Name: How to transform work with Claude for Excel and Claude for PowerPoint
Uploaded: 2026-02-12T19:16:06.200Z
Duration: 1 h 4 s
Description: How to transform work with Claude for Excel and Claude for PowerPoint

Transcript for "How to transform work with Claude for Excel and Claude for PowerPoint": Okay. Welcome, everyone. I'm Nancy from the Anthropic Marketing team. We'll go ahead and let folks jump on for just a few more seconds here, and we're just very grateful for you all joining us today. Awesome. Okay. Well, it looks like we have a lot of people on the line, and we're very, very excited about this topic. So, again, thank you so much for spending, the next, you know, half hour to an hour with us, and we're just very, very genuinely excited about this topic. You know, at Anthropic, we've seen a lot of transformation, you know, and we we think that the same transformation that we've seen in software engineering is also coming to the rest of us across the organization and that we're at a very unique inflection point. Last Thursday with Opus 4.6, we, you know, announced a few exciting updates that we think will really kind of change what's possible for everyday work and make it really easier and faster to delegate work to Claude and go from things like raw data and unstructured files to really polished Excel models and PowerPoint presentations and, you know, a lot lot more. So today, you're gonna get to hear directly from our product team behind the strategy for how we're thinking about these, office work based agents, and you'll also get to hear directly from our customers as well, who are using these tools in their day to day work, and we're just exceptionally excited about that. So let's go ahead and dive in. So we have a couple of quick housekeeping items for you all. The first is that, you know, in answer to the top question that we always get in these webinars is, can I get a recording? And the answer is yes. We will be sending a recording to all of you within twenty four hours. So if you need to drop early or you wanna share this with a colleague, we have you covered there. Secondly, we want this to be very interactive, so please use the q and a tab. We're gonna have a lot of time for questions, and we'd love to see, you know, a lot of engagement for this group. So we do have a lot of folks online, so we might not be able to get to every question, but we're gonna definitely do our best to address as many as we can. And finally, we'd love your feedback. You know, we always send surveys at the end of these webinars, and, you know, it really does help us make these a lot better. Great. So in terms of what we're gonna cover today so first off, we're gonna start with a quick walk through in terms of everything that we've launched. And this will also include a quick, product demo so that you can see the new features in action. And then we're gonna close with a customer panel and q and a from the audience, which is always the best part of these, you getting to hear directly from your peers on how they're using these early tools to get a lot of value and drive adoption across their teams. So with that, I will hand it off to, Nick and Vinitra to, really kind of step through what we launched last week. Amazing. Thank you, Nancy. Good morning, good afternoon, and good evening. My name is Nick, and I lead our product efforts for Cloud for Financial Services. Hi, folks. I'm Denitra, and I lead our product efforts around PowerPoint. Alright. And Vinitra and I are recovering consultants and bankers. So we've spent a lot of our lives in Excel and PowerPoint. That's why I think we're especially excited to share what we've been working on, which are hopefully practical, useful tools that we wish we had in our hands a decade plus ago. You know, I think a lot of the difficulty in adopting AI has been needing to change how you work. Right? But instead, here at Anthropic, we really want to meet you where you are. This is a big reason why we've launched Claude in Excel and in PowerPoint, where Claude is really a thought partner for you directly within the application, really helping you through deep data analysis and refining your creations. Empowering all of this is a state of the art model intelligence from, as mentioned, Opus four six, our newly released model as of last week. And our cohort product really connects to a full suite of tools, data connections, and skills, and you can really imagine Claus seamlessly moving across your desktop application into Excel and to PowerPoint as well. Just to make that real for you guys, let's show you what that looks like in action. Next slide, please. Now, you know, let's recap what we just saw. Tom is a local pizza shop manager who really want to see how he should be expanding his business, which, of course, requires brainstorming, financial analysis, and obviously convincing his boss on the right decision. What Tom was able to do was that he had Claude Cowork research across a large number of messy documents on his local desktop to come up with initial set of recommendations. Now he narrowed that down together with Claude by having Cowork producing a first draft of a comprehensive financial analysis, which he iterated with Claude again directly within Excel. And finally, a CEO ready a presentation directly in PowerPoint, which he refined again directly with Claude. You know, as you can see, Claude is really your thought partner throughout the entire journey of doing research, financial analysis, and creating outputs and presentations. Next slide, please. Now to make all of this possible, we've worked really, really hard with our research teams to improve our model capabilities. You can see this on a new benchmark we're calling real world finance, which really measures cloud performance across about 50 investment and financial analysis use cases across research, spreadsheets, and slide decks. Now these are tasks commonly performed by not only finance analysts, right, but also managers of local pizza shops, managing directors and SEPs at the Eshaw and RBC who we'll be hearing from in a bit. So, you know, we we're really excited for us building within financial services to push some of these very generalizable knowledge work functionalities and capabilities forward on both the model side and the product side. But we're still really early. You know, on our side, we work really closely with the research teams. We've only started finance specific training just a few months back. You can see on this benchmark, CLUCK OPUS four six improves over SONNA four five by 23 percentage points. And SONNA four five was our state of the art model just a few months ago. You know, we can really field intelligence not only from enterprise and consumer customers, which Vinitra will share in a bit, but also our partners building on top of the raw model intelligence, which is our API. Now these partners have really robust benchmark evaluations, and they've been blown away by Oopus as well, including a few names you may be familiar with, Shortcut and Hebia, who are both building powerful spreadsheet and PowerPoint agents themselves. Now I'll pass it to Bediktur to go a little bit deeper into these capabilities. Yes. Thank you so much, Nick. Well, we're super excited today to first announce a slew of improvements to Cloud and Excel, including many features that we know many of you have been eagerly waiting for, such as being able to interact more natively with elements such as pivot tables, being able to do conditional formatting, and add and edit charts within Excel. We've also added several usability improvements, including auto compaction so that you can keep Cloud and Excel running longer on the complex task that all of you work on. So we're really excited to announce these additional features and support for the kinds of advanced Excel features that power users like all of you probably will end up using in your day to day work. We also like to talk a little bit about the difference in outputs that we're seeing between OPUS 4.5, which was our state of the art model, just before we announced OPUS 4.6. And we're actually finding that OPUS 4.6 is much more capable, across the board. It's able to think through complex, tasks like commercial due diligence tasks, and it's also capable at standard formatting for finance professionals. So you should be able to let it run even longer on the tasks that you'd like to work on. Second, we're also super excited to launch a whole new product surface. It's now in research preview, and it is clawed in Power Point, which is a similar add in experience for those of you who are familiar with Claude in Excel already. And we're excited about this because Claude is now able to directly read your slide masters, directly follow your templates. And, because it's a sidebar add in experience, Claude can make very precise and granular edits to specific slides or objects that you've got selected. Finally, we're also excited to say that Claude is able to work with many PowerPoint elements natively, including charts and tables and diagrams. So we're really excited for this to now be in research preview and for those of you on enterprise team and max plans to be able to test it out. Again, with the model continuing to improve, you'll see improvements in outputs here as well between OPUS 4.5 and OPUS 4.6. So, of course, earlier versions of Claude were able to make basic boxes and diagrams, but Opus 4.6 is gonna be much more performant in adding in PowerPoint native elements like charts, and you'll be able to edit them as usual once they've been added to the slide, which is pretty great. We think this new product surface will be perfect both for generating full presentations, that follow your corporate templates very precisely, but also allowing you to then iterate and make granular edits on those slides, which then brings us to what have we been hearing from our users on this topic. We've been really excited to hear positive feedback from the users that have had a chance to try these capabilities so far, including HG Capital and Deloitte, and we're hearing that they're finding both Cloud in Excel and Cloud in PowerPoint to be powerful tools that accelerate their day to day workflows. And speaking of users, I'm very excited to have a couple of users joining us today to talk about how they're thinking about both these products, but AI more generally. So thank you so much to our panelists for joining us today. On the line with us, we've got Andre, a managing director and head of commercial innovation at RBC, as well as Nikhil, an SVP and co head of generative AI business strategy and transformation at DE Shaw. So we're really looking forward to hearing from them about their experiences and how they're thinking about the ecosystem. Nick, I'll throw to you to lead us from here. Amazing. Thank you for the wonderful overview, Vinitra. Now, Mikhail, Andrey, I am really excited to spend time with you both. Andrey, this is our second time on a Anthropic panel together. I think these panels are some of my favorite moments at Anthropic. You know, we spent a lot of time with our research teams, with our product teams, building great models, and what we like to think as great products, but we learn so much more about how AI is actually being adopted from industry leaders like yourselves. So my first question to just kick us off, can you both share a little bit of background on your role? And what is your vision for AI within your organization, and what is its role in day to day knowledge work? Maybe we can start with you, Andre. Sure. Thanks. Thanks, Nick. Thank you, everyone. So my name is Andri. I lead the AI and digital engineering team here at RBC Capital Markets. RBC has a long history in innovation and especially AI innovation. We are ranked number three in the AI, in which ranks banks, international banks, in their AI maturity and capability. We're very proud that, you know, back in 2014, our CEO, Dave Mckay, invested in Borealis AI, which is our state of the art AI lab, in Toronto. And since then, we've been, you know, heavily investing in the space. And since, Generative AI sort of, like, really kicked off, we have benefited significantly. We have used cases stemming from capital markets research to the investment bank, to sales, to trading, and and any everywhere in between. So my role is to, have the vision for the products and also the, like, engineering of of of the products and make sure that we are utilizing AI to the best to serve our our clients as best as we possibly can. Amazing. Nicole, over to you. Yeah. Well, thanks for having me. So I co head a team that drives Gen AI strategy and transformation across D. Shaw. And before coming to, I spent fifteen years or so studying people and how we adopt and adapt to new tech paradigms of all kinds, whether that's cloud computing, mobile, social media, obesity medicine, you name it. I think what's so interesting about AI specifically is that it's sort of different, at least to me, than any of the other sort of tech paradigms we've been dealing with recently in that the capabilities of AI are emerging, So we don't really know what it's good for yet, and we have to do the hard work of figuring out exactly where it adds value in our own sort of individual and collective workflows. And then we evaluate again as soon as there's a new model released because then suddenly it changes again. Right? So it becomes this sort of constant discovery problem, essentially sort of research race. And I think that research is as much about better understanding ourselves and our organizations as it is about better understanding of the technology. Right? Like, we need to break down the tasks that we do to understand where AI actually provides an advantage, where it doesn't at least yet. We need to build a much more fine grained understanding of all our own skills. Like, what does this actually mean if we say we have amazing judgment or I'm good at building a model and so on? And we need to sort of actually try to tease that apart. So I would say, from our perspective, we don't yet know exactly where AI fits into our work today. I think we're learning very, very fast, and we see that it fits in in many places already. But we are doing our best to stay at the forefront of that discovery process. That's really how I see the mission. Yeah. Mikko, I think that's great. And, hopefully, it's our job at Anthropic to show you what the forefront really looks like from both the model and product capability of perspective. I I really like what you said. I think, you know, in a world where Claw can do a lot of the basic manual data cleansing, financial modeling, PowerPoint presentations, what do we do with all of that extra time? Right? I think a lot of the time should be spent thinking and raising our judgment and getting to the core of what we do as finance analysts, right, which is really understanding opportunity, identifying those opportunities, and doing the thorough diligence and analysis that we need to. So I'm excited for a lot of these capabilities to, you know, grow with us as well over time. Now, Andre, you know, RBC was one of the first customers to come to us, right, about PowerPoint capabilities. And spending a few years as an investment maker myself, these things are really manual and really hard to solve for. Could you maybe share a little bit more about the problems you are hoping to solve with PowerPoint and AI and just some of how some of our newer products were able to address some of those problems? Problems. Absolutely. So, you know, as you as you know, Nick, but, you know, I'm I'm sure a lot of members of the audience are are probably in the space as well. What you know, a lot of what especially a junior investment banker does is they spend a lot of time building, PowerPoints. And PowerPoint is such a crucial similar to Excel, actually, both of those are such a crucial, app applications in finance that it's really hard to understate. And what we wanted to do is, you know, can we help our analysts be more efficient when they're generating these decks based on, you know, some of the, like, the historic decks that we have, taking, you know, best in class, and then build something new on top of those. And as you mentioned, Nick, it's a it's a really tough problem. And and and, you know, we we had some of our own internal iterations before we start engaging with you guys, And, we start hitting, like, you know, different walls. Right? Like, it's easy to generate text. Right? Like, we know the models are really good. It can generate the content. Content is actually the easy part. But can you get the boxes to line up? Can you get the tables to show up correctly? Can you get the color scheme to follow through as it as it should? And those you know, it sounds kind of trivial, but it's actually extremely difficult. And that's where Cloak or PowerPoint has really, I think, like, changed the game, frankly. And it's been, like, pretty powerful in that in that setting. Similar, you know, in Excel as well, but definitely in PowerPoint. Yeah. Amazing. You know, I think we are also, as I mentioned, just getting started for these capabilities. We still have a long ways to go. Cloud is not perfect yet, but we've been really encouraged by how quickly some of these capabilities have been evolving and improving over time. You know, I often get the question, where do I get started with AI? Right? It is such a powerful system, and, you know, Clock can do a lot of things for you. My recommendation has always been just give it a shot or experiment with it and see where it takes you. I'm curious for both your vantage points. What can you share where you're seeing impacting your own work and what's really surprised you about ways that it's transformed your own workflows? Maybe, Mikhail, I can start with you. Yeah, sure. I mean, I hear you on experimentation, right? Like, that's the only way to really learn. When it comes to Excel specifically, initially last summer, there was suddenly the ability to start working with Excel files in new ways. And I feel like back then, we discovered some amount of value from it, but it was somewhat marginal. Right? But then you look at the the changes that's happened since then, and it's just it's pretty breathtaking to me. So for me now, like a lot of my own work with Excel is stuff like survey analysis, these kinds of things, tracking tasks, relatively basic. But still, in terms of exploring data sets, visualizing datasets, and so on, we're just now in a totally different place. I feel like the moment we're at right now with Excel is not too far from where software engineers were with ThoughtCode last summer. So we're maybe sort of six months behind, but we're really starting to see that sort of like, oh, now I can actually do stuff. So, you know, I was working using it last night and doing sort of every step from, like, the fundamental merging of data sets, just like taking all these, like, you know, somewhat broken sets that need to get into a single sheet and then start exploring it, visualizing it, and so on, I think it's still very much a step by step process. It still makes tons of mistakes. You still need to sort of babysit it quite a lot. So we're not there where I can just say, Hey, fix it. But we are at the moment now where I can really sort of take each step with me, and honestly, it means that I don't have to do too much manual manipulation in the sheets myself. Yeah. I think one of the things that, we've noticed as well is, obviously, the first versions of these products end up being sort of an MVP testing to see whether they're meeting the core use cases of our users. But then it's actually pretty shocking how quickly we can learn from the real life use cases of users like you and all of your peers. And so that really helps us think about how to make these products more performant for the specific ways in which users want to use them. And we, of course, filter that back to our research team so that, you know, OPUS four seven or five or six will be even better at all of these tasks than they've been before. Yeah. That's. right. Maybe one thing I should should mention that do think is maybe specific to the Excel plug in or, like, that way of working versus when you just work through the cloud is it's much easier suddenly to suss out some of the hallucinations when I can just ask the model to give me all the data as formulas that I can track all the way down as opposed to some sort of hard coded number that, like, I don't really know if it did the calculations correctly or not. So the work of evaluating the outputs has become a lot easier to me. And that that's why I think we've been so excited about these net new form factors where Cloud is actually just sitting in the application with you. Right? Prior to these applications, Claude was, of course, doing a lot of iteration with you directly within the desktop app or the web app and then creating this first draft. But you have no idea where a lot of these numbers came from. You can have to one shot, you know, one task with Claude, but that's not how we really get work done. Right? As a finance analyst, I used to spend, you know, eight hours a day within Excel, refining my assumptions, you know, building up my my projection of what the future looks like. I think I really like what you said, Miguel. I think, you know, it's really important to keep in mind what I call the complexity curve. You can think of what Claw can do in essential essentially sort of ladder steps. Right? You know, you're not gonna give Claude a 20 tab leverage buyout model to to start with on day one. You're gonna have to build your trust with Claude incrementally over time, find out what it's really, really good at and where it's not, and that's where we really push the boundaries of how we work with these AI systems. You know, the main feedback I've gotten in the market is cloud is amazing at just looking at raw data. And that's really where all of your workflow starts from the very beginning. Right? Trying to piece together the patterns, the trends, the risks, the opportunities across these disparate datasets that was never able to be done, you know, before you really had cloud within these applications. So I would really encourage those on the call to start thinking about some of these use cases where it's really, really manual, impossible for you to do yourself. Already You pro probably already have the the leverage buyout, you know, template fully set up. So cloud is not going to be fully at value added there, but it can really help you explore these data sets and then sort of expand your mind in terms of where to look. So and that's kind of what we're seeing on the Excel side for sure. I am pretty confident that Clock can be a really good leveraged bio modeler in just a few months as well, but we can sort of, you know, take it from there. Andre, I am curious to hear a little bit more about your personal workflows as well. Can you tell us maybe one surprising example of where you've been using Cloudlitly? I mean, I would say, like, like, almost every single workflow for me personally has been touched by by Claude. I still you know, even though, like, I lead a team, I still develop. So, obviously, like, Claude code helps there. But then Claude for Excel, Claude for PowerPoint, I need to, like, you know, build a lot of decks to, to showcase what what we're working on and what what team is working on and and so on. And then help me sort of accelerate that significantly, help me go through, like, just just basically be my my assistant, like, almost everywhere in my in my workflows. And, I mean, I completely agree with with sort of, like, the the current limitations, but I'm really excited about sort of, like, what's to come. And I kind of to your point, Nick, about bringing cloud or bringing AI effectively, like, to where you are or bringing it with you, like, everywhere you go. And and it like, starting to collect that context. And we kinda do that right now, like, internally through different ways. But but as as the space matures even more, we'll start seeing, like, effectively, amazing assistants that have so much context and can like, are also experts in in in helping you with whatever you niche question or product that you you you. need. I couldn't think of Claude as, you know, giving me parts of the Ironman suit. Right? I am not a designer or an engineer by trade, but I am pushing pull requests, building my own prototypes all the time myself. Right? I think you can think of Claude as really an extension of all these things that you wish you could have been able to do, but you're not good you did not go to school for it. Right? Claude can give you all of those capabilities, very wrapped up in a nice way. And I think, you know, I would really encourage those on the call again to think about things with a little bit of persistence. You know, I often see posts on Twitter, for example. Cloud did not make this beautiful leveraged bio model a 100% correctly. Therefore, I'm not gonna use it for Excel task ever again. I think that I would love to just maybe shift the paradigm a little bit. Try it for other use cases that you don't want to be spending time on today. Right? Manual formatting, manual adjustments of, you know, model that models that need to be debugged, the data analysis use cases I talked about before. So really explore the depths of these capabilities and understand where things are working well, and all of these really need to be fit into your own personal workflows as well. You know, I think I'm I'm really curious. You both are leaders, obviously, driving AI transformation across your organizations. So let's maybe talk a little bit more about AI adoption. From your vantage point, what do you think are the biggest challenges you're seeing as people start engaging with these new AI systems? Maybe, Mikhail, I'll start with you. Sure. Where do you want to start? That's a long list. I'll pick I'll pick on one. I think one thing that I'm tracking very closely right now is the impact of AI on skills. So how do we build and maintain skills when AI does a lot of the work for us? And, really, it becomes a question of which skills are we building and how do we do that, what skills create sort of real advantage right now. And with that, like, when is it important to understand the inner workings of your Excel sheet, let's say? Right? And one way that I think about that, at least currently, is my goal is to help most people in our firm become amazing race car drivers, so to speak. They don't need to know everything about how the car works. Don't need to know how to rebuild the transmission or so and so on, but they do need to know how to read the track, how to, you know, feel when the when the grip is is sliding and and so on. And and that's sort of one set of skills that we need to nurture. Right? Like, how do we use these tools through their optimal performance? But then at the same time, you also sort of know that every good race team needs a good mechanic. So who becomes the mechanic? How do we make sure there are mechanics that also can go really deep and can can debug the the whether it's an Excel sheet or a a bunch of code or whatever it is when it's really necessary. So I think the question that we're working on right now is how do we make sure that we build the right skills and which people should build which skills from an organizational perspective. I love this analogy. I'm definitely gonna steal it. But I also was thinking about how to extend it. I think the the point you're making is also that different people driving the same car in f one, for example, might get completely different results because they're better or worse at steering the car to do what it what they want it to do. And I think we can think of AI tools similarly. The more you develop a familiarity for what the model is good and bad at, the more you, kinda feel like you can steer toward the right output that you want and kinda make it more performant for the use cases that you are really interested in improving. And so, yeah, I love the analogy. Yeah. I might even take it one step further. I actually think the model intelligence itself is just the engine. Right? I think we're all going to be driving different cars. There's a lot of ways for us to customize. our individual race cars with things like model context protocols, the plug ins that we just launched, how we use cloud across Excel and PowerPoint and others, and, you know, our philosophy on at least the product building side. You know, the model intelligence is going to be generally extremely powerful. But on the product harness side, we really want to make it super customizable, and that's why we've launched open source things like MCPs and skills. Alright. Andrey, what do you think about the biggest blocker? What is that for adopting AI within RPC? I think it's sort of understanding that not everyone is is sort of, like, maybe at the level that, let's say, the four of us are. Like like, if we're, like, really advanced, like, we we we know the latest, like, advancements of of Claude and, like, capabilities we can do and we cannot do, and we've fallen in this space. But a lot of other people just, like, maybe tried it once and then, like, asked it about the weather, and then they, like, they dropped. And then, you know, they kinda move like, went back to the way they do things. I think change management is a is a big question or sort of like a one one of the big challenges. And it's it it it has to happen through a lot of expo like, people just, you know, kind of like, we we encourage people to try it as much as possible. We have a a community of of what we call AI champions. It's basically a group of people that come from different areas of the business. And, like, we meet every week. We share, like, ideas within the champions, but then they go out and then they distribute those, like, ideas and best practices amongst their desks and and, you know, the people around them and and and so on. So so it's almost like a like a natural way to kinda spread the word, and and best practices and things like that. So so I would say, like, it's it's it's kind of understanding that, again, like, it's a journey for everybody. Not everyone is at the same level in of of of, sophistication. And that's not not it's not a bad thing. Right? That's just that's just, like, the natural progression of things. So we just gotta we just gotta help our our our teams to, like, feel more comfortable and and just just grow. actually. curious about that, and would love a a quick follow-up on it. I I'm curious if you have any kind of tactical advice for the folks on the call of what you found most effective to get the word out about what models can do or even how folks should be thinking about using AI. Is it the small group of champions that can show real use cases using the company's data? Is it more training? Something else? It's it's it's it's the first two. It's basically simple use cases that, like, you might not think about because you might not necessarily be in that, like, that deep in the weeds. And it might just be like, hey. You know, like, a really simple Excel spreadsheet that, like and you're using Cloud, like, for Excel, and then then you just ask it a certain way. And then, like, it's a shocker to, like, to to people. Oh, like, oh, that's possible. So, like, things like that, but it but it has to be tangible. Right? It has to be, like, what's the what's the outcome? Like, why at the end of the day, it's like, why why do I care? Why do I need to, like, use this tool? And having people having the AI champions do it in in a way that's that's personal to the teams, that's been really helpful for us and kind of help grow adoption significantly. Yeah. I love that. I almost think of it as a sort of hub and spoke model for at least what we've seen enterprise adoption. Right? We spend a lot of time with the organizational centers of excellence, AI champions, which you all really have that start start disseminating some baseline knowledge to all of individual teams. And then we've seen our most successful customers even nominate force nominate, while in while while in tell, I suppose, a few individuals within each organization to really push boundaries within their AI adoption, and then that sort of knowledge gets disseminated across the organization. There's no way for Anthropic to be building for every single use case for RPC and D. E. Shaw. Right? Y'all are much more intimately familiar with your systems and your processes than we are. So, again, that's why we come back to we want these systems and products to really be fully customizable, and that's why we're excited about things like skills and and plug ins that we've launched. But, you know, I think it's great that these systems are really powerful, but we alluded to this a little bit earlier. Trust is everything, especially in finance, right, whether for internal or external deliverables. This is probably gonna be a hard question, but how are you all thinking about building confidence and trust in AI assisted work? And I think the other folks on the call would certainly again appreciate pragmatic practical advice. Mikhail, maybe we can start with you. Yeah. I mean, it's a tough question. Particularly, I mean, in my environment, we have an incredibly high bar for accuracy. So what do you do when almost fundamentally these products aren't going to be totally accurate? And there are sort of different ways into it. One thing that One sort of old idea that I sometimes use to sort of structure my thinking around this is there's this old idea called the trust equation that people talk about for trusting other people, right, like how we build trust in other humans, which is the idea that trust is driven by at least three factors. One is reliability, another one is credibility, and the last one is intimacy. So if you sort of take those in turn, like, reliability, obviously, is the easiest way to build trust, like get people's stride for themselves, see that it works, see that it works again. It works over time. Great. Right? And, of course, there's, you know, large amounts of technology, infrastructure, and so on that can help us get there. But there is this sort of problem, right, which is like, well, how do you get people to try it again and again if it didn't work the first couple of times? Like, how do you if you don't fundamentally trust it, like, you're you're not gonna build that reliability. So So then we're sort of left with the other two, right, so you can instead lean on credibility, for instance. I think this is what, Andre, you were talking about in part, like building champion networks and these kinds of things because credibility is really, like, do other people you trust trust it? Right? And I think that becomes one of the ways I think of, like, when we build networks and how to share information around these use cases. It's sort of like, well, it's not enough that we just share a bunch of stuff because there's already so much, like, slob and information overload. We really want to make sure that we pick out the folks who have high credibility, the people who have sort of clout internally. That's not always the more senior people sometimes it is, but it can also be the people who really know something about a topic. Like, how do you get their ideas? Right? And and think of think of that from a communication infrastructure perspective. And then lastly, there's this thing of intimacy. Right? So we also tend to trust other people when we know sort of why they think and do the things they do, right? So when you apply that to AI, it becomes this question that we talked about a little bit earlier of, like, so how do we help people understand why the models work the way they do, not at the level of, like, what does Cloud do or Cloud or Excel, but rather sort of, like, what are the generalizable AI skills that everyone needs to build? So how do we make sure everyone has that intimacy that they sort of know when to say, Oh, I need to dig deeper in that result. Is that really true or not? Right? Or know how to break down their own tasks and know which task to delegate to a model and so on. So, like, that's sort of one way to think about it. Like, we need to build those three elements, and you can have the lots of initiatives up against each. You know, I think it's also incumbent on us as model providers and product builders to guide our users like yourselves for the right use cases. I'm seeing some, you know, chatter in the q and a of demos and tutorials. We have a lot of resources out there to share with you all, and we can certainly talk about those a lot deeper. But, you know, that's why we want to show you and orient you all towards things like data analysis before you even think about doing full end to end financial modeling and, you know, moving some text boxes and fonts around on every single page before you create a net new presentation. So I think, you know, thinking about, again, the complexity curve that I talked about before is super important. Alright. Maybe one last question on my side. You know, I think we're moving really quickly as a market, as an industry. What do you both think the next twelve months look like? What is one thing you're excited about or nervous about? Andre, maybe we can start with you. I'm actually, like I'm pretty excited, pretty bullish. I'm, like, excited about, like, all the updates coming up. I think, you know, it's we're already seeing huge unlocks, not just in terms of efficiencies, but we're like, huge unlocks in terms of, like, what we can do, what we can cover, and then how best we can build new products, cover new ground. I'm just, like, really excited about what's to come. Have AI, like I said, like, everywhere. Like, at least in your digital experience, have have it everywhere in your digital experience. Amazing. Glad to hear it. Mikhail, what about you? Yeah. I mean, I guess I'll echo some of that sentiment. Like, I don't think we I don't know what's to come. Right? All it looks like is it seems like things are getting really weird, and I kind of like the weird. Like, I think it makes me curious. Right? So there's this sense of, like, this actually unlocks new questions about ourselves that I find really, really interesting, and I feel like every time there's a new increase in capability, we learn something else about something new about how we've organized ourselves historically and how we might wanna change that. So personally, I look forward not just to technology improving. Of course, it'll be great if you can keep pushing the boundary there, but I look forward to sort of all the ways in which we'll have to reorganize and rethink what we do as a consequence. Yeah. A thousand percent. You know, I've been with Anthropic for almost two years now. I think never before have I been more excited and a little bit nervous. I think about the pace of progress on our models and product capabilities, especially for how it's starting to transform knowledge work. So I'm excited to work with all of you all in helping to educate your workforce and all of those on the call as well and really adopting these in a safe and responsible way, I would say. Are really starting to see our users to elevate what they can do with their their time. Right? The time that they're not spending crunching numbers, the time that they're not spending moving text boxes around. So we really want to expand your capabilities, as I mentioned before, together with Claude and excited to continue on this journey with you all. And I think we're just getting started here. I think that's all the time that we have for the panel. We do have about fifteen minutes for q and a, which is great. I think there's a lot of questions on the line. Alright. Maybe you can start with just a few questions that we've seen. There's a lot of questions around we've launched Excel and PowerPoint, but now what about Word? We're certainly thinking about expansions to other surface areas as well. You know, I think we've seen PowerPoint, Excel to just be really these core workflows that analysts spend a ton of time in every single day, but we're definitely open to other options and other applications as well. And there's a lot of questions also around how we think about the handoff and the trade off, right, between the desktop application and the the other surfaces. In general, I think we would love for Claude to be everywhere. Right? Claude, you know, we again, we want you to meet Claude where you are, and you're not spending time directly in our desktop application. Right? So expect Claude to be in more services around in in sort of the future. Any questions on your side? yeah. I was just, gonna say maybe a one question I saw that might be relevant to get, Mikhail and Andre your thoughts on, from Ernesto is, when leadership evaluates AI adoption, the bar is typically ROI and revenue impact, not incremental efficiency gains. How do you build the business case for embedding Claude into workflows when the value is often diffuse and hard to attribute directly to top line growth? Think it'd be useful to hear your perspectives on it, Mikhail or, Andre, whoever wants to jump in. You wanna go first? Yeah. Sure. So, I mean, you gotta, you know, you gotta also, like we you're the phrase ruthless prioritization. So you gotta be, you gotta prioritize where you're applying AI almost ruthlessly. Like, everything has to to to have a reason. Like, why are you doing it? Is is is it improving process? Is it saving time? Is it, you know, like, clearing a bottleneck? You gotta you gotta be able to answer those questions. So, like, for every project that we undertake and every sort of, businesses that we focus on, we we we try to see from from from that lens. Again, not every business to like, to to the question's point, like, every problem will have a, like, hard dollar number tied to it, but there's usually other metric. Right? Like, does it improve quality of life for employees? Does it improve time to market for clients? Does it improve client satisfaction? There's there's so many ways of how you can you can track different metrics. So but you just you you gotta be you gotta be very clear as to, like, why you're doing it and what's the benefit. Yeah. I'd agree, but I guess I tend to separate a little bit between business cases for things we're trying to explore and see, can we create value here? And then the business case when it's there because the usage is already growing within the firm, we're starting to see the uptake. In the second part, often the business case becomes so obvious. Software engineering right now is sort of obvious if you look at it for two seconds that there's a lot of value there for anyone who is working in that field. But for a lot of these sort of like early stage, like, will it fit in, how do I change the workflow, I think a big trap is looking too much at a, particularly in a near term business case, a, because there's a massive experimentation cost right now that is real and may actually slow us down a little bit in the short term. But it's back to the point I made earlier on. Like, for us at least, part of the priority here is to be at the forefront of learning, and that has a cost, and you need to bake that cost into the business case. Right? So then you'll often end up with something that doesn't look appealing in the short term even if it's very appealing in the long term. And the second piece is simply, like, because this is a constant learning approach, we don't quite know yet where the value is. So we need to set ourselves up in a way where we can learn along the way, And that doesn't lend itself to the classic business case of, like, I know ahead of time exactly what the value proposition is, and I'm going to do this massive project, and I'm going to track the value over time. I think, increasingly, value is something that we're discovering as opposed to predicting. Totally. Yeah. Maybe a few more questions around just our product road map. So, you know, things are moving really quickly at Anthropic. Right now, Claude is only an agent without a lot of context persistence beyond the immediate file it it is working on. We're actively working on this very common request. We want Claude to jump, you know, back and forth between different Excel documents or between Excel and PowerPoint and CoWork. So know that that is very much on the road map. We're excited to share more updates when we're ready. Anishra, any questions on your side? Yeah. I got a couple questions around how to think about the various ways in which Claude can generate PowerPoint. You could obviously do it through just claude.ai. You could do it through Cowork. You can do it through the PowerPoint add in. And folks, I think, had some questions around when they should be using which one. I think in an ideal world and where we're going is we would love all of these to kind of be working off of shared context. I think we're on our way there, but not quite there yet. In the meantime, the way that I think about using these tools is if you have, for example, a template or an existing deck that you want Claude to follow really closely or replicate for a different say, a pitch deck for a different customer or an analysis for a company that you wanna do for a different company and things like that, starting with the add in is gonna be a great place because Claude will work directly within the deck to make, edits. I think if you're looking for more of of a one shot kind of new generate a new slide deck for me type use case, you can do that anywhere, via co work, via, just kindacloud.com, the add in, and all of those kind of have access to the same ability to generate slide decks from scratch. So, hopefully, that helps people a little bit with thinking about when to use the add in. We found that the add in is really great for iterative, granular use cases where, you're trying to say, for example, to Claude, I want you to move this box or I want you to edit this part of the slide because it has that context and can work in that directly. Do. you know audit for what it's worth, at least for me with Excel, there's been a shift where for until quite recently, I would do most of my Excel work through the main cloud portal because it was just better, and I think that has shifted now with 4.6 where you can do a lot more work directly in the in the application. But there is, of course, I mean, a big maybe there's a product request for you guys. Like, the ability to pull on other documents that you get from being in the main cloud surface is is a real differentiator, right, like, where you can, say, update this in relation to whatever style guides you have or other memos and so on. And once you're like, with the plug in, you're more sort of focused on the on data that's all contained within the file. So I think that sort of drives two different use cases. A 100. sure. Yeah. Miguel, I think two big words and themes I'm seeing in the in the comments would be context and traceability. So in terms of context, we again, as I mentioned, we only started with a simple thing that works, which is cloud that has only access to your existing file that is open for these plug ins. But we very much want to quickly bring in things like MCP servers, local file access, and and skills as well. So all of the things that you have available on cohort, hopefully, should be coming fairly soon on our side. And then when it comes to traceability, especially for the Excel set of use cases, you kinda wanna trace the lineage of the changes that Cloud has made. I think Cloud is already pretty good at telling you its changes. And we have these, you know, citations boxes where you can click on those and see exactly what changes Cloud has made. But we want to make it easier to roll back and take a stew take a few steps to, you know, revert clause changes if destructive ones were made. So a lot of those changes would be forthcoming in the next couple of weeks for us as well. Again, all of this is really to increase I like the the three words that you you mentioned, Mikael, reliability, credibility, intimacy. I think it starts with one and two. Right? We want it to be reliable and credible because it has access to all of your core documents. But, you know, it's reliable because you can trace its responses and then be able to go back when things were done wrong. Awesome. Alright. Let's take another look here. One for you, Nick, that I saw, is from another Nick, saying, I saw an article about Opus 4.6 running for weeks straight with Cloud Code. Can you draw parallel to the capabilities of Opus 4.6 to tackle long running tasks or high volume tasks in Excel, e g spreadsheets that are five or six annual periods? How does it work in larger workbooks? Yeah. For sure. I think what I the mental model I always really like to refer to people would be the model intelligence and then the product harness we build around it. So I think of the model intelligence as really the brain. Right? We've been spending a lot of time in these past couple of weeks and months making sure Cloud knows how to build a DCF model effectively. And now it's actually starting to be fully capable of running long horizon tasks, but it also needs the arms, the legs, the fingers, the tentacles to make those actions come true. Right? And what we've done on the Cloud Code side is we've built the product harness, which is a set of agents and also sub agents. So there and and teams of agents. Right? You've seen some of those recent releases as a part of Opus last week as well. Teams of agents that can all take on smaller parts of the task and come together. We are excited to bring this architecture forward to other surfaces as well. Cowork has that as well, so that's why Cowork is so powerful today. We don't currently have it in clogged yet. It it just has one arm and not all of the the arms and legs, but we hope that that will come fairly soon as well. You know, maybe a question for you, Andriy. You know, I think there's some questions from Michael, for example. How are large banks and large enterprises getting comfortable with their sensitive corporate data, especially being transmitted to firms like ours? How do you think about the security compliance process as you adopt solutions like ours? Oh, it's a really complicated question to answer. It did literally, it took us maybe, like, two years to build the right guardrails around it and build the right framework around it. And I I would say, like, we're not fully there with every piece of sensitive data that that that we have. We are incredibly proud of of our security teams and and and how we approach this this this problem. You know, our and then our, like, data privacy is sort of, like, our number one priority. So what we do what we've done is we have 18 different risk teams, each looking at, let's say, a partner like Anthropic in in their respective ways, whether it's cloud security, whether it's data loss, whether it's legal, whether it's reputational risk, all kinds of all kinds of reviews and and and due diligence, tests that we do. And, again, like, not every single dataset gets to go. So, well, the way we organize it, we have classification of datasets, and the datasets that have been cleared can go, and we can use, anthropic models. But the ones that cannot go, then we use models that we run on premise. Makes sense. Right. I think we only have maybe time for one last question. So let's make it a fun one. If you are both to capture your feeling about AI in two words, what would those two words be? Mikael, we can start with you. Maybe that's the first word. Yeah. I would just say, oh, weird. I think that's where I'm at. Yeah. We don't disagree. We have a fun, hopefully. Andre, what about you? I I don't know. Again, I'm pretty positive. I'll say, like, loving it. Like, I'm I'm loving it. I'll ask for more. More and more and more. And I have this view that, you know yeah. Yeah. We're just more. More. We'll we'll give it to you. Alright. Yeah. There. I think that's all of the time that. we have. Vinitra, did you want us to round it out? Yeah. I think our moderator, Anne is C, back on to help us wrap up here. Everyone, thank you so much for taking the time to join us today to hear from our product team and some of your peers in terms of how, they're starting to use these very kind of, like, early frontier capabilities. I'd want to plug a constantly developing more deeper use case based, guides and demo videos and tutorials, especially as our products and models are constantly evolving. So we're gonna share those following this webinar via email. And I also encourage you to take a look at the tutorial section of cloud.com. That actually already does also have a lot of great tutorials on using Cloud and Excel and PowerPoint for various different use cases, and we'll constantly be updating that together with our Ministry of Education team. And we will also continue to run more one zero one and two zero one level webinars on these types of capabilities. So keep the feedback coming, and we look forward to seeing you again at one of these soon. So, yeah, we very much appreciate it, and I hope you all have a wonderful rest of your day. Thank you again to our amazing panelists as well. Yes. Thank you so much. Thanks, everyone. Take care. Thank you.