Name: Document agents use cases in action
Uploaded: 2025-09-23T16:13:11.521Z
Duration: 57 min 8 s
Description: Document agents use cases in action

Transcript for "Document agents use cases in action": Hello, everyone, and welcome back to Templafy Talks. I'm your host, as always, Jon Teney Acos, senior principal customer success manager at Templafy. Yes. There are a lot of words in there. This is part two of our document agents one zero one series. Don't worry if you missed the first one. You can go on our website and find all the information and, follow-up and watch that again. I'm joined once again by our cofounder, Christian Lund, a man whose voice is so nice that when Templafy is finished, he can get a job on Calm. Now as always, we're using Goldcast. If you want to leave questions or, have any comments, please write them in the chat on the side. My colleague, Casper, hello, Casper, is gonna be monitoring the chat. We're coming to you live from Copenhagen. Please, so we can test it, tell us where you're coming to now. Awesome. So last time on Telefy Talks, we discussed a little bit about the state of AI in the industry. We gave an overview of how our document agents work and how they fit into existing workflows. Before we get into our second episode on document agents, I do wanna take a moment to give you some Templafy news. There's been a couple of big developments at Templafy, and I think it's worth sharing here as well. If you subscribe to our newsletter, this won't be news to you. But, firstly, content insights. So last year, Templafy made it, available for you as admins and owners to get insights about what content is being used inside of the admin center. This was, the use this was, sorry, because of feedback from you, and so we've made this possible. But you're never happy. Of course, you're not. One thing that you've requested is you want to see user data. Now if you opt in and if you to do that, contact your customer success manager. But you opt in and you can actually have access to, user data, and we're calling it self-service user data. Very clever name. But, yes, as I mentioned, if you want to opt into that, contact your customer success manager, and they'll take you through the process. Super simple. Secondly, Templafy is getting a face lift. Now one common piece of feedback we've got from users is that the task pane can get in the way. Well, you spoke. We listened. On October 10, you'll be upgraded to a new collapsible task pane, which, as you can see, pops open, and then you can easily close it upon request, giving you a lot more space to work. Not only is this task space less intrusive, it also gives the ability for users to pin their favorite assets, bringing the content that they need most closer than ever before. So all of the same access in a nice compact package. So for more information on content insights and the new collapsible task pane, you can see the, articles in the chat now. Awesome. That's enough from the news. I can get rid of my papers. I don't need those anymore. So document agents. So last time we talked about document agents and their place. What we didn't really talk about is do we still need the documents if we've got agents to do it for us? Yes. It's a good question. I'll start by saying, thank you for hopping on and thanks for all the rejoiners and thanks for all the engagements we had on the on the last webinar as well. It's always very useful. And, last time we did discuss what is actually the role of AI AI in the context of document generation, and we concluded, that obviously that is huge. And, and that's also why we've been investing so heavily in getting closer to what we can actually do with it and to get to accurate accurate outputs and to help users do stuff. But one thing you're asking is is are documents even important? And, I think a good way to look at it is that documents, the word documents stem from, documentation. So it's the meaning of documents is to document, work and specifically knowledge work. That's kind of the way that knowledge workers are doing that. They don't have the advantages of architects and others so that can document their works and buildings and stuff. It needs to go into documents. So documents are essentially a vehicle for information, to be handed over and to oftentimes drive value. And, if you look at the numbers, they're pretty staggering in terms of how many are actually being produced. And the numbers we're looking at now are not even including emails. So it's it's it's it's in the hundreds of billions that are being produced every day, so it's pretty massive. And if we take a look at, what that means for individuals, it's also, especially for knowledge workers, like, pretty extensive how much work is actually going into business documents. So we don't think they're gonna go away soon because they definitely play a big role. But we do think there's a ton of opportunity to take this, have your work day, out of what is now being spent on producing content and including documents to do something else, typically what you're hired for. And that's essentially what we're trying to do now with, with the document agents. So, they're definitely here to stay, and there's definitely a lot of work to do to get the to get the work out of them, so to speak. Yeah. Awesome. Thank you. So I guess, on that point then, so we've talked about you can see on the screen, we just imposed there twenty eight hours a week. Yeah. That's obviously a lot of time. Time is money. How are document agents gonna start to sort of shrink that and give users a bit of time back? What we're looking to achieve? Yeah. It's to answer that question and also to get closer to the specific role that AI and document agents are actually gonna play, I think a good way to look at it is actually this slide. Because, when there is so much work going into it and when we have consultants and knowledge workers just spending tons of time, creating business documents, it's important to look at what they're actually doing and what can be done to support it. And, actually, what what is interesting is that, above 85% of all enterprise documents created are actually bespoke. And, that means it's been difficult to have technology supported because the way technologies have been working up until now has been either, if you look at the the chart that we have here on the slide, either on one end of the spectrum where you have the templatized world, where you have, like, templates themselves or prepopulated content that is available to users for users to reuse, which is definitely helpful. And services are out there, including some that we come with that allow you to distribute content. It's very helpful. On the other end of the spectrum, that's why you have the repeatable documents where you can actually put them on formula, and that's why they qualify for classic rules based automation, where you can put in structures and rules that allows you to, to build them out on the back of certain information you capture from the user. But it requires them to be very structured. You need to know very much about your use case, and you have to put in also a lot of work Yeah. To actually get to a point where you can do that. And then you have the interesting part in the middle, which is everything that you can't really put formula on, or you don't necessarily have, like, a clean use case, or it's just bespoke because it's new every time, which is very common for con consultants, for example. The work they do is very bespoke, And that has been a very underserved area from technology because you can't really put it on formula. So the the abilities that we have until now on the two sides of the of the curve here are just running short when it gets to those, those areas. It's not that you don't have tools in the middle that can support. It's just limited. This way you run into the problem where users will reuse old content because it fits the intent that they need this time. And that's where, obviously, the risk content. So as you were. Sorry. Just thought that was an important point because, you know, this is sort of sort of the trends that we see. Yeah. I mean, it's it's the way that that a lot of users would work. Like, I would imagine I can't ask the question out loud here for for the audience, or I can't, but I can't see the answers directly where I sit. But I bet that the majority of the people in this, digital room have been doing something similar to this. I did, like, a deck last week. I now need to do something pretty similar. I don't wanna start from scratch. So you pull that one out and you optimize. You customize it to what you now need to do. But it's in it's a it's a bespoke event. Mhmm. Or you just start from scratch with something. So you might have a template, but it to getting from a template to an actual finalized output, there there are definitely some steps. And what that requires is essentially, you know, cognitive ability, mostly. Unless you can put it on formula and it's very repeatable like the example I'll show later as well on the proposals where you need a high accuracy. It's almost the same document every time. It's just the information that goes into it is slightly different depending on where you're going to. When it falls into this bespoke area, you really need brainpower. Yeah. And that's, of course, where those other technologies have been limited, but that's where we have in the word artificial intelligence that can then, come in and support the other AI, which is actual intelligence, the human. So it's interesting that you say that because one of the things that, I know that a lot of users do, and I would never do this, is if you don't have the correct template or you don't have the right, you know, sort of best practice slide, you're gonna use a document that worked for that purpose last week and maybe take a slide or two, repurpose it, change, you know, a couple of bit of text in there, and then reuse it that way because you don't have the specific, need served by your templates, for example. So this this kind of where the documents fit in that bell curve is, you know, I can relate to that from my personal experience and also from the experience of our customers as well. Yeah. I I don't actually, I don't trust you when you say you don't you don't do that. I should start saying that everyone in this digital room probably have the same type of habits, and it's for good reason. That you look for something that worked and you repurpose it. Yeah. And a lot of things just end in this bespoke world because as you might be helped, like, by with templates or even some of the more, very structured documents, you know, a lot of the work sits on you. And the problem is that it requires or the reason rather is that it sits there is that it requires cognitive ability, where the other the other sides of the spectrum here kind of fall short. And that's, of course, as well where AI and specifically agents for document creation comes in very strong because what you're ultimately trying to replicate is how people think. And that's what we really that's what really is needed to get close to that, area of bespoke documents. So it's more collaboration between the two AIs. Mhmm. One being the artificial intelligence and the other being the actual intelligence from the human. And the more they can work together, it's it's a it's a that's that's kind of a good way to get faster through this process with higher accuracy outputs and less risk, actually. Yeah. So really what is new here and also what we'll be discussing today is that it was a job for people. That was the only way to solve it, to put it very simple at least. But now, there is actually very, a lot of opportunity to put just with AI, and that is also what we have been gonna be putting out the majority of the focus on today. In that specific, mid part of the of the the bell curve, what is now possible and how you need to look at it to be successful around it. Fantastic. Awesome. So should we take a look? Yeah. There's a couple of things that, we wanted to make sure that this is clear for people on how to how to actually start using agents. And, at the end of this, there is, definitely demonstrations on how we do it at Templafy. But as well, I think you should look at this as a general approach to this. What are the questions you would need to ask any vendor that you wanna enter in in into with, if you wanna get in this direction? So we're happy to help, but I'm sure others are as well. So, the first thing I would be looking at is what agents should actually be doing. And I mentioned a little bit before, but at the end of the day, it's the ability for businesses to now get to a point where best practice is worst case. One enormous problem that exists across any business I ever met is that not everyone is equally equipped to build out business documents. You have different skill sets, obviously, different types of expertise. And even if you're able to identify the person who's best at it, it's difficult to to replicate or scale because there's a capacity limitation. Good news on agents, and one thing they're really good at is that once you you you identify best practices, you can scale from there. So you're never worse off than best practice. That's the good news. Now what they actually do in this mix of that actually hasn't changed. The the process of building a document is essentially the same. You have an intent. You need this document as a vehicle to get to an output. But what goes on with the agents in the middle is that they do a lot of the stuff that otherwise is relying on a person to do, and you can't necessarily replicate and putting into it like a classic automation with rules kind of structure. So what they would do is on the back of your intent is to understand what you're trying to do and then build out the best structure for you, meaning the template, but also the structure of the document going in. It would connect you to the best models, the best data sources that informs you essentially the knowledge that is required for you to build out whatever you're trying to do. Then, of course, take care of the compilation and ultimately perfect the document so it also looks good and is presentable. Mhmm. That's essentially in a very sort of simplified form what they do. But, a very important piece, once you start looking into using agents more for this work, is to be very aware, like always, on what you're trying to achieve. Because there's a difference between I'll take the example from the real world again because that's what we're trying to replicate. If I wanted something done, there's a difference between me going to the first person I meet at the coffee counter, giving them a little bit of information about what I want, and then expecting, like, a massively good output versus creating a a pretty good brief and then going to a specific person that has a lot of expertise in that area Yeah. And then ask that person to build out the outcome of potentially a group of people helping me to do that. So and and depending on what you're trying to do, both things can actually be pretty good. Mhmm. There is definitely a path where the general approach of throw anything at me, I'll give you a good answer, is definitely useful. And there are other scenarios where you actually need more expertise if you are to use it for anything. And that's what we'll be diving a little bit into and also leading into the demos at the end, so we are sure to put some context around what that means. So I'll start at the bottom. Like scenario for many customers today, all actually available as a blank template. And if you're gonna finalize anything from that, you're you're on your own from there. Mhmm. I'm not saying that the template itself cannot have quality, because you might have structure, branding, legal, disclaimers and stuff that could sit within such a template. So it's not nothing. But a lot of the work of actually completing the document would sit on the user. Then moving on to the first tier of what document agents should do, that would be the generalist. So the best general person you have in your company, if you were to take it outside of technology, just in a digital format. Throw anything at me, and I'm pretty capable of answering things to a certain extent. Those are great, and we'll give you some examples of, for example, how to build out training material on what these what these folks can do. It's it's definitely pretty impressive, but they're all limitations because they need to know everything about everything. They can work on the back of a template, and they have access to a ton of knowledge, and they you can set them up to use specific models and those types of things that puts in a lot of security around them, but there will be limitations to how final you can get Yeah. Just on the back of of, you know, how capable is it possible for them to be without more dedicated expertise. Completely similar to what you would see if you ask real people Yeah. About the same thing. So there is a job for the employee to do something here. Looked like you had a a question. So it's that that bespoke content. So I don't have a template for specific intent. This will get me 30% of the way there. And the things that, you know, that you would usually risk by using old content, like, you know, misusing old, branding maybe, images, customer information, all that risk is gonna be taken care of or some of that risk is gonna be taken care of here. Yeah. The way you should look at this is actually the comparison between these two is like, are you better off if you're I well, I'll put it in another way. I've never met anyone that is done when they just have the blank template. Right. It's it's at best a starting point, but you're definitely not done. So it's a matter of how done can you be. And given it an intent about what you're trying to do, these agents can help you take that further. Not all the way. That would be just a stretch. We don't wanna promise anyone that that's the magic. Mhmm. We've seen that from some of the market, but though that's actually the promise, We don't think that's correct. We think they're gonna be able to help you to a certain extent, but if you wanna go further, there are definitely ways that you could do. So once you get closer to actually understanding what your specific use case is, where it's less like, you know, I need something that can help me do something to get to a good starting point to, actually, now I wanna do a pitch deck, which is one of the examples I'll show in a second. Awesome. Then you know what you're trying to do, and you also have better understanding of what good looks like underneath that, which means that you can then start setting up, like, an army of agents that are actually very capable of doing specific things. Yeah. So that could be, for example, understanding very, deeply what it means to do a pitch deck. What is the structure that needs to go into it, and what is the content that must be there. And the ability to collaborate with other agents that are good at certain parts of building a pitch deck, which could be bio slides, pricing slides, other stuff that sit there. That is something that quite requires more expertise and potentially also access to other data sources. And it sort of puts you in the in the in the place where you get more deliberate on the output. So as you start being more specific on the use cases, this is where this makes sense. Yeah. Good news, that we've already seen from our own experiences is that as people start doing some of the work on the more generalist approach to it, that also captures information about what are then actually the use cases. Because on the back of what you're actually asking, you can see what you might wanna upgrade to more specific, types of processes for to help the users, to get there. So once you get into this, that's when you get into higher level of completion with higher level of accuracy, lower level of risk, and also sometimes something that can really impact revenue, for example. So a very strong step up and you sort of take down the amount of work that the users would need to do to finalize it. I really like this example, in both of these cases because going back to that bell curve slide when you said it used to be a job for people, you can really easily, tell from the descriptions that you're making is that, you know, as a user, if I wanted to make a pitch deck, I'd know exactly who to go to. I'd need to find that information, like, you know, manually or online, however, but there is that capacity issue. So instead of me having that intent, knowing who to go to, and then creating my document, the artificial intelligence, the agents are doing that for us. Yes. That's really cool. That that's the point. Again, like, I've put it out before, best practice is worst case because now you're always speaking to somebody who's really capable of what they're doing, but they don't run out of capacity. So they can continue to do the work all the time. So I'm not saying you're you're necessarily a 100% done. It also depends on what type of information, how you feed the agents, with the information that is required, completely similar to a normal person. And, but you will definitely get further down that road as you start doing that. The required capability underneath is, of course, that you go from having, like, a general agent that requires nothing to all of a sudden having, the ability to set up, like, as I as I said, like, an army or workforce of agents, and you need somewhere to structure those. So one good question to to ask, speaking to vendors about this, how is that structure supported? Yeah. Is there a way for us to control what these new digital coworkers are actually doing, which models they connect to, which data they source from, all that type of thing? Because this really sits in the realm of helping the user, yes, but with a certain level of control from the businesses as well, not the least in enterprise. Well, you need to know enterprises. Obviously, risk is the the number one concern when it comes to AI. And if they don't know the sources they're going to, the last thing you want is a business document to start hallucinating, not coming from a trusted source. So to to your point, you know, if they're they're exploring other vendors that do this, they need to really, really dive down on exactly how, what works behind the scenes because I think you make a really good point there. Mhmm. So you mentioned around the universal agents and the the custom agents and how they work. You did at the start mention that side that kind of those really hyper focused templatized AI driven workflows. Yeah. How do they fit in here? Well, I would say that's the last step, actually. So it's a good segue. So once you get super hyper specific on very dedicated use cases, the example that I brought, as I mentioned, is a proposal. That is where you actually potentially wanna narrow down the areas where AI even do work. Mhmm. So it's another type type of of logic you put to this because throughout those documents, you sometimes have no notes where that doesn't need any type of elaboration or or interpretation or anything. It's just one to one that needs to go in from a data source from a to b. So you definitely have those structures still, and oftentimes, it's it's combined throughout such a document including a proposal. One example that we have there is that parts of that proposal is built out by just classic rules based automation with logic, and there is no AI involved. And other parts are actually elaborating on information that sits in a CRM system to pick out information from the customer you're meeting and elaborate on that to put it into the right context, which AI is really good at. So here, it's just a matter of controlling where do you want AI to work, but it does allow you with that level of orchestration to get super specific on the outputs to a level where you can definitely trust it much more than you would be able to if you didn't have that. Awesome. Yeah. I was keen. And I said to you before, can we take a look? I'm still I'm really keen, and I know the users are all leaning in. Yeah. The people are all scrambling. Suspense. Yes. It's killing me. So I know. Apologies for keeping everyone waiting. It's just important to understand what you're actually looking for to to figure out if things are good or bad. But we are getting there now. The the final thing I would say is that I already mentioned that, but in the sort of universal agents or the generalist space, that's where you would have, like, one agent taking care of the whole thing. Mhmm. So throw anything at me. I'll build a full deck. I'll show an example of that. Yeah. Other exam, when you get further, you still have somebody taking care of specific documents with more specialists going into that, but then collaborates with other agents that are good at specific parts of the deck. And then finally, where you get into saying, well, in on the advanced AI automation, where you're very deliberate on specific placeholders where you want AI to work and also areas where you don't want it to work. So it's a lot about narrowing down and being very deliberate on why you want help. And, again, it's similar to what the world is like in the real world. When you work with coworkers, sometimes you want them to be very specific on certain areas. Other times, you want them to be a little bit more all over the place. So, what we brought today are a few examples from it. So these are examples of what it could be, like, on the on the more general use cases are, like, training materials, company updates. Those things that are allows you to be elaborate and where it's also good to get to a good starting point that you can then work on and finalize. We're into the marketing briefs and and business case and pitch decks are more in the middle where you get closer to use cases where you can be more deliberate and then all the way to highly specific, high precision type of documents with contracts, full send financial reports at the end. So we didn't bring everything for everybody. Those are just examples, but we did take, one from each to give you a good idea about how to apply agents, and that's what I'm gonna show now. Fantastic. Good. Well, without further ado then, we're going into the universal agents. This is one, again, that I'm working in the role that I have in customer success. I am constantly starting from scratch, adding slides, building like, everything from start to finish will be my intent because there isn't templates for things like you mentioned, just creating a training for a customer. So I'm excited to see how we're gonna solve that just using the intent and the the kind of universal agent. So take it away. Let's have a look. I will do. So, as was said, the training material agents, in terms of what we're trying to do with them is, like, I'm preparing some training material. There is a prompt from the user and intent, built me an outline of a training presentation for our HR team and, of course, a reaction on the other side. But what happens a little bit, around this is that the expectation from the user in this scenario is typically, I wanna get to a great starting point. I don't necessarily think out of this intent that I'll get, you know, fully fleshed material is just done. And what we're gonna apply to it, as you said, is then the universal agent, because we don't have that's the best choice right now. Had we had one specifically for training presentations for HR teams, we would have pulled that one out. But in this case, we don't, so we're just gonna use the universal. The, the business impact that we're trying to sort of impose to this are the three ones that we're always optimized for, Save time. This one is pretty good at it. Reduce risk. This does definitely some of it. Increase revenue. For this use case, maybe, but that would be very indirect. So we won't we won't say that we we have, like, a ton of impact there. So let's see how it works. I'm just gonna I'm just gonna copy the the prompt as we see it here and see how well the product actually performs. Fantastic. So what I'm gonna do is actually to leave the show view of this presentation, and, the way to access Templafy's agents can actually be done from multiple places that are go, obviously, to a website. I'm gonna show an example of that too. It can be embedded into workflows across Salesforce and other places. And, of course, it works directly inside of, inside of PowerPoint and across the office applications as well. And that's where we are now. So might be maybe that's a good place to start. There wasn't mentioned about the the expandable task pane that John mentioned at the beginning. So this is actually what it looks like. And from here, there is also the ability to, to trigger the process of building out a business document. So this is the conversational view that we have specifically created for for the agents. So it's one way to get the agents going, not the only one, but a very good starting point. And, as you can see, there are some suggested, starting points for the user as they go in here. But since we now have a a deliberate prompt already that I copied, I'm just gonna put that one in. Build me an outline for a training presentation for our HR team. So this is the request that I'm gonna give it, and let's see where it goes. So, actually, what we would first do is to see if there are any sort of high res of these fully automated, templates that would be available for this purpose. Had I had some of these, which you'll see in one of the other example, it would actually prompt me to say there's a good starting point for you there. In this case, it's, it's just gonna ask me to start using building with AI because that's the best option that we have. So I'll start that process, and then, again, the thinking process really starts. So it's now providing me with a very good structure of what it intends to put into the document so I can review it. Mhmm. And I have the opportunity, as well to iterate over it, give more information back. So it's definitely truly conversational. If there are things that are missing or things that I wanted to add, I could certainly do that. But since this looks pretty good to me, I can continue to what is now the next step. And this is then the outline view that allows me to go into more specific details on the individual bits and pieces that would actually go into the deck. So, I have the ability to unfold each one of these to redefine some of the information that I think should go into that particular area to have it longer or shorter. Let's say I wanna have another, like, emphasis or a stronger emphasis on this particular part. I could put in more or fewer slides to it, get rid of sections, obviously, add new sections that I think are I think are missing or move things around, in in this in this part. So that's what this is for. It allows you to do a lot of the prework before it even it even gets in your hands, which means that it's a way to take off some of the work that otherwise would need to be done inside of the office applications. But in this case, I'm gonna go ahead and create it, and then, the agent's going to work. And in this case, it's gonna use the the the the universal AI assistant to first define the plan, what that it sets up for itself, potentially including other agents as well Mhmm. Is what it does in this step. In this case, it's just gonna do everything on its own. So it go through to to define the plan for itself and then progresses into actually creating the slides. So as you can see here, now it actually does the work. So like with any other any other tools, so most of this out there, it takes time to do this, but it's maybe, like, 1% of the time it would take for you to put this together. So it pretty quickly goes through these, and you can actually see the names of the of the slides that are actually being created here. So it you follow the progress of what's actually going on as it lines out the information that is required to go on to the slides that we have here. And in this case, it kind of we we sort of let it loose to create what it thought would be great. But, of course, it could have narrowed it down to, like, having a shorter presentation. Of course, that's gonna impact as well Mhmm. How much time it take. But for for a deck, of this sort, it's it's still, like, pulling things together with, with, I think, pretty impressive speed, which, of course, is not valuable on its own if the quality is not really useful. Yeah. Exactly. Like the the state the status bar at the bottom as well. Yeah. You can get And you can kind of follow-up our of our your work. So it's, that's that's good. But now it's created the deck that I asked for. And, my expectation was 30%. And one thing we like to optimize for is that we don't put out information that we don't we can't stand for. So in the events where we have, like, high rest quality, we put it in. But there are also areas where we deliberately leave it open rather than just putting stuff in in that is not useful anyway. So as you can see throughout the deck, it has created, you know, definitely some of the work, but there are also areas where it's left stuff open. Yeah. So it did pull up pull together, for example, the slide for the, for the bio slides and, where we have, like, the missing people images of people, which is definitely something a lot of more specialized agent would be able to do. Here, it doesn't have access to it, so it doesn't try to do it. It wouldn't put in random pictures. But it also does other stuff where it actually is a more, like, full experience of what the what the what the deck should actually do. So in essence, it's it's pulling together, like, the complete deck that we asked for and and piece that together to the best of its ability. And I would say at least two or 30% grade in terms of being pretty much pretty well down the road. And, of course, in combination with some of the other tools that we have available and many of you are aware of here in this audience, there are abilities to, of course, work with us. Because then you have other other tools where it's easy to find, like, the best imagery and things like that, where you can easily find the stuff that is required for for for inserting images there. I'm just gonna search for people. And, so if you were looking for a specific person, that would, of course, be an ability to insert that person there. So, potentially a little bit of a of a random example, but I hope you you get where I'm going. Let's say that this is Alex. There is actually ability to find, these images and and do it more manually. So that kinda sits in the in the 70% range of actually going through the process of doing some finalization to it, where it's a combination of just putting things together to a certain state and then allowing you to continue your work. So simple example of what you can expect Yeah. From the more, generalist agents, the universal agents that would that would just help you to get it get to a good starting point. But it is a but that's the point. Right? It is a really good starting point. You've got the a good structure. You've got everything is on brand by the looks of it. Right? It uses Templafy, the template that you've got set. The fonts are there. The styles are there, and it works symbolically with Templafy. So everything else that you need is right there waiting for you as well. So, you know, you said 30%. I think one meter you could argue a little more, but, but still a really good example. Yep. So, that's the starting point. Any comments, obviously, welcome, on the chat. If there are any questions, we can follow-up on those. Yeah. We will. But, also, don't worry, that you will learn at the end of this session how to get your hands on the document agent. So patience, please. Let's move on to another example. So I'm gonna go back to, to, the deck here that we were looking at before to to tee up the the next example that we have. So just gonna go back to the overview. So in the mid example with the custom agents where we're looking for a little bit more completion, the example that we brought today is the pitch deck Mhmm. Which is a very commonly, produced decks across our customers. And similar to the other one, there is an intent from the user. I have attached my call notes or call this user just had with a customer. Based on those, please create a pitch deck for the pharmaceutical company I was speaking with. Right. So that's the intent. And, the expectations coming out of this is different from before because before it was more like, let me get started, like, get to the starting point. Whereas this is more actually a deliverable, for review. So I don't expect it to be completely done, but pretty much down the road because I've given it more specificity. And also, there are better agents to solve this specific thing. So we have specialized agents that can jump on and do specific things on pitch decks. So that's essentially the difference. It also means that some of the impact that you can make with this would also change. So time saving, yes, absolutely. The risk reduction, absolutely as well. It sits there because, you you're taking out some of the, some of the some of the factors that can be done, just backed on on on knowledge rather than having to to figure out what might be the best thing here. So there's a lot of risk reduction going into as well outside of the template, of course, with having that structured in a certain way. And then the increased revenue is one that comes in here, of course, since we're now looking at pitch decks. So it's supposed to do something good, and it would ground its information and structure in something that has actually worked well in the past. So the whole context is king, is really coming into play with this because what happens on the background is that it uses some stuff, outside of the prompt that the user informs, this process with, which is the call note as you'll see in a second. But the agents themselves are also good at grounding the information that they put to you in something that has worked well in the past. So big difference even though that the process seems very much the same. So let's take a look at how it works. Awesome. Let's have a look. Good. So I'm just gonna leave this now. And this time, I won't start the process from within all fine. I'm just gonna go to the website so you get a sense of, what that looks like too. So as I said, like, obviously, you can go from anywhere from your phone to to work with Templafy in a completely similar way that we looked at before. So it's not like a massive difference. But, and it's also embedded into, as I said, also to other tools as well. So similar type of experience, but I did copy the the the prompt the prompt. Yes. Let's put that in. And, also, let's add the the file notes or sorry. The call notes that, that was I put into the prompt that I actually have. So these are the notes that I have from that particular call. And it's just a word file? That's just a word file. That can be processed from, like, other good tools, like, that would do transcripts and pull together, everything that you've done on a call, to put that together onto a document of some sort, and then you can use it, repurpose it to use it here. So that's one example of how that is quite, quite capable. So So once you start doing that, it's a good example of of having a combination between the user giving good instructions to what to their intent, what are you trying to do, but then starting to add context. What is the back you should be building this on? And as you start combining that with since this is a pitch deck, we would understand that, it looks like we need to trigger the pitch deck agent who is the best to do that with which again is very good at, connecting to the the important data sources for building pitch decks and other agents that can do that. So it's kind of a flywheel of context that goes into this, which really influences the end result on the back of this. So, let's go ahead and try to do it. So similar to what we saw before, we now have better information and we can start building this using AI and triggering the best agents possible to actually go through this process. So very again, like, similar type of process just with higher accuracy in the output, It actually does the same. And it's very deliberate for us, but we prefer to have the user doing the same the same thing every time. They shouldn't be bothered with this. This is one agent or the other trying to come in to help. It's more a matter of can I get to where I need to need to go to? From a user perspective, this is much about can you help me get my stuff done with an accuracy level that I can use. From a company perspective, it's about, you know, injecting yourself into that process to make sure that as the user wants to get stuff done, you can help them make sure it's done with the highest possible level of accuracy. So that's exactly what happens here as well with, you know, all the stuff that should go into this pitch deck, completely similar to what we saw before, the process of setting this out. And with all the structures there, the ability to flip these things around or to optimize the individual pieces, and then ultimately to go ahead and create the presentation. So, again, the the process is very, very similar, to to what we looked at before. But what actually happens behind the scenes is a little bit different because of how things are set up. And that's a good way for a good place for me to maybe put in a little commercial on the third, part of of the webinars that we're doing here, because, that looks more into how to actually set these up, all the technical bits bits and pieces, connecting to different models, all that stuff. So that's for a different webinar, but that's actually what goes on. A lot of that stuff goes on behind the scene here. Well, I was gonna mention at the end, but it looks like you're trying to take my job absolutely fine by me. But so you can see here, it's the same as what was happening in the last example. You can see every step and what's happening in the background, and, obviously, then it's going in that example you talked about where the humans used to do this, in this case then, the agents are going to pull from the right sources. They're going to see, and pull in the right template, and they're pulling the, the contextual information from that document. And so the user we're aiming here for around 60% close to completion, and this is a company sorry. A customer facing and outward facing deliverable. So compliance, and the sort of correct sources, correct information is paramount in this case. Yeah. So I'm excited to see, what comes out. Yeah. Okay. The presentation is ready, so we can now go ahead and download it and just pull this one down. So, and to pick it up from our, our download folder. So I'm just gonna go over there to pick it. First time on the PC. So it put together the slide and, you know, definitely, it looks looks, pretty reasonable in in terms of the output. So looking through the slides, it definitely looks like this is a this is a a pretty well generated pitch deck. And in the general approach, it's kind of it's done a lot of the job. But what I wanted to get to here is that oftentimes when you get to this level, there might be stuff that is still, missing potentially or extra things that you actually wanna do with it. And I wanna share a little bit about how you can also continue on the iteration inside of, inside of using using our tools to do that and using agents to, to get closer to that. So there is the ability directly here inside of, a PowerPoint as well to, not only as we saw before, build full presentations, but actually iterate over generating additional slides. So if you go down a little bit, there is a section here at the bottom where I actually want to, include a section for the, for the ROI. So I did, prepare a little prompt for what that could be. So I'm just gonna put that in here So but you can see it. I did put, like, a little bit of context to what do I want to happen, with the slide that I'm building. So giving some providing some information to what a slide should do and then essentially having a very similar process to, what we've been looking at earlier on how to actually process these things. So this is an agent, but for just a slide generation as opposed to the entire deck? Exactly. So this looks specifically into creating some of the stuff that you might be you might be missing. So whereas the other one would be pulling together kind of the full context, putting up the structure, building out all the decks in the middle, what this does is that it allows you to build these extra pieces that you might be missing. So it's more it's closer to the the iteration and perfection phase where you're trying to do do additional stuff that otherwise might be missing. So that's what this piece is doing. So it's very similar to the experience to what we just looked at except that these are just very specific slides. But, one important piece is that on top of it being able to, leverage, like, several agents, that are available, it also is able to take the context of what's actually is on the slide deck to use that as context for putting it together, which is why the example of putting together an ROI slide is actually quite useful. That's really cool. So it's now generated, and, let me just go ahead and insert it to see how well it, it actually works. So it's actually generated the ROI slide here with, with pulling from some of the, the predesigned, stuff that exists or the content that exists within the company Mhmm. On what a good way to portray, like, an hour slide would look like, and then essentially piece it together on the back of the information that we have on this deck. So that's an example of, going through a process pretty quickly, adding context, piecing a deck together, and doing a little bit of iteration on the other side. So, of course, tons of more things to do here, but, but, definitely, hopefully, that gives a good impression of some of the direction that we're taking. Yeah. It really does. And the question I know is gonna come up is can you use the, the slide agent however way you're describing it? Can you use that in any presentation, or does it have to be in a in a particular like, an agent created presentation? No. It works in the exact same way where you could have, like, generalist agents that are essentially capable of doing anything you throw at it through a certain level of, of, of high risk, so to speak. And then you can have more specialized agents with this one that is particularly good at doing ROI slides for that and has, like, a certain extra capability around that area. So both things apply in the exact same way. These are just kind of your little workforce of agents that you can apply for different things. But it would identify the right agent depending on what you're asking always. That's kind of a built in thing for it. So, yeah, it works across. Fantastic. I know this is, gonna be a very desired feature for a lot of our customers. And this is one of the requests that you get a lot. Right? You know, can you make this slide look pretty? Or I wish there was a way. And I actually had this in a customer training. One of the we were doing talking around AI assistant. An AI assistant is very much focused around text generation, and text, iteration. And this particular person, they said, I wish there was a way I could write a prompt and have a slide, and now we can. So Yeah. Now you can. Now you can. So, yeah, excited about that. And another good, question to ask other vendors in terms of is that a possibility you have around us. I'm just showing off some of the stuff that we're doing on the back of the request that we're having as you're saying. So but I wanted to get back to, the final, example that we also prepared for for you, where we actually move into the area where we still have agents doing work, but we're very deliberate on where exactly they should be doing the work for us. So it's in the advanced AI automation area, and the document type we'll be looking at is a proposal. So similar to what we've been looking at the other, there is definitely an intent from the user that we're trying to pick up and an outcome that we're trying to get to. And in this case, I need a strategic proposal for a Skype for Samuels that positioned us to win against the competition is actually the prompt that sits here. On the two sides, we have now expectations from the user rising. So in this case, they would prefer to have a ready to ship deliverable with very little work that needs to be done on the back of that. And in order for us to get there, we're now using advanced AI automation where we plug in agents as part of a mix between between AI and also rules based automation that sits within this. So it's going to, affect also the impact that we can have. There is definitely more configuration going into this type of setup, but, also there are better outcomes on the other side. So save time very high, reduce risk very high, and also now increase revenue very high because the accuracy and precision and how it's grounded and how it's controlled is different from some of the other stuff that we've been seeing. So I'm just gonna copy the, the prompt here and move over to, the website that we looked at before. And we'll just paste it in here, and let's see what it does. So as I would expect, this actually comes up with extra suggestions for me compared to what we were looking at earlier. Because now, we actually have a pre prepared template for doing proposals that has a lot of built in dynamics and rules to it. And, it still gives me the option to build this completely with AI, but actually, it suggests the better option of getting to the highest possible level of quality that I can get to. So I'm gonna select that one. So what happens here is that on the right hand side, further instructions to give to this particular template pop up. And it's been pre filling out some of the information that was already given, through the prompt already, So I don't have to do that, but I will have the opportunity to actually change some of that. And the instructions here are things like, who's the client name, what is the industry we're going to, that we're going against, Is there a sub industry we should be aware of? Are there specific people? And, essentially, those gating questions can be many. But what they do is that it inform, the underlying template on what it should actually do, where it should go capture, information and content and data, And it also informs where it needs AI to perform, its job. So the agents that we've been looking at earlier can still be used. It's more it's just more deliberate areas of the document there with there were that where they would actually be, be coming into play. So as I go ahead and download the document here, it creates for me. There you go. And I now have the ability to open it. So as you can see, these are this is a pretty fully fleshed proposal document that has a lot of specification, a lot of detail, also more than we looked at before when looking at the pitch deck, for example. So it is a combination of putting this together with rules, which means that some of the slides are just set slides that requires no changes to at all. They're just as is. Or to information of specific areas of the document that actually comes from just rules based logic. But it also uses the AI agents again just in very specific areas. So one example of where where that is actually, happening is on this slide, on the slide number nine, where this piece on the challenges and the solutions is actually something that is picked up from the CRM system. So, this in in in this area, it's been looking at some of the challenges and solutions that we've been discussing with the customers on this particular area, and then it picks that up and repurposes it to put into the slide and combines it with other parts of the deck, which is just completely rule space. So it's an example of being able to combine things where the best capabilities are actually used in the best possible way. The other example, and that's the final one that I would give you, is also the ability to continue to iterate over these slides. We looked at before how you can build completely new slides, with, with the agents as well. But there are also, agents that are even purposed down to the detail of just iterating over text. So that is definitely also an opportunity, an an option here. So just highlighting some text and then going into, selecting what you wanna do with the text. Let me get rid of, of the highlight there. So going in here, I'm just gonna rephrase this piece. And as you can see, it pops up with one of the purposed agents that is really good at rephrasing stuff and gives another suggestion to what this text could be that I could easily replace the the current text with with a simple click of a button. So that kind of, concludes the three demos that we have, prepared for you. Yeah. That was awesome. I, I've definitely learned some things that I didn't even know. But so, Christian, how should we wrap this up? What's the the message you would like to send? Yeah. I think we've been through a lot. Thanks for staying on. If there's one thing I would want to conclude on on the slide that we looked at a little bit earlier here, on what we've actually been looking at because we promised that what AI can do and what we're trying to do is to get closer to all of those bespoke documents in the middle of, otherwise, were very difficult to get to. We've been trying to put on some context on what to how to think about this as you start, looking into potentially onboarding solution for this and which questions to ask vendors around this. We've obviously been able to to show a little bit about what we do here at Templafy. But, and and hopefully, that's been helpful. But at the end of the day, it's about having the, you know, the more generalist universal agents to support some of those throw anything at me type of use cases and still get help to a certain extent. Once you get closer to the to understanding the use case and what good looks like, that's where the custom agents really come in with higher specificity and precision in the output. And if you wanna go all the way, it's still it's very much, it's very much the the the rules based automation now just blended in with AI that can that can probably help on some of those cases. So, I think that's a good that's a good place to conclude. Awesome. Well, thank you very much for being on Templafide Talks again. So two things. Firstly, if you want to learn more about document agents and see them in action in person along with a case story from, ABB, join us in London on October 14. There's gonna be a link in the chat. You can just sign up there, and we'll see you in London. You might even, recognize a couple of, familiar faces at the event. And if you want to get your hands on document agents now, you can. Document agents is live. We have it on general release. There is a link another link in the chat. Yes. There is a lot of links. But you can sign up now. We'll get in touch with you, and you may be lucky enough to get on a call with Christian. He'll take you through, the setup process, and we can get you started. So, thank you very much for being part of Tenify Talks, for joining part two. Part three is coming very soon where we're gonna have, as Christian spoiled a little earlier on, interviews with the team that created the document agents, show you how to configure them from the back end, and really, drive more value from the solution. So from myself, from Christian, from Casper in the chat, and Isabella behind the scenes, thank you for joining Templafide Talks. We'll see you on the next one. Thank you.