we build custom agents. it'll connect a little bit more. But we know that the other thing that's really important here, and I touched on this a little bit, but generic AI outputs uh are aren't usable. And in order to have a successful AI solution within the business, you need to be able to trust what's happening. It needs to be able to work at the enterprise level. And so again, having the context of all of your content, your audience, and your brand data uh supports that and the control that you're offered, all of the workflows, permissions, roles, uh the flexibility to bring your own models, all of that, you have control on what that experience looks like. And then there's consistency, right? we with uh with BrandKit and the ability to put those rules in place, you can make sure that AI outputs are producing at scale and consistent in the way that you want. So, Agent OS, what we're going to kind of dive into, it has kind of three main components that we're going to explore today. The first is Polaris, which is a conversational AI companion that's available throughout the platform. Can help you answer questions, carry out tasks, or offer insights. you there are custom agents that you can build through agent builder to build out and carry out very specific uh sets of tasks and then there's the digital concierge and this is something that you could publish to your website or app and is a conversational AI that can engage with website visitors or your users and again based on how you create or design that experience could answer questions um access account information personalize or tailor the website experience which we will explore a little more of right now. Have a little bit more details here, but I think what everyone wants to see is just the actual experience. We're just going to do it and then very clear. Yeah, exactly. All right. I'm going to pass it over to you, Tim. Awesome. So, what I think we should start with though, when we say all um you know about all that agents stuff, um this is not released yet, right? Most people here know about it. we have talked about it openly. Um, but it's going to get a lot more love than it even has had right now. So once you start using it, it is actually a lot more stable than it is now. And I'm not saying it's unstable, but we really need to make sure that this stuff works at scale and it's always like repeatable and, you know, deterministic and stuff, right? So for you guys, that's really important. So um, can I have my screen on this on on the screen? Oh, here we go. Nice. So what I want to talk to you about today is it's basically a similar demo that we did at Contentcon, but we have a little bit more time today. So we can talk a little bit more. I'm hoping low will interrupt me here and there to actually give some more context, but also ask your questions. We are here for your questions. We love the interaction and if a question actually leads us to a slightly different area, we might take it there if we feel like, hey, this works for the whole group. So feel free to influence us with your clever questions. Yes. And you know, I'll chime in, Tim. Yeah, exactly. We're like Lo and I have this thing where I just, you know, I'm a product person now, so I keep speaking and Lo just has to, you know, jump in somewhere and then make me chill. So the what we want to show here is that like we have this story that we created and this really also shows how you can use the data that you can collect based on how users use your website to actually look at that data and based on that create your content. So we're slightly flipping the script, right? where traditionally you would just know your audience somehow through some research or some analytics you already have and then you figure out I'm going to build a bunch of pages and see if this converts. So now what we want to do is flip that and say well there's a whole bunch of data that we have collected for you based on what all your visitors are doing and we can now give you through AI um you know basically hooks into saying hey your opportunity lies here why don't you make that content so it's more datadriven and that's what is really interesting here and so the the story of the demo is basically here we're so I want to go on a holiday actually can you can you switch back to my screen real quick as we just while we're setting the stage here, Tim, just before we jump into this, I think this could be helpful if we just put what I have on my screen up because we're going to we're basically going to show you an example a pretend customer. Oh, yeah. This fits. Yeah. So, let's let's maybe give the plot here, but let's we're going to show you as a a Red Panda Resort is a high-end luxury resort catering to premier vacation travelers. We're going to pretend we're part of the marketing team there and we're trying to cater towards that audience influencing them to purchase some kind of vacation package. Right. Right. So now let's pass it over to you and you are a person who's interested in traveling somewhere. Right. And so you're on a social feed and you're being targeted with an ad from Red Panda. So, this is part one of this demo where we're showing you me as the end user and how we can use our AI in different spots to make personalization work more and more. And the bigger the context is, the the smaller the audience gets and at one point it's one and one to one with me with an AI agent. So again, I'm looking to go to some really lovely far away place and I really love like water sports and adventurous things and I've saved up my money and I now see this Mel defe post here which is exciting me and they're saying hey there's actually a bunch of offers that we have. So, of course, I'm going to click on that link. And so, by clicking on this, I'm already showing a little bit of intent, right? So, there's a little bit of context already added by the fact that I am interested in looking at the packages of the Red Panda Resort. And so, we have this little debugging tool here on the page. This is just for us to show you in the demo. You wouldn't actually put this on your website, but this is actually very handy because our data and insights tracking tag JavaScript is on this page. And you can see, hey, I'm inbound from social because I have a UTM campaign in here, right? So, it put me already in this audience and I'm currently currently anonymous and I'm a standard customer. I I haven't shown any interest in potentially being someone who pays more or less or just someone who browses. But based on the page that we just saw here, there's already a bunch of information here about what's on that page. And so these interests and these are basically topics. And these topics are generated by the JavaScript just looking at the page and already figuring out, hey, if I now go to another page, what you'll see is that like in a few seconds, these numbers start to update a little bit. You see, suddenly there's more here. like water actually changed a little bit and there's a whole bunch of things that it's now looking at but the most interesting one here is that it actually says hey we have a lookalike model so because I just um Tim can you zoom in a little bit possibly I can better cool so what I did I just clicked on like a water sport reef reef snorkeling thing and so when I actually also go to sailing these are like pretty expensive ensive excursions, right? Just $2.99 a person for just a few hours. It's like really fancy stuff. And you just saw that based on what other people have been clicking on the website that actually ended up spending a bunch of money that became premier customers. I now start to look like them a little bit. And so we didn't exactly set this up ourselves. This is actually the AI in our data and insights tool figuring out, oh, you kind of look like a premier customer. And if that score gets high enough, we're just going to upsell you. And so I am now not exactly telling you an agent story just yet, but soon. And so right now, we're almost over that threshold where maybe this um when they feel like, okay, you're probably going to be a premier customer. So now I went to the fourth page, all in that same category. It's like, okay, we actually have exclusive packages. So well, just show me now. And so what now happened? I actually went back to that previous page. This is also the list page with all these packages, but this is a different variant. And now they selected a product for me. And this is already quite personalized to the context, right? Because I looked at like sailing and parasailing, those kind of water sporty things. However, I want to go for a week, not just four days. This like a Maldiv is far away from Europe, right? Yeah. That's like travel days if I spend the money. And so what you just saw pop up here is our concer and this is an agent OS agent. And so hey, it looks like you're interested in these adventure packages. And so like I want to go a week. And so this agent knows about your platform, your content stack platform. It knows about the products you have. It knows all this stuff. And so I just I want to go a week. and it actually figured out, oh, there's actually a 6 day all-inclusive thrillseker package. So, it just kind of kept that the same, but made it 6 days. So, that's already cool. But I really like diving. So, I can just say, can I dive? And of course, what you see here right now is somewhat curated because we have to be able to demo this really well. Um, but you can also make this slightly more open and um make it choose very specific products to show because right now this is kind of like like everything here is super personalized because it's now saying, "Okay, we have scuba diving, but you need a certification for potty open waters." Like, yeah, I actually have one because of course I've prepped for this. I already have one of those. And so now that it knows that I have one, we can actually now change out one of those. So, let's change that sailing. And so, this agent now is changing the personalization rules for me. And I just changed sailing here to scuba diving. Look at this. This is just magical. And where it gets even more cool in my opinion, when we now look at the data, I am now also a petty certified audience in the personalization. So this agent from content stack or now from you and you use this it actually can talk to all the different products in content stack and say well now put this person in this um audience but there's also a bunch of other stuff it could do and so right now this is where we leave it but what I can do is actually well let's just book this thing and so it actually personalized the three things for me but if Brandon or Low had done this and they had clicked on other things this package would have been different and we chose to do it that way you don't have to you have all control that you want. But so let me just add my email here. While you're doing that, like I think too one of the opportunities with like putting together packages like this, right, is it can be data informed. So if you have an adventure or thrillseeker package, you might want to prepopulate it with, you know, the events or activities that are most popular based on your data, right? So it's not just random packages or random events. And so the final thing I wanted to show now that I have paid, what actually happened is now I'm a full premier customer now. And it I'm of course my likelihood is 100%. But it also noticed I gave my email, right? So I now have a customer tab and you you can actually see how much money I spent doing this demo over the last few weeks. I'm spending an ungodly amount in this report. But it now an upgrade of some kind and yeah exactly please like but the interesting part here is that it now had a record already in Salesforce like we have a Salesforce integration in this demo and it just merged that with whatever that anonymous thing was and now you can see that this is a much richer like there's much more stuff and water sports is like a 100 obviously because I keep you know doing that stuff and so there's all this has now also changed because of the history that I already have here and so I know, we're going to have to go towards what all these agents are doing and how this works, but we wanted to show you that all this datadriven stuff is really helping you to um use your agents effectively. And so, right now, we went from I show a little bit of intent, but I'm completely anonymous to getting more context and more and more based on what I click. It was kind of a self-driving AI in our data and insights. that kind of at one point decided, okay, I'm just going to, you know, upsell you this thing. And we didn't actually save much in our cookies. You just have an ID and that's it. So for GDPR, this is fully safe. We have to say that. I think it's important. Um, and of course, you get to choose how much you do with that, but by default, our stuff doesn't actually scrape user data and does all the crazy things. You choose what you put into it. It's important to say. Yeah, there's a there's a question here that maybe we could just address that it's saying does this information push over to our personalization system like Salesforce or SAP. So I think what and we'll maybe show this a little bit but when we show how like custom agents are built those part of those abilities or tools that you can give it access to are your third party systems and and and tools across your ecosystem. So yes, the answer is yes there and based on how you want to customize or deliver this experience. Yeah, because we now chose to then merge that customer record that I have in Salesforce with whatever my anonymous thing was. And so now it's, you know, it's it's my I've spent 10,000 nights here apparently, which is kind of fun. But like it merged that, but we chose to send that and merge that you because um some companies who are maybe in France or whatever don't have the luxury of doing that. And so it's up to the people implementing how much data you share. Um so we went and I think like for governance in general, right? Like you want to have that level of control regardless of whether there's like a policy or something that you need to adhere to. It's important for governance in terms of your organization anyway. Exactly. Al all those things are slightly boring in an in agent demo. You have to say them because it's just really important for businesses. So, I'll say one thing here. So, we kind of went from anonymous to showing intent to getting more context and then going to a one-on-one personalization with this agent than to buying. And so, you start when personalization starts, it's kind of just, oh, it's interesting. It kind of works well, but at one point it just hits you and you're like, oh, that actually really knew what I wanted and I converted. Right? And of course, this is a demo, so of course I'm going to convert. But just put your thinking cap on and see how you can use this stuff yourself. So this was the part where we just experienced this thing as uh as a as an end user, right? But now let's think about how are we going to actually build an experience like this from the content st side, from the CMS angle, from you know the data and insights angle. And um what we tend to see is like before I talk about whatever this lovely graph is here in our story, let's go back to my screen. Yeah, I bet. Yeah, we have a couple more questions too. Yeah, this might be a good address. Let me answer the question. Um okay, so one of the questions here is how is the context fed into the LLM? specifically CDP and site content. Is there a way to feed other data sources like PDFs? Uh, and again, I think what you give an agent access or context of is up to you, but yes, you uh can customize what context is given. But things like you mentioned PDFs or other types of content, we'll go into a little bit more. That's all part of brand kit and things that you can put into your brand kit to give it the context. Uh that I love how this question spans like three of our products. So yes to all Mang and um what you just saw was anonymized data that our CDP actually figured out based on what people have been browsing like. Plus, when we started doing that chat with the LLM or with the agent, the context of whatever we were chatting was sent to that agent as an automation. And we'll talk about what that means in a sec. And that automation was able to kind of figure out what is the data I'm allowed to touch and how can I grab that and if what the tone of voice, all that stuff. It like it touches multiple products, but you have all the flexibility that you need. And so, yeah, I'm not sure it fully answers the question, but all that stuff works between our products for sure. Yes. All right. So, is there another one, Brandon, we have to talk about? There's another question. Um, can the data that is learning be something easily viewed and accessible to be shared with marketing? So, they can use that to identify potential campaigns or promotions to do an offer. Um, what can we see or This is literally on my screen right now. So, I'm going to show you this in a sec. Yep. This is because we realize that that is the crux of what you guys need. You don't need analytics. You don't need only data. You need actually a mix. Yeah. And so, I guess what we're trying to kind of going back, we showed you a little bit of what a customer experience might look like on a website, but now let's shift gears a little bit. Let's pretend and take the role assume the role of the red panda marketing team. They've always focused on relaxation topics related uh that they found that's always kind of geared towards the premier customers they target. But now they're looking for additional opportunities to to to market to that audience. And so what else besides just relaxation could they focus on either in content topics or product offerings or whatever that could be appealing to that type of audience? And so they are trying to figure out, yeah, what what does our audience care about, right? What what topics do they care about? And that's something we've always kind of had to either in guess or infer based on what tools we have. But let's show you how data and insights does it for you. Yeah. And so just like Brandon was saying, if we look at this is data here. Um these are essentially the topics that we saw in that little tool we just had on the page with all the bars, right? And so these are the topics of the pages that we have. And the bigger the circle, the more people actually were interested in that. And the higher the circle in the graph, the more content that was created about it. And so on the left side here, you can see standard customers are at this point in time some generally interested in culture, food, relaxation, serene stuff, stuff like that. And then Premier customers are actually much more interested in adventure, sun, luxury, hiking, all-inclusive. And so if we actually have a look at, let's say last year, we can actually render that here and have a see the change because we started this last year and you can see that we're kind of sitting in the middle. We don't have that much data just yet. And there's because this is big and it is high. We we just made a bunch of relaxation content like there's eight pages on it. But if you now actually look at what how that changed over a year, you can see the inertia of where things are going, right? So culture, food, relaxation went to the standard customers and adventure and luxury and hiking and all those things went to the right side. And so hey, hey Tim, what if uh what if like I'm not a data scientist and you're speaking a different language to me right now here Brandon. Exactly. You're saying a lot of things and they sound I know you're smart and you're speaking in a way that sounds like you're smart, but how do we how do we like relate to anyone here? How does anyone make sense of this? Yes. And this is where the magic starts to happen because I was about to say I just needed 25 sentences to explain you this silly graph here. But what if we just did this? Now, we just asked the Polaris agent to see just look at this graph for me and just give me the best opportunities that I could use this week where people are going that are essentially premier likelihood kind of customers. Where are they spending their, you know, browsing behavior? What are they looking at? And as it turns out, it's thrills. It's all inclusive. It's story ideas about hiking. And so the AI, if you look at this, it actually looks at an agent ability called a tool called opportunities, which is an automation. We'll talk about that in a sec. But for the technical people on the call, this is essentially the MCP protocol. And this agent figured out I have a tool for opportunities. So I'm just going to fire that and then come back with some information. And that tool you can scope to what information it's allowed to see. Obviously, we'll see that in a sec. So, we just had this non-deterministic interface that I just opened on this page and say, "Give me some help." Like, I don't know what to do. What is the best opportunity? And so, now that we know, let's do hiking in thrills, right? So, now I can do Hey, Jim. Real quick, there's a there's a question that we can maybe address right here. How are these keywords pulled in from the content? Does it come from all field types and content models including JSON, RT, and Texonomies? I think worth calling out here that you can uh these topics can be pulled from yes all of this if you want if you don't have anywhere to start you data and insights if you install the the the real-time events tag and start pulling this in it's all going to be uh inferred via via machine learning and AI and we'll get you started on those those topics if you don't have them but if you do need them to be customized to what you already have or matched to your taxonomies or whatever And you can uh set them up that way. Exactly. So Corey, we have in our data and insights, we have like a connection. So what you can do is any system you have, you can connect into it. And so you can have all your content stack, fields, models, URLs, whatever you have, put it into the CDP and it will just do it. But like what Brandon was saying, this one actually doesn't use that. This is just a JavaScript tag that looks at the theme of the page, uses machine learning, and then builds up your database. So you can do both. Yeah. But either way, you're giving context, right? And that's what you want because if you didn't have this data and you ask Polaris, you might get something else, right? But that's the power of this. You're using data to make informed decisions and now your agents are also using that same data to make informed decisions to give you a good output. So, what I'm now going to say is create me a page about the thrills of hiking in the Maldives and please use our hiking assets because I know we have some information already on hiking. And so, um, we went from we we basically can just ask Polaris kind of anything to do with what you can do in content stack as an editor. And so, what it's now doing, it's going to start reasoning. It's like, okay, so it I need assets, right? because I asked for that. So it now it looks at the asset agent and so this asset agent is now going to be kind of figuring out, hey, do I have hiking images? Sure, let me just have it black. And so now it's doing all these abilities. Look, I'm going to get in this case actually it doesn't really do what I was hoping it would do. It's now getting all the assets and filtering it later, which works, but sometimes it just finds hiking assets. And so that's the interesting part, right, about how these agents sometimes work. Um, so now it's f it most likely found images. And so now this orchestrator agent is going to then just open up another agent and says, "Hey, well, what is a page? Can I figure out what a page is? What are the fields that I have?" And so if we just let this run for a sec and take a step back, like what this is actually doing, it is kind of acting like your assistant somehow. I'm not sure we we generally say that word but it's helping you and it's kind of figuring out hey okay I need to make a page let me figure out if I have a content type for page let me just get that now and then find the fields create the and it already has the right images and it we are using brand kit here so it knows the type of writing images titles it knows exactly how we tend to do this on this project and so it's using all that and it's like it's even filling like SEO metadata and actually it just made the page and so now this is the big thing. Let me just click open in visual builder to see what this built for us. And maybe this is worth calling out that this is I think one of the key differences between like an automation and an agent is an automation you can give it a set of deterministic rules or paths to follow or even put some kind of like business logic in there to loop or conditions what whatever but what will happen as soon as it if it can't find what it needs or you just told it to create a page right you didn't give it a content type or anything like that in automations usually those types of things will fail because they'll run into some kind of issue. or we'll have some kind of error or something like that that doesn't produce a result but agents can reason right we have given it a set of abilities a set of other agents that we have given it access to that it can go and work with to find the best solution and I that's I think one of the biggest things to to to call out here is it's doing stuff that we may not know what it uh needs to do but to carry out accomplish interestingly it did it differently than I expected but then look it actually made a page But to be fair, this is not what I expected. It's only the hero. So, um, sometimes there's a hiccup. That's the first. This is one of the first times we have seen that because also when you looked at the reasoning, it was starting with looking at images rather than asking, hey, what is a page? Page, right? So, it might have done a different approach. Um, no, it doesn't actually matter now. I I could do it again, but it takes a little bit too much time. Um, why don't we show what al think I was just going to say I also think it's important like in a situation like this this is why you would never leave an agent up to its own devices right and then just publish this like it's not meant to be a replacement is meant to be a helper to make you move faster but you always want to go in and check cuz something like this can happen or even if it made the full page right you would still want to go in and look through that and make sure it's okay exactly and so Um, I think it's also worth calling out all the steps that we kind of took leading up to where we are right now. If you ever wanted to put new content out, maybe we just quickly illustrate what I'm reading it while you speak. I just want to redo it. You you speak. Yeah, that's a great idea. But I think what if you're trying to figure out one, you got to figure out what type of topics or content you want to talk about or create. Then there's all of the ideating and brainstorming that maybe takes place somewhere outside of the platform, requires other people. Uh maybe there's approvals, things like that, all these things that happen. I I know there's lots of lots of customers I've spoken with who it takes weeks to get a piece of content up and just given all the all the red tape and everything like that that that that's involved. But we just went from nothing to a page that's ready to be published. And maybe it's not at 100%. And we're going to try and get it a little closer this time. But if we can get that get it there 90% then getting to publish stage taking having that take place that much faster is uh just where we see some of the huge advantages here in efficiency gains. Yeah, it's a huge timesaver. And it's also I think like worth calling out that we didn't just guess. We're not just saying hey we think this might be interesting to write about, right? It's it's data informed, right? It's backed up by what we what our audience is actually engaging with and cares about. So, um you feel a little uh you feel better and more confident about how you're going to market with that kind of stuff. You know what's interesting, Brendan? It replies differently now. We I either open AI just changed its model for different reasoning or we did. I'm going to check that in a sec because this is really interesting. Did you ask the exact same question as first time? Like I'm just going to do exactly the same question here. Okay. And the blessing of and the blessing and curse of uh being subject to other LLMs is that you know this is a great match actually great bridge here Brandon because we now chose to just use this is I think open AI we can also choose entropic and whatever but if you have your own agents or your own credentials maybe you're on AWS Bedrock or you're on Google you can just add in your own credentials and use your own agents like that's totally possible. So while this is running because now it's going to the entry agent first. So I'm I'm thinking this is good. Um these agents you can kind of customize them. And if we look at can like we can see here. Okay. So our the custom agent I'm going to open now actually uses OpenAI but you can of course um you can actually say here let's go to the next one. You can actually choose a bunch of different ones here right and so we are actually adding a even more here. Um, so these agents are right now this is a si a very simple custom agent that doesn't do too much right now, but this is just for you to see how you can set it up. To be fair, this interface is going to be a little bit easier once we actually launch it because you can do so much. It's almost overwhelming, but it also gives you the freedom to do a lot of cool things. So you can give it some instructions. Then yeah, this is where we would put like the goal or what its behavior is and then Exactly. Yeah, the abilities here. Yeah, this is where the magic happens because a lot of times agents if you just go to like a generic like cloth desktop or whatever it's kind of non-deterministic, right? And you kind of have to hope that the model understands you and give you something deterministic every time, which is really hard. These tools make you much more able to have a deterministic outcome of what you want, right? And so we have all these different abilities that you can give it like you can give it like an automation. So in content like automate you can drag and drop and build your own automations with your own inputs and outputs that are always the same. You can add code, web hooks, anything you want like you know send something to Slack or you know update a database somewhere. Whatever you want to do you can do in automation. So an automation can be a tool for an agent. Um, we also have sub agents like an entry agent that has a whole bunch of baked in stuff that it knows about entries. But you can also see we have actions and actions are basically all the action steps that our automation product offers. And so this will be an ever growing list of stuff, right? You can create releases, you can send messages to Slack, you can update entries, create, you know, Asana tasks and like this is basically never ending awesomeness of fun things you can do. And you know, you can make notion pages or like whatever you want. I think it's just worth calling out or repeating here that if you haven't been using automations, automate and using automations, they're not going away, right? An automate, there's certainly a use case for very deterministic workflows. And if you want that to be a part of an agent's abilities, then that's what Tim's explaining here is that you you could have a very strict automation workflow that you assign as an ability that an agent could carry out. But um and I think that's one of the key things worth mentioning here. Yeah, we have a few like we got some questions. Yeah, we got some questions here. Sure. Yeah, let's go page done creating. Yeah. What is an example of how this could be used in a non-marketing site like government or education? Um one example I'm thinking of is you could have an agent. Let's say you have a government or educational site and you have policies or lessons or courses, right? You could create an agent that always suggests related courses or related lessons or related policies. Um like also if you have something like an internet that is for internal use. Um you can also use it as a page builder. If you have set layouts, you can have it create new layouts based on components that you already have. So there's a wide range of use cases um not just for a marketing site or even a e-commerce site, right? Yeah. And you can even just have it look at maybe um like let it run every night to see if there's any violations to your rules or if there's spelling mistakes or you know it can do anything an agent can or a human almost right as long as you configure it right. And so um this is part of that configuration part like there's this brand kit integration here and I wanted to highlight that before we look at the page we just created. So uh on this one in this project I don't have a brand kit but this is an example of one of there's a question that's asking if you can have multiple brand kits and so yes if you if many you can see right here you can select from the different brand kits you had available but yes multiple brands or those who have different brands under an umbrella or something like that can all have different different brand you can also make one brand kit with like 10 voices right like the voices can you know change up so We can go there quickly because we I feel like we don't show this enough. Um let's see if in this project I have more. So we have a red panda brand kit actually and see there's a bunch of voices here. So let's do the fun family adventure like this voice emphasizes excitement togetherness playful bl this kind of stuff right and you can actually say casual persuasive light-hearted but then also you can add insights and sample content but you can see this is a lot of lines. You can actually have 80,000 lines here of information or maybe this is characters or words. Anyways, it's a lot. And so in my personal brand kit voices, I fill this up and I say, "Yeah, use the stop words that I use when I write, right? Be cheerful, but also be a little bit spicy because that's how I write on LinkedIn, right? I can put all that in here. I can say keep it short. Keep it only to two." Okay, so Christine, it's characters. Thanks for that. And so there's lots here. And then we also have a knowledge fold where you can basically add any sort of data. And so it uses the data in a knowledge fold to inform itself about how it can speak. If you put your products in here, it can actually talk about those products in what you generate. So this is a very important part of what you do in these um what you set up in these agents. And well now that we tried it the second time I really wish this time yes let's see but brand kit in general is how you provide your context layer right and it also helps you with governance but it does lots of things in terms of your brand voice and knowing and you can input about your organization along a context sort of hierarchy right so not just the organizational context but who you might be talking to such as personas but who you are right you're asking this question as xyz role you know so I might ask it and say well I'm a technical marketer think about me in this role as you provide me with an answer exactly so let's have a look at this page this page looks a little yeah it's a lot it actually understood the data model and it understood we had all these components and it used brandkit to write this copy also And so all this stuff went through the brand kit to figure out what can I write? What images do we have? So it figured out all the right images, right? Family adventure stuff or hiking as a couple. You see that it it did copy up a few images. It reused them. But this is also not something that actually replaces you. It gives you a base. And so and yeah, you might need to swap like like local wildlife, right? That looks more like a that image probably fits more with the couple the the couple eggs or whatever. But you can here, right? Right. So, yeah. So, this kind of looks nicer now. So, how about we So, let's publish this and then come back because I do want to show um the personalization stuff because this Well, we do need to We have a We do need We have quite a few questions we want to make sure and get to as well. So, want to do questions first? I'm all good. Yeah, let's do questions. Let's Yeah, sure. If I build a workflow in Automate, can that be exposed through MCP tools? Uh, do you want to just touch on the MCP stuff real quick, Tim? Oh, I know Patrick has this question. That's a I know you were going to some have some sort of question like this. So, right now, our content stack MCP server that you can download from npm does not have that ability. Um, but it will in the future. So Christine and I are fully on board and we really want that. We want the MCP that you install, let's say in your local desktop cloud or in Jet GPT to be able to also get these automations and be super integrated with agent OS. What you should know is that agent OS itself, how agents actually work is all MCP protocol and it's all like the tools that they use, the way you set up the tools yourself, that's all MCP protocol. That doesn't mean that the MCP product that we created is the same as this. It just uses the same stuff. I hope that makes sense. But Patrick, we we're due for a call anyways where I talk about this a bit more deeper. Yeah. Um so next question is says um from Eric, this is a great example of building a new page from data. Let's say we want to tweak messaging. Is it able to analyze pages across the sites and make broad updates on the messaging? Yes, I cannot show it now, but yes, very simple question. I was like, this is going to be a very short answer, but that is an incredible use case for a custom agent. You can make a custom agent that runs every night, for example, that just looks at certain amounts of content based on data and maybe it looks at like some external tool for anim for, you know, SEO or whatever and it can actually do updates and present those to you in the next morning. Or you can actually say, I don't think I have anything installed here, but we could actually have like an automation here that you just click a button and like, oh yeah, there's a few things that Christine, I know you're in the chat that you actually added here. I won't do it now, but you can make your own anima automations. When you hit a button here, it will fire up one of those automations and then it can grab all the pages and give you options and change messaging and like update it here and say, "Hey, are you happy with this?" Yes. Let's go. Like there's lots of ways into how to do that. Um there's another way also that's actually quite interesting. If we wanted to is I can actually grab the AI assistant here and have brand kit in here and actually say well let's grab the Brad Panda brand kit and optimize for SEO for example. Like I can just click this now and let's see what comes out. It didn't do that much, but it's now explore the secret hiking paths of the meld. It's slightly like snappier. And so this fit our brand kit basically, but I'm sure there are more questions. I think this relates to Paige your question as well about agents learning or or adjusting to to what you do is certainly if you're building into a custom agent and based on you can either create a very granular or specific way that it's adjusting like Tim kind of just mentioned where it's looking for things or based on just its behavior it's learning from what it's engaging with and what it has context of. If it learns, we'd have to ask Christine for a bit more details. But what I tend to do is not really rely on it learning. I would rely on just adding a bit more instructions in its context window because then you control it much more. Right? So Paige um once you start using this um just experiment with the the you know the context window of where you can say you do this but not this and then like output that but not that because I'm sure if you want to do this right you're going to have a lot of instructions because you own your brand you know that stuff right and so um over time these agents will get a lot better at figuring out your all your pages and just making a similar one right I've had it where and Patrick you're like this one where I made a proof of concept MCP on my local machine that actually when I say make a page it actually looks at all the other pages and figures out the data model how they are used and it just makes me a page that's similar to the homepage right so but I instructed it that and that's a really interesting one and so page sorry that it's not a super direct answer but it's so open right now and things are changing so fast um maybe the best answer is this is how it is now imagine how it is in two months. Maybe that's the best answer. Um, so we have some more questions. Corey asks, "Is this going to work with highly flexible pages? For instance, pages that use modular blocks with a lot of different block options each with their own options on configuration. That's what this is literally like. Look at this. This one, this is a module block for a hero. It has a headline and image details. It has you can go full image or you can go half image. We have text color, dark or light. We have layouts. So this one is like this is a decent amount of how many options I'd give to a block. Like this is rich text. Like this one doesn't have too much. It's like two columns, title, body. Um well this one is I'm sure this one is slightly more comp complex. Let's say this is an oh this is just an images block with sub images. So this one is not too crazy but as you can see these are those modular blocks and we having access to that schema how far you can push that limit to be honest we we have to figure that out we know there are some very complicated modular oh yes block instancations yeah but it also works with references and stuff like I've have some personal projects where I also say just add me five of the latest articles and it will actually find the reference and put them in so it can do all of those things. Uh there's a question that's asking if Claris will work across stacks or if it's stack agnostic. uh at launch it will just relate to one stack um as far as what beyond that uh could probably expand that but no not in scope uh for kind of immediate release but as far as I know it now and this is why this is still not fully finalized right of course knows more but the idea is that once this is actually live in a few months when you open this in this stack on this page it has the context of this thing so I can ask it something about it if I now go to another stack and I open Polaris, it has context about that stack, that stack. So, it's context based of where you are. And so, that's why it also um that really helps you to be able to say just change the image on this page on the hero and it will have that context and is able to do it. Interestingly, it already has the context. When I say do it, it understands, but it we didn't build in the tools yet to actually then change it right now while on the page. But that's something that's coming. Christine, please hit me up in the chat if I'm saying things that are not true, but as far as I'm aware, this is still good. Um, and then Grace asked, "How do I get the chart page with Polaris? I don't see Polaris on my on my content stack." So, I think Grace, we're the chart page you're talking about is audience um audience insights and that's through the data and insights um functionality on the platform. So you would need to um install that and then also add the real event tag so you could collect data and once you're collecting data then you would get um that chart populated but it's separate from Polaris. Polaris just uses that information um and helps you synthesize it and Polaris is a part of agent OS which will be coming out in a few months. Yeah. to we have to update your plan for you. Talk to your closest TSM and we will help you. But this is a great segue into what we think is probably the best action step you can take from this call to get ready for agent OS and and when it goes to GA and becomes available is in order for it to have the context of your audience data and and engagement. Installing the audience insights app right now, which is free and available to all CMS customers, is a great way to start gathering those insights so that when you do decide to start taking advantage of that and when agent OS comes out, you already have kind of a builtup uh repository of engagements from from the site over a few months, right? Which is what you want, right? Because you wouldn't really action on just a couple weeks of data, right? You probably want to have at least a month if not a little bit more to start making decision. Minimum six weeks low weeks. Yeah. And so I would go two months to three months minimum like the more data you have the more the more clever it gets and the more the CDP can do. Yeah. And so for this demo, we kind of just we actually made our own data. So we can really show you this demo always the same, but like you will just get automatically filled up. And we'll and I one thing that we'll follow up with in our in with the recording is some instructions on how to set up the audience insights app uh for those who haven't done so yet because again if you want to take one action or one one thing away from today of how to prepare or put yourself in the best place to start using agent OS when it becomes available set up your uh your insights your audience insights app and and JavaScript tag on your site start collecting those those insights and then you can start asking Once agent OS comes out, you can start engaging with Polaris, asking it all kinds of questions to uh to get specific answers context specific to your your audience. I'd also say if you have the opportunity now to set up brand kit and if you don't have that yet, we talk to your CSM to figure out if it's future and if it does because this is also something that takes some time, right? How do I figure out what my voice is like and like low you know a bit more about that. So maybe you want to take them. Yeah. And just, you know, it takes some time to compile all of these assets and things. So you can start having those conversations now because if it's not you, someone in your organization probably has a lot of this information already written down and documented somewhere as far as your brand guidelines or the writing tone, the video tone, the tone in which your organization overall likes to present itself. And also things like what shouldn't we say or what kind of terminology we should use. So you can start compiling that information now as well. Um I think because we're we're nearing the end here, Erin has a really good question. Can you put that on Brendan? Yeah, I think this is what you showed in the initial demo was showing a threshold where we would offer that that vacation package until you passed a certain threshold of what of that lookalike model to show that you you look or resemble like a premier customer. So same going into any kind of context. This one I think this is more on the LLM level. Yeah, I think and we actually we actually internally set a few of those settings already. Um, but we're finding a way like are we going to just open this up for everyone to set like these very specific model LLM instructions you can set based on this or that we just do it for you because some of the products we will will just we will fix what the model is but others of course you can get to choose and so there will be a whole bunch of levers that you can start changing to make this work well for your organization like we haven't I don't have the ability to show it now because we currently do just do it internally. And uh Scott, I know you asked about uh some are is there going to be any additional cost or pricing? This is something we're still trying to actively uh figure out, but likely and also dependent upon if you're bringing your own your own LLM, your own credentials, but likely be a a credit based system uh for building out custom agents. But more details to come there on on some of the other the other experiences within agent OS. But also the pricing stuff will be around building out custom agents. This app though it it comes from the marketplace but it's currently hidden. So if it's turned on your organization that you get the data and insights and the connection stuff, then this app will just show up. So it's it's not exactly something you find and have to turn on. We will just do that for you. Yes. And if you need more information about that, you can reach out to your CSM or us and we will make sure you get those resources of how to set that up. But again, we'll include it in our in our follow-up materials along with the recording. And uh and so yes, that's kind of into the last question. Will there be info shared about what's required setup wise? And so look out over the next couple of months on different steps you can take. We'll be we'll be reaching out with different different instructions. Like we said, start with the data and insights app, the audience insights app. Get data and insights set up. That's what we think is the most critical thing to to start with. And then we'll be uh in touch over the next few weeks over uh things you can do now to to put yourself in the best position to succeed. This feels like a pretty pretty good place to wrap up. Yeah, it's a lot of information today. If you have other questions, feel free to reach out to us. Reach out to your CSM and we will see you next time. Awesome. See everyone. Thanks for coming. Oh yeah. Thanks everyone. Mention that or you cannot really try it now, but we have if you don't have a CSM because you're not our customer yet, you can also do a free trial or try our explorer accounts. Very true. Get a little bit of a taste. Sign up for free. All right. Thank you everyone. Take care. We'll see you next time. Cheers. Bye.