Excel Copilot Agent Mode: What It Does and How to Use It (Full Transcript)

[00:00:00] Speaker 1: Hi everyone, Kevin here today. It's an exciting day. This is our first live stream on the channel and joining me. We also have our first guest. We have Carlos Otero and Carlos is from the Microsoft Excel team and here on this YouTube channel and also on LinkedIn. I know our audience loves Excel. So really excited to have you join us today. Do you mind pulling your pull the mic down? Sorry. Let's see that. So yeah, Carlos is on the Microsoft Excel team. Yeah. And so Carlos has been thinking really deeply about AI in Excel. And today we're going to, we're going to dig into a feature called agent mode and we're going to start off by just looking at what even is agent mode. Then we'll look at how you could use it. Some practical examples. We'll do some demos and I think it's probably going to transform the way people work in Excel.

[00:01:10] Speaker 2: I'm excited. Like the, you know, spreadsheet AI space is just really fascinating around. There's a lot of energy. There's a lot of innovation and agent mode is really an inflection point I think for us and where you can really do with AI in Excel is really exciting.

[00:01:23] Speaker 1: Right. And the thing that I find so fascinating is I think Excel just hit its 40th birthday. That's right. So it's been around for a long time.

[00:01:30] Speaker 2: Yes.

[00:01:31] Speaker 1: And it still feels like there's so much innovation in the spreadsheet space.

[00:01:34] Speaker 2: Yeah. We actually had a party on campus like for the 40th anniversary. It was a blast. We had people from the original like crew from 40 years ago all the way to, you know, folks now and you know, I think that in the 40 years of Excel, this is maybe the most exciting dynamic moment in the history of spreadsheets really. It's just so much opportunity, so much more to do.

[00:01:55] Speaker 1: And what I think is so interesting about Excel is it's a really deep application and you have so many just millions of businesses around the world who use Excel. But I think sometimes, you know, there's so much data out there and getting insights from the data, making sense of it. Sometimes that can be challenging.

[00:02:14] Speaker 2: Yeah, absolutely.

[00:02:16] Speaker 1: And even if you are an expert at spreadsheets, it takes time to get those insights.

[00:02:20] Speaker 2: Yes.

[00:02:21] Speaker 1: And I think that's where some of these new innovations can really help a lot.

[00:02:25] Speaker 2: Yeah. And it's really interesting because I think it helps the entire spectrum of users based on what we're seeing, you know, in our testing and working closely with our customers. It can help someone that doesn't know how to use a spreadsheet really unblock the power of Excel. And it can help someone that really has, you know, someone that spent 30 years working in Excel, learn new patterns, learn new things, stay up to date with the latest and greatest, you know, ways of solving problems in Excel.

[00:02:52] Speaker 1: No, that's awesome. Yeah. Yeah. So there's this thing called agent mode. Could you give us an overview of what that even is? Like what is agent mode? What does it mean? Yeah. And I like to when so I've played with it some and I like to refer to it as beast mode because it feels like it just gives you so much power.

[00:03:08] Speaker 2: Yes. Yes. Conveniently have some slides that are prepared that we can jump in if we want to. Let's do that. Yeah. Yeah. So it's good. So I'd like to let me just project here. So just want to spend a couple of minutes, a few minutes kind of working through this and then we'll do we'll do some demos, which is I think what most people are wanting to do. But I just wanted to frame how we think about intelligence and AI in Excel, because even though agent mode is what we're focusing on today, that's not the entire part of the story. It's part of a of a larger system of AI in Excel. And we're really trying to infuse Excel with intelligence throughout its surface area. So here you see three pillars. This is how we currently think about infusing AI in Excel. And this is evolving. Part of what I mentioned earlier is like it's just a fascinating, fast moving space. So this strategy is obviously is evolving very quickly. But here's this is how we how we're thinking about the space right now. So on one side, we have copilot chat. Right. And this is you can think of this as the consistent, coherent chat based experience that you you have across the entire suite. Word, Excel, PowerPoint or WXP, as we refer to it, you know, at home at work, sometimes feels like home too. And that's the that's the same consistent experience you'll get in the M365 copilot experience as well. And this has your baseline chat experience for all your AI needs, including some Python based analytics and, you know, an advanced analysis that you can do there. Then we also have AI that we're trying to infuse into the grid directly. Right. And make sure that it's anchored to the habits that people already have. So if you're building a formula, you're working with charts, you're working with pivot tables. We want to bring AI there to help you and facilitate those flows as well. And then the third pillar is agent mode, which is beast mode, as you said, is really the AI that it's an Excel expert. It gives copilot the ability, the superpowers to deeply understand Excel right to the grid and then read from the grid in a pretty, pretty dynamic and agentic way. Oh, thank you. Is that good? Yeah. Great. All right. So today we're going to focus on agent mode. Right. And agent mode is really, as I said, it brings an expert partner, an expert collaborator that knows how to work in Excel, can help you alongside your journey and can help you build, you know, build solutions in Excel. Now, a little bit about our journey. Right. So we've been working on AI for spreadsheets for a long time, even before the current LLM wave of innovation landed. Right. And we've learned a bunch over that, you know, those those years. One of the things that we've learned is that most of the chat queries that we get in the chat experiences that we've had in Excel are related to wanting to edit the grid. Right. Commanding, formatting, building stuff, modifying stuff. So it's a really strong signal there. Now, the thing is that several years ago, we didn't quite have the technology to fulfill that need in such a dynamic way as we do now. Like the technology, the models, the AI and our understanding of how people are wanting to use these tools. It's like they've evolved to the point that we can start doing some really cool things with agent mode. The other thing that we've learned is that people, you know, appreciated that Copilot in the past explained how you might do something. And in many cases, we allowed you to insert those recommendations, but it didn't do it for you. You still had to do that. You still had to do the manual stuff.

[00:06:32] Speaker 1: I've got to say, when I when I've used Copilot in the past, I ask questions. I'm like, hey, like, how can I do this? I get the explanation. Yes. But like, I've just been itching like, hey, can you just do it for me? Exactly. Yeah.

[00:06:43] Speaker 2: Yeah. So that's consistent with what we hear. Like, no surprise there. And obviously, this whole vibe, coding, vibe, productivity is it that's kind of the future. Right. So we see this across different productivity tools, not just spreadsheets. So that's kind of a consistent thing that you'll see across across the suite for the for the core apps, Excel, PowerPoint and and Word, which also have agent modes, by the way, and behave in similar ways. All right. So a little bit about some updates and some news that we that we have. So there's three things that, you know, we'll work through some scenarios for agent mode. But I want to emphasize three things that we've announced recently. One is that it is supported on Windows. Right. We the initial release was for Excel on the Web, but it is now supported in Windows. It is now we have Web search, which means you can go out and fetch publicly available data and augment your analysis. Yeah. And then the third is that we support Claude or Anthropic models, which is really exciting. We have the best models from, you know, from OpenAI and Claude and Anthropic powering agent mode.

[00:07:50] Speaker 1: So when you're when you're interacting with agent mode, you could go in and choose the model that you use. And so you could get a feel for, hey, which model they prefer and what situation. Exactly.

[00:07:59] Speaker 2: Exactly. Exactly. So that that's, you know, really exciting just within the same experience. You don't need to go install something else. It's all right there. Now, availability specifically. Right. So in Web, actually this week, yesterday, we went to GA. Good timing. Yes. Perfect timing. I was, you know, like it's hard to predict exactly when those things are going to land because a lot of pieces have to come together. But I was really hoping that I would come here today with those news. So now it's a gradual rollout. Right. Like usually we don't we don't turn things on 100 percent right away, but it's gradually rolling out to production, to general availability in Excel for the Web. So you can go to Excel for the Web and use it now if you have the support to compile licenses. The second piece is that Windows is available in the beta channel. That's the Insiders program. And it's it's the capabilities in Windows also. I mean, actually, both Web and Windows, we have this program called Frontier, which is a early access to AI capabilities for all of Microsoft. So there's certain capabilities for agent mode that are in Frontier, like Cloud, for example. Cloud is something that we just announced and that's only available in the Frontier umbrella. And we'll test it. We'll iterate. We'll prove it. And then we'll we'll move it forward to GA. Got it. Yeah.

[00:09:10] Speaker 1: OK. Yeah. Oh, very cool. Now, this is this is really good background on it.

[00:09:15] Speaker 2: Great. So do we want to jump? Oh, actually, one thing I wanted to note, I also have a technical I have a technical diagram here. I don't think we need to spend a lot of time here right now. Maybe if we have time at the end, if people have questions, we can go in there. Why don't we come back to how agent mode works? But I wanted to go to just kind of browse through this real quick. This is the main support article for agent mode in Excel. And we just updated it yesterday, you know, as part of the GA launch for Excel for the web with all the detailed information about availability and status and what to expect. Right. So this is all the support information. There's some important pieces here to note. It does require a copilot license. So right now it requires for an enterprise M365 copilot license or a premium subscription, which is on the on the consumer side.

[00:10:03] Speaker 1: Got it. So as long as you have a copilot license either in the enterprise or a personal copilot license on your subscription, then you should have access to this.

[00:10:11] Speaker 2: Correct. Correct. And that will give you access to all of the GA kind of capabilities.

[00:10:16] Speaker 1: There's also information in a GA generally available, generally available capabilities.

[00:10:21] Speaker 2: Now, if you scroll further down, you'll have instructions on how to use it, how to how to invoke it, some some example prompts, what to expect, how to you know, how to different knobs they have. But I want to call out a section down here, which has all the information about the Frontier program that I mentioned, how to join it and what features are available in Frontier.

[00:10:41] Speaker 1: So for example, this this Frontier program, that's for organizations or enterprises to make the like feature sets available to their employee base early.

[00:10:50] Speaker 2: It's also also a consumer. It's both. Got it. So here's exact information on how to access the program. And it gives you access not just to stuff in Excel, but across, you know, Microsoft, like early, like all the cool stuff we're doing with agents like the analysis agent. And I mean, a lot of those are GA already, but some of the leading edge AI work you.

[00:11:14] Speaker 1: So if you want to be on the bleeding edge, kind of the latest tech coming up, you should opt in to this program. And is it as simple as saying like, hey, I have an account and I want to opt in or?

[00:11:22] Speaker 2: Well, for consumer, it tends to be that case for enterprise. It's a little more. Yeah. OK. You have to work through your IT department, but IT admin.

[00:11:29] Speaker 1: So so make sure you get a lot of cookies for your IT admin. Yeah. Work it out.

[00:11:32] Speaker 2: Yeah. If we have any folks on the IT, you know, IT folks joining today, like please take a look at this. If you want to be in the bleeding edge, really a Frontier kind of our company. Very cool.

[00:11:42] Speaker 1: Cool. Yeah. No, that's that's great background on agent mode and kind of the AI capabilities. But I think what our audience loves is actually seeing it in action and kind of seeing us get our hands dirty and kind of test it out. Yes. Let's do it. So let's let's jump in some demos. And one thing to note, most of the videos that we post on this channel are produced and we know we make sure it's kind of the happy path or a nice flow with live demos. You know, we'll have to see how things go. But I think we the demo gods should be on our side today.

[00:12:13] Speaker 2: I was joking about demogods. I was joking with my son, Max, earlier today that I really wish demogods are with me today. And he was like, you mean the demigod like Maui from Moana?

[00:12:25] Speaker 1: Right, right.

[00:12:25] Speaker 2: I was like, yeah, yeah. Demigod and demogods. I hope they're both they're both on my side today. So, OK, let's jump right in. I want to get right into it and show the three things that I noted are new that we are we're kind of we've announced recently. One is desktop Windows. We're in the desktop Windows app right now. This is a Windows app in all its glory. Right. Yeah. Then Cloud and Web Search, which are the three the next three things. So the three things I talked about earlier. So to get into agent mode now, it's all part of it's all part of the Copilot umbrella. Right. So to get into it, you go into the home tab and then you open Copilot. Now, I have to be honest, this is one area that you should anticipate evolution. Right. We are working on we're exploring different ways of making it easier, more discoverable, kind of adjusting the the way people discover and fall and leverage these capabilities. But this is what it is today. Right. But you should anticipate some some changes on the on the details and that support page will have all the latest.

[00:13:27] Speaker 1: And then a lot of that is just seeing kind of how people interact with it. Are they finding it? Are they? Yeah. And then kind of making adjustments based on that. Yeah. So kind of taking that user feedback to it. Yeah.

[00:13:36] Speaker 2: Yeah. And there's a lot of brand new patterns that we're we're inventing here, which is exciting. Right. But that requires us iterating and learning from from the feedback. So, OK, so now this is the chat experience that I mentioned earlier. This is the the same consistent experience across WXP, Word, Excel, PowerPoint and the M365 portal. Let me collapse this. And then to get into agent mode specifically within the tools dropdown, the first one will be pinned as agent mode so that you will always see that as your first option.

[00:14:06] Speaker 1: Now, that basically enables agent mode now. Correct.

[00:14:08] Speaker 2: OK, so that gives Copilot essentially the like you can think of it as giving it edit rights for your workbook. Right. It enables the capability, the smart, the Copilot needs to read and write agentically or autonomously on your behalf on a workbook. Right.

[00:14:27] Speaker 1: OK, so then I'll note So this is kind of equivalent to going to your coworker down the hallway, bringing him into your office and saying, hey, can you make some edits to my document?

[00:14:36] Speaker 2: Yeah. Or maybe just giving them a link. Right. Right. Right. In modern days, you would send them a link where you share the workbook with them and give them right access as opposed to read read access. And then they can edit your workbook. Right. Or even you can collaborate with them. So you can think of that that way. OK, so then we have Web in this button here and then you can you can toggle it on and off. And then up here.

[00:15:02] Speaker 1: And so just to clarify, so Web gives it the ability to go out to the Web and fetch data that's kind of externally available. Correct. Yeah.

[00:15:09] Speaker 2: So that's a good note that so agent mode currently only has access to the context in the active workbook.

[00:15:16] Speaker 1: Got it. OK. Right. Outside of search. So when that's toggled off, it's just looking at the workbook, turn it on. It can look at the workbook and go out and find data all over the Web. Correct.

[00:15:25] Speaker 2: But it doesn't have, you know, security is core, like it's critical part of what we're doing here. That is we spend so much time with the security teams. So there's a lot of really nuanced, tricky, technical security angles to to agent mode. So currently, agent mode doesn't have access to, say, creating your workbooks or look at your other files in, you know, or look at other things in your computer. It doesn't have access to your computer. It only has access to the workbook that you're looking at.

[00:16:00] Speaker 1: OK, so that would include if you have multiple worksheets and whatnot, it could it can look across multiple worksheets within that workbook, including other worksheets within. That's right. OK, correct. Got it.

[00:16:09] Speaker 2: OK, the other thing I wanted to note is up here, there is a drop down to select different models. So I'm going to and this is where this is what I was talking about earlier, where you can pick OpenAI models, you can pick Anthropic models.

[00:16:21] Speaker 1: And this is the Frontier right now. Claude specifically is Frontier. Right.

[00:16:25] Speaker 2: You'll see here a little disclaimer. So I am running and this is part of the why the I hope the demo gods are with me. I'm running like some internal bits here that has everything turned on. Right. Even some stuff that's not fully released. So you'll see I'm using Frontier features and I'm using Claude. Got it.

[00:16:41] Speaker 1: And then just kind of out of curiosity, from your own experience using this, what what do you see is like the pros and cons of the different models? Or is it just, you know, you try the agent mode with one model, then you try it with another one and kind of see the results. Like, are there any strengths to one model versus another?

[00:16:56] Speaker 2: Yeah, that's a good question. And I would love to get feedback from the community on that. Actually, they have different tendencies, you know, like some one of them tends and I want I I would I would like not to bias too much and like to get some, you know, some information from the community here. But you can see differences in how, for example, how it uses Excel artifacts, like how the types of formulas that are right, the formula patterns that it uses, the types of formatting that like some of one of the models does more, you know, a little bit different layout and a different, more colorful layout. So it's there's differences and nuance on how Excel is used, which kind of mimics personal prep, like there's no right or wrong for a lot of these things, right. So it's kind of hard to tell which one is objectively better, because it applies Excel in different ways. Got it.

[00:17:48] Speaker 1: So it almost sounds like experiment with the different models and see what you have a preference for. Got it.

[00:17:53] Speaker 2: And to be honest, also, we're still we're kind of learning about this ourselves, you know, as we as we play around with the models ourselves as well. Okay, so let's run let's run a scenario here. So we actually have a question here.

[00:18:07] Speaker 3: So real quick, we have a question from GNS Global Solutions. Hopefully the chat can hear me. Okay, he has he says he has a co-pilot license, but he does not see all your showing in the in the software. He needs some insights. So basically, they're not seeing the same things that you were got you guys.

[00:18:25] Speaker 2: So there's a couple of things that could be happening. It's only available in GA for the web, right Excel on the web to so and we just started rolling it out yesterday. So it might be the case if she's trying this in Excel for the web, it might be the case that she's in the proportion of users that haven't gotten to bits quite yet, right? Because as I mentioned earlier, we roll things gradually. So just be patient in a couple of weeks. It should be we anticipate that we'll, you know, we'll reach 100%. It could also be it could also be that IT and her company has some of these things turned off. There's a part of the security and compliance kind of requirements that we have and which is incredibly important, right? To give users and companies control of when and how they want to roll out the latest and greatest technology. So it might be that IT has some some stuff turned off. It could be that she's trying Windows, right? And Windows specifically you have to be in the insiders audience to which is should be accessible to her as well. So it's kind of a tricky question. I'd need to know more information specifically about her environment too, but it sounds like it's starting to roll out.

[00:19:38] Speaker 1: It's becoming more available to more and more users. So with time more people should have access and trust me.

[00:19:44] Speaker 2: We are like we're moving as fast as humanly possible to try to get this stuff out. So please be patient. But if you if you try it on Excel for the web that you know, I think that that's your highest chances the best bet.

[00:19:58] Speaker 1: So yeah, exactly. Okay.

[00:19:59] Speaker 2: All right, cool. So I while we were talking I ran this prompt, right? So I in this is intentional to trigger web search and it's using Claude, right? So I asked it to pull the top 10 gold producing countries and the most recent year for each country include production volume. I create a table has some very specific data requests and some very specific Excel requests, right? So you can see that as we were speaking it added all this information. This is a table. I do love that one of the things that we that's important for us. For instance, you can see here. Actually. No, I don't want to add a new one. I want to see the existing conditional formatting manage rules. You'll see here that this is actually a live conditional formatting rule, right? So this is just to illustrate like part of what we are attempting to do here as well as that the artifacts that agent mode inserts are refreshable are live are auditable are editable by the user right now. This was a web request. So you'll see in the chat, you know, it has information about what he did. It has specific sources so you can visit these websites and then you'll see also has all the sources that it scanned through to to kind of find the answer, right?

[00:21:14] Speaker 1: Got it. And so this was with web turned on. So that's why it's going out to all these different sources and kind of pulling this information.

[00:21:19] Speaker 2: Correct.

[00:21:20] Speaker 1: And those like at least at a glance there. It looks like those are high quality sources of information.

[00:21:24] Speaker 2: Yeah, and obviously you can nudge it, right? So here it was generic. I didn't specify a particular source, right? But you could say use this website use this table. Here's the URL go fetch the data from here. Got it. So it can certainly do that and then you'll see you'll see the information reflected in the you know in the chat. So now here I run another one just to illustrate it's it's it's a conversation, right? So you can you can have multiple turns you can you can adjust the results and here I asked it to using that data right to create a pivot table and aggregate it by Geo grouping, which is one of the columns I pulled here content. I mean, you don't have to use the exact same words because it's it's AI. It knows right? Yeah, so it used that Geo grouping the continent field to pull this data and it also created a chart got a pivot chart in this case. So again, this is all refreshable.

[00:22:19] Speaker 1: So this is a pivot table. So it all ties back to the data on the previous sheet. And so if that data updates then this update this view. Yeah. Yeah. Yeah.

[00:22:26] Speaker 2: So this is kind of what I was mentioning earlier, right that it goes to the full spectrum like I still vividly remember the first time I learned how to use a pivot table and it was magical for me. But before that is it was like until it was so hard like I couldn't even I didn't even understand what I would use a pivot table for until I constructed my first one, right? But for a lot of people that's just like so hard, right? So now you don't you don't have to learn it. You don't need to learn the specific steps. You can get the value from it. Now. I would still recommend learning how to use it. There's a lot you can do manually as well. That is really valuable. You can augment you can partner with the AI and modify it to specifically do something maybe more nuanced that you want to do. But but it can get you like most of the way there.

[00:23:11] Speaker 1: Yeah. No, I think what's always challenging is there's so much data out there and you have questions about it. But then kind of what adds friction is you have to know how to use these different tools like you have to know about pivot tables. You have to know how to set them up. You have to know about fields and then that I think that blocks a lot of people from getting insights from their data. So what's what's really interesting here is you just asked a question and then here, you know, you get this view that shows you exactly what you asked about. And so it almost abstracts away the tools and kind of the steps that you have to go through to get it. And instead you ask your question and you get a response. Yeah, that's I think it it opens up kind of the power to a much broader audience.

[00:23:54] Speaker 2: Yeah, exactly. And in this case, you know, in this example, I started from scratch, but you can envision that you already had the data in here, right? That's why I did two steps. I can imagine that you already had this data. Someone from your team sent it to you and you can augment it. You can analyze it again with Excel native artifacts and objects that are refreshable or connected that everyone in your company knows how to deal with already.

[00:24:13] Speaker 1: But I even look at this like, you know, setting up a pivot table. That's one thing you'd have to know about Excel. Setting up conditional formatting. That's another thing you'd have to know. And here it kind of simplifies that whole process.

[00:24:22] Speaker 2: And then the web search, right? To identify this and bring it in.

[00:24:24] Speaker 1: And then even for people who say know how to do conditional formatting or pivot tables, this basically streamlines and accelerates that whole process. Yeah.

[00:24:32] Speaker 2: And it's compound, right? Because I actually could have put, I could have put all of this in the same prompt as well. Right? And that's kind of the beauty of this too. Like I am finding that as I use it myself. Because now, like anytime I get a spreadsheet that someone sends me at work, I try to use agent mode, like exclusively to, you know, it's a funny example. We have a whole oof tracker. Like when people are on vacation for work that, you know, every quarter we send it out and people need to manually go in. It's just so we can track who's in and out. And so we can plan for, for capacity for different things. So last, you know, a couple of weeks when they send it out, I was, I'm not going to do this manually because there's instructions. It says, if you're out, put a one here. And of course it has conditional formatting and sophisticated workbook. So I just use agent mode. It was just like, fill this out for me. I'm going to be out the week, you know, this week in December, this week in December. And it did it for me. So I, I just gave it the instructions. And, and, and in fact it, it was able to parse the instructions in the workbook because as you know, like people often put instructions on how to populate, especially in the workbook itself. So yeah.

[00:25:36] Speaker 1: And then, and then what kind of like in that example where you're filling out this, you know, vacation tracker, it's going in, it's making changes to your sheet. And I'm sure there are probably a lot of people who are, are a little apprehensive about like giving access to the sheet, having it make changes. Is there any kind of, is there a way to undo or step back if you don't like the change that it made?

[00:25:55] Speaker 2: Yeah, that's a great question. That's, that's a really great question. Yeah. Of course you can undo, you can undo what the agent does. Obviously there's also workbook history that you can rely on. Like Excel has really sophisticated kind of version history that you can go in and go back. We also have change history, right? So in, in workbooks you can go and see the changes that you made or your collaborators made or the agent made. So you can go and kind of undo that and can track. We don't have some of the, I guess, how do I say, that's a space of further investment for us, for sure. So like currently you can undo, but that's an area that we're actively kind of looking at. How do we make it even easier to understand exactly where the agent is working? What changes it made? But currently you can, you can undo, you can undo and you can rely on Excel's built-in tools to kind of, you know, undo stuff and go back to a prior user. Got it.

[00:26:53] Speaker 1: Yeah, that makes sense.

[00:26:54] Speaker 2: Yeah. Great. So that's, that's the demo in desktop. I wanted to now go to a few other real case scenarios as well. Do we have any other questions or should I jump onto that?

[00:27:08] Speaker 3: Hold on one second. Yes. We have two other questions chat. We have one from Jose. I think it's Jose. Hopefully I'm reading all these right. But is agent mode power query or sorry, could agent mode and power query be possible? That question.

[00:27:27] Speaker 2: Could it be possible? Absolutely. Yes, this is software. Currently, no, currently there isn't. So power query. So power query from within the agent mode UX is not connected. I do believe we have some AI within the power query experience itself. But currently what I'm showing here is not connected to power query. But that's a great one. I love that question because that's definitely on our, on our vision, right? To continue to get deeper and deeper into some more of the kind of advanced capabilities in Excel and facilitate, facilitate using that through agent mode. So you can think of agent mode currently as having, it's a little bit more integrated into the grid, right? So formulas, tables, pivot tables, charts, conditional formatting, shapes, layout, formatting. So all of that agent mode can access. We haven't quite tapped into power query or power pivot or some other capabilities like that quite yet. Great question though.

[00:28:28] Speaker 3: Great. Awesome. And then there's one more, another question from GNS global solution. She's asking, can this agent allow relational or relationship data, like how access work?

[00:28:39] Speaker 2: That's actually related to the prior question, I think. So the way relational data scenarios are traditionally supported in Excel's through power pivot, right? Like power pivot uses the same, the same engine that power BI uses, like the, the analysis services engine or vertipack. So that's actually the tool that you would use traditionally in Excel. If you have a gajillion rows of data, you load it into the data model, you draw relationships between multiple tables. Currently it's not supported. Agent mode is not supported there, but that's kind of along the same lines of what I was talking about earlier. There's, there's a lot for us to go into in the future and more sophisticated areas. Yeah. Yeah. Great questions. Oh no, we can jump into, yeah, we can jump into some of them. So here I have, depending on how much time we have and kind of how the conversation goes, I have a few different scenarios queued up that we can explore.

[00:29:34] Speaker 1: Let's do that. Yeah. And so we've transitioned now. So we were in desktop. Now we're on the web app here. Correct. Got it. Okay.

[00:29:40] Speaker 2: Yeah. So this is Excel for the web on the browser. One little tip that I love. There's a, there's a URL. If you do Excel.new, it opens a blank workbook for you in Excel. So this, you know, like a little tip there that I love just to create quick, you know, quick, quick workbooks. So, but let's, let's trigger this flow here and then we'll explore some different scenarios here. So here I'm asking it to, from scratch, build a discounted cashflow model, which is kind of a classic financial model thing that, that you see in finance. Right. And I've asked it to, you know, there's some assumptions here, historical data. I want discount rate of 10%. So it's kind of a compound sophisticated thing, but that actually was faster than I was anticipating here. So you see, as we were talking, it kind of created it. Right. So this is kind of a little bit more of a real world scenario. And you can see it's still working. It's still iterating. It's still adjusting, but it's kind of putting results as we speak. So you see the progress as it's happening. You see the progress as it's happening. Right now, I, as I mentioned earlier, I used to work in finance myself and I had to build a few of these back in many, many years ago. So I know how to build these from scratch, but like this took a minute, less than a minute. Right. And this is all like should be connected, right? It has formulas. It's all kind of, you know, it looks pretty, pretty reasonable, pretty coherent, all alive. It's still working. So it'll probably like that's the beauty of agent mode as well, that it, it iterates, right? It will try something. Maybe there's an error. Maybe it got the syntax wrong. Maybe that layout is not right. So iterate until it feels good about it, fulfilling the request that you, that you sent.

[00:31:34] Speaker 1: Yeah. And then I guess having, having worked in finance before, how do you, when you, when you look at the results that you get here, how do you feel about like, I know one concern with AI is just accuracy of results. Yeah. Like how do you feel about the results? Like, does it feel like it's generally spot on with what you would have done when you were in finance?

[00:31:51] Speaker 2: Yeah. I mean, I have to be honest. It's not perfect, right? And that's why it's a copilot, right? Ultimately the user is the pilot, right? And you, you're accountable and responsible for the output that it generates. But we've designed it in such a way that it, it, it tries to solve the problems through verifiable results in Excel, right? And it has access to both input and output. So it can, it can reason over the actual output. It can reason over the syntax of the formula. If there's a formula error, it can correct that. If it says a hash value, hash in a, oh, that's weird. Let me go sort that out until it finds a. So from a numerical sense, like it's pretty, you know, it's pretty accurate. Now from a, from a stylistic sense, maybe it's, it uses a different formula that I would have used, right? It uses a different pattern, but that's also kind of interesting. And here actually I was, I changed tabs as we were speaking. These are a few other ones that I had queued up from before, like this kind of cashflow model. Can you get the similar, similar idea? These are ones that I use the exact same prompt earlier. In some cases it added charts. So it's all kind of your assumptions and your formula. So it's all, you can see like multiple runs give you kind of consistent results to some extent. Not, not exactly the same, but kind of consistent. One of the things that I, that I've loved in our user research that we've done is working with Microsoft Finance. Like at Microsoft, we have like a really sophisticated Excel user base in the, in the finance department. So we work closely with them and we were meeting with some folks in finance that have been using Excel for like 20, 30 years and they were doing some modeling scenarios. And one of the people we were working with was surprised by the formula pattern that agent mode used. It's like, oh, like that? I didn't even know you could use, I think it was using let or even lambdas. I forget, I forget the exact pattern, some newer formula patterns. And they were like, oh my goodness. Like I can, I can, I've been doing this model for 20 years, but this formula that agent mode use is a lot more effective and efficient than the way I would have done it.

[00:34:02] Speaker 1: So you almost learn by the output that you get from agent mode. You look at it and it's like, wow, that's a more efficient function than I could have used or arrangement. Yeah.

[00:34:09] Speaker 2: Yeah. And actually in my experience, when I remember when I was in finance and I was learning how to use Excel, that's a lot of how I would learn. Like I would, I would try to deconstruct workbooks for my colleagues and like look at the patterns, right? Oh, like I could apply that. Like that's a much cleaner way of solving the problem that I was trying to solve. So you can use agent mode to teach you new patterns as well. Right. You can ask it, Hey, look at my model. How would you improve it? Like, look at this formula. How would you make it more efficient?

[00:34:38] Speaker 1: Right. Right. Yeah. And then even taking a step back. So I know a lot of people who are kind of seeing agent mode for the first time. How do you recommend that they get started with it? So like here, like some, some of these prompts are fairly, like, you know, this is a fairly complex involved prompt, but how do you, like if someone's just kind of, you know, touching it for the first time, how do you recommend just experimenting with it and getting started?

[00:35:03] Speaker 2: Yeah. Yeah. So one is in the support page, we have a few suggested prompts so they can, you know, they can look at that and, and, and leverage that to watch this recording and look at some of the prompts. Like, you know, we, I'm happy to share, I'm happy to share some of the prompts I use today and even some of the workbooks that we have for people to play around with, but I would say just explore it. There's even ask agent mode. How can you help me? Like one of the, and I don't have it in this flow, but one of the prompts that I really like is just explain what your role is and what your job is and ask agent mode, how might you help me?

[00:35:37] Speaker 1: Right.

[00:35:37] Speaker 2: I work in chemistry. I have this model where I have to predict some, whatever chemical compound, how it might react. Cause people use Excel for everything, right? Right. Right. So how might you help me build five sheets in this workbook with examples of how might you help me be more efficient in my, in my work? Yeah. I love that one. That one creates like some really cool stuff.

[00:35:59] Speaker 1: Yeah. Yeah. Oh, that's very cool. And, and so it sounds like in that case, agent mode is also, it could be conversational where you're getting feedback and so it doesn't necessarily have to go in and edit the sheet.

[00:36:12] Speaker 2: Precisely. Yes. You can, you can, I mean, it's, this is where the, the. M365 chat versus agent mode kind of overlap comes into play a little bit. If you definitely do not want Copilot to edit your workbook and your scenario is pretty clean cut. Just explain to me how I might do this. Uh, you might want to use chat, but if you want, you know, Copilot to edit your workbook, then you use agent mode, but agent mode is also capable of just answering without editing the workbook. Right. For sure. And then when, yeah. Yeah. So like I have a few other scenarios just to kind of work through just for the sake of time, I'm not going to run all of these live, but I did run these previously, right? So here's one example of market analysis, which is a common scenario that we see. Right. So I ask it to aggregate some data for EV market. I want forecast. I want EBITDA margins. Just this is, I mean, you can customize this to your business needs, right? But here I'm asking for it to create a, a pretty thorough market analysis, pulling data from the web. Uh, and you can see it has data for Tesla, BYD, Rivian has historical information, sensitivity, analysis, charts, key findings. Like it's pretty extensive.

[00:37:27] Speaker 1: That looks really thorough.

[00:37:28] Speaker 2: Yeah, really thorough. Right. Yeah. So I, again, I could have done, gone manually to pull all these things. And since this was a web, uh, you should see the same kind of sources, like everything is cited. Right. So you can see where it pulled the data. You can even, one of the things that I love, and here actually you put sources here, so it must have, this is what I was going to note. You can ask it to, to note the sources in the workbook as well.

[00:37:51] Speaker 1: Oh. So when you see the data on the sheet, you can tie that back to where it got here, here, here, here it is. Yeah.

[00:37:56] Speaker 2: Right. So I think in this case it did that automatically. It didn't have to ask it. So it puts the sources in the workbook itself. So if someone else, uh, reviews the workbook and doesn't have the chat history, they can also see where the data came from. Right. Right. Yeah. Yeah.

[00:38:09] Speaker 1: No, that's, I mean, that's incredible. I mean, this, like just thinking about how much time it would take to pull together a view like that. Yeah. And now you're getting it in just minutes. Totally. That's incredible. Yeah.

[00:38:18] Speaker 2: So here, let's shift gears a little bit. So the examples I've shown so far are ones where we start from scratch, right? With, which is valid. A lot of people create new workbooks every day and like they have to do new analysis and build new models, but obviously there's a lot of cases when you start with data already in your workbook. Right. So here's one scenario, um, where that's the case. Imagine for a second that I, you know, I manage a team or a company that has different departments. Right. And they've all submitted data for the year for the quarter in this workbook. All of this data is different. It's laid out differently, different headers, but I have data from sales, marketing, finance, inventory stuff, support and whatnot. Um, so here I have a completed view where I asked, here's the prompt, right? So you can see kind of the history, but I, I asked, Hey, I have some data, the different teams I managed have submitted for me. Uh, can you build a report? Can you summarize key trends? Can you build a dashboard? Can you highlight, uh, key, key insights? Right. And it added these two sheets, right? So you can see this is the same exact workbook I was showing earlier with the data, that's the existing data. And then at the end, as I ask it, it add two sheets, one with a summary. Uh, and these are all pulling, so you'll see the formulas, right? These are all life. This is pulling from the sales one. This is pulling from the marketing tab. This is pulling from the inventory top. So this is live refreshable as the data updates. I think it has conditional formatting, like pretty cool layout. And then it built a dashboard and has charts and I don't like again, like, and it's not perfect, right? This is, this is one piece that it's not perfect. Like you can see here, actually, no, I thought there was going to be some overlaps, but yeah, there's some overlaps here. Actually. I haven't changed this since this is all raw from the agent. Right. But there's some overlaps. So it's not perfect, but I can easily move things around. I can, I can augment it. I can change the layout. I can move things. I can even ask it to do that.

[00:40:11] Speaker 1: You can continue, I guess, refining it with the agent. Correct.

[00:40:15] Speaker 2: But I mean, the truth is that certain things are just more efficient. If I need to move a chart, just move the chart.

[00:40:19] Speaker 1: Yeah. Yeah. Right. You know what, one thing that's interesting is so on, on the YouTube channel, uh, one of the most popular videos is how to build a dashboard in Excel.

[00:40:28] Speaker 2: Yeah, of course.

[00:40:28] Speaker 1: And, uh, cause I mean, so many people have lots of data and they need some way to visualize at a high level what's going on. Uh, and, and seeing this where you just have multiple worksheets and you can just pull together a quick dashboard and then you just tweak it and, uh, and, and the fact that it's live. So if you bring in additional data, it just updates.

[00:40:43] Speaker 2: Yeah.

[00:40:43] Speaker 1: Uh, that, that is pretty neat.

[00:40:46] Speaker 2: Now let's see, I have a couple more here. Maybe this session expired. Let me go to this one.

[00:40:52] Speaker 1: Yeah. And, and so this is all, this is all live, right? It's all live. Yeah.

[00:40:55] Speaker 2: So as you see, I have a bunch of these workbooks queued up and this is all in the browser. So this session I think expired.

[00:41:02] Speaker 1: And there was the, the new Excel logo.

[00:41:04] Speaker 2: I have the new Excel logo. Yeah. So I guess this, this demo might not work, but here I can just explain, um, and I can share this separately. I don't know if you can, uh, if there's a way to share kind of stuff after the fact, but, um, in this one I have, I have a demo where I ask it actually to teach me a formula pattern in Excel, right? And actually we can, let me, uh, let's see, let's do this. Let's be, let's be brave. I'm feeling brave today. We can do this. I'm going to grab my prompt for that.

[00:41:35] Speaker 1: And I saw, I think when I saw the prompt, it, it was how to use, was it X lookup? X lookup. So a lookup function. Uh, and I know like, I guess the, uh, the predecessor, uh, V lookup and then X lookup. And I know that's a very common scenario in Excel where you have, you know, two different lists of data and you have to make a connection between them. Yeah. Uh, and, and sometimes it's a little confusing of how you do that.

[00:41:56] Speaker 2: Yeah, exactly. X lookup is a good one, right? It's just, it's just. X lookup is so great. It's so much more flexible. Yeah. I've always been in the V lookup camp, suppose the index match camp and V lookup I think has a little bit of both in there. So, but, but most people like are not keeping up with, you know, like they're not going to Excel training. I mean, they, they do. There's a lot of great, the community is awesome online. We have all the MVPs and all the, all the community that put great content, but, um, now you can also leverage agent mode to help you help you with it. So let's, let's run that in the, on the side. We can check back in with that, uh, in a little bit, here's another scenario that I have this very common in finance, um, kind of accounting and reporting. I'm doing a variance analysis here. So I asked it to compare data between budgets and actuals and generate kind of a absolute variance, a percent variance, and then a status based on some rules that I have. Um, again, this is all from scratch, right? It created this report. It created the layout. It created the formatting. You can see like that's subtitles. I nicely formatted the total. And of course, if you have specific patterns at your company, you can explain that in the prompt as well. Uh, but it just created all of this from scratch. Uh, here's another one that I, that I love as well. I asked it, um, to actually, let me see if the, no, I lost the chat history on this one, but in this one here, I asked it on that same workbook to highlight. Uh, let's see. I asked it to go through this variance workbook and I, before I had seated some formula errors and I asked it, put a comment wherever you find some discrepancy or a formula error, don't fix it yet. I want to review it. I am, I want to take a look to see if it's legit. And it did that like, and it added this comment, right? Uh, suggested fixed to this. There's a mismatch. So if I inspect the column, the formula, I look at that one above. It's doing a subtraction. I went above subtraction and then it's not, it's a pattern that's off. So I, I planted a few formula errors here intentionally. Here's another one. There's a mismatch. This, you know, I have a minus one. The formula is different, right? So it can also help you audit either your workbook, someone else's workbook, which I love like this whole error, error identification space. Uh, yeah, it's just, it's great for that.

[00:44:32] Speaker 1: So it seems like kind of building skills in prompt writing are probably very important, right? Like if, as long as you phrase an effective prompt, you can essentially have it do whatever you need it to do. Yeah, exactly. Yeah. Yeah. It has access to all the different tools within Excel.

[00:44:45] Speaker 2: It has access to all of the prime, you know, the, the, the most common, the most used tools, as we talked about earlier, maybe Power Query, some of those things are not available. Yeah, that sort of thing. Yeah, exactly. Comments, shapes, text boxes, you know, charts, pivot charts, tables, pivot tables, conditional formatting, slicers, like all of that.

[00:45:03] Speaker 1: It has access to all of that.

[00:45:04] Speaker 2: Has access to all of that. Got it. Okay. So here, here it kind of finished as we were speaking, the, the scenario, right? For the XLOOKUP. So it has a full sheet with, let's see. Oh, there it is. Okay. There's some actual live examples, right? Okay. So it has, it added in a matter of minutes, three examples with live XLOOKUPs that are pulling the data with the right syntax, and then you can kind of inspect that.

[00:45:33] Speaker 1: And so, so it's, it's really almost like a learning tool too now, right? Exactly. You can have a build a sample spreadsheet. You can look at the, you know, look at the formula, see how it's working and test it out for yourself too. Yes, totally.

[00:45:44] Speaker 2: All right. I have one more. Let's finish up with something fun.

[00:45:48] Speaker 1: Let's do it.

[00:45:48] Speaker 2: So, so I have to say this is, I've been at Microsoft for a long time and the Excel team for a long time. And this is the first time that I've worked on anything that my kids have gotten excited about. So in the early days, and this was kind of one of my bars I had internally for knowing when we were ready to ship is whenever it could produce art.

[00:46:07] Speaker 1: Was that the true goal behind agent mode to make yourself cool to your kids?

[00:46:11] Speaker 2: Well, it just kind of happened. Might have been part of it. Yeah. So, so I have, I have a couple of prompts here. I'm going to run one live and then we'll explore the gallery. But last night I asked Max and Luca, actually, let me, I'm going to run it in. I want to run it in Claude. I've been testing Claude heavily since that's kind of a new thing. So I asked Max and Luca last night, my kids like, I'm going to do a live demo. Hope Maui, the demigod is on my side. What art would you like me to showcase? Right? So they wanted to do a modern painting on one sheet and then a marble painting on another sheet. So we'll let that run and we'll come back to that.

[00:46:54] Speaker 1: Cause I guess in a sense, like you look at Excel, you have the grid. Those are kind of like pixels, right? Exactly. You can manipulate that to create art. Yeah.

[00:47:02] Speaker 2: And it's just illustrates the point, like Excel is such a flexible tool for creativity, really like to solve problems, but people make literal art in Excel as well.

[00:47:11] Speaker 1: I've seen some incredible artwork that people have made in Excel, just using the, just the grid and changing the color of each one. Yes.

[00:47:18] Speaker 2: So here's some, some results from the prior run. So here's the, I think this is the abstract one from Max. This is all agent, right? Here's the marble one from Luca. Cool.

[00:47:30] Speaker 1: Interesting.

[00:47:31] Speaker 2: Yeah. Here's another run where we did the marble run. No, this is the modern painting. Uh, and of course I didn't specify here. Here's the one I just ran here. I didn't specify anything specific. I can, I can say, Hey, use a 1000 by 1000 grid. Right. And it's going to be a lot more detailed. Yeah. So I just did this for expediency of something simple. Uh, here's another one where it's like marbly. So I don't know. It's just kind of fun, but I just love, it's just to illustrate also the point that it has access to like, just modify the grid in very flexible ways. And we're still discovering what all it can do. And that's where like, I'd love the community's feedback, uh, you know, to help us kind of iterate and make it great and make it better.

[00:48:12] Speaker 1: And on the topic of feedback or questions, uh, I wanted to check if we have any questions from the audience.

[00:48:19] Speaker 3: Yes, we do. We have quite a few actually. Um, so I'm going to try and run through these as quickly as possible. Um, so currently I use agent mode in Excel and web. There's a couple of people asking this question too. Uh, but I still don't see the option to select Claude as a model. Is it because of it restriction?

[00:48:35] Speaker 2: Claude is frontier only for now. So you might see agent mode in Excel for web, uh, with the, where we released GA, which includes kind of the baseline functionality and web search Claude. It's kind of early access frontier program only. So if you're a consumer, if you're using a consumer account, then the website that I shared, the support site has, uh, information on how to join frontier. If you're an enterprise, then you need to work with your it admins to, to join frontier and there's instructions in there as well. So I suspect that's the, that's the thing, but you know, it has the latest open AI models as well. So you won't be able to use Claude quite yet. Uh, but you can use open AI, but again, we're, we're trying to move as fast as humanly possible to push this out as well. Great question.

[00:49:22] Speaker 3: Great. Awesome. So this is kind of a two parter. Um, but how does agent mode handle protected Excel files? Can it update unprotected cells on a locked sheet? And can agent mode work with the Excel files that have been tagged by Microsoft purview or for data classification and protection?

[00:49:39] Speaker 2: Yeah, that's a good question. We support all of the EDP requirements. That's, that's not a strength of mine on the precise, um, you know, precise behaviors, but, uh, agent mode on GA is EDP compliance and enterprise data protection compliant. Uh, so we, we do respect sensitivity labels. Um, and we, we have protections to make sure that we respect what it is setting for, for the workbooks. Obviously, if the workbook is read only like agent models or can't, uh, can't edit it. Uh, so it has, it respects that as well. Great question.

[00:50:15] Speaker 3: Cool. Next one is for a Microsoft three 65 license in Excel desktop. Is it possible to have agent mode as an add on, like it is in the web version of Excel without having a paid copilot license?

[00:50:28] Speaker 2: As an add on, I guess, uh, I, uh, I'm not quite sure. I understand that. I think it, it requires a supported copilot license. Uh, that's the copilot license.

[00:50:42] Speaker 3: My, my assumption is that she's referring to it as like, Oh, I see an add in.

[00:50:47] Speaker 2: So yeah. So maybe that's a, yeah, sorry.

[00:50:49] Speaker 3: I said, I probably said add on.

[00:50:50] Speaker 2: Cause I think add on there's like M three 65 subscription add on thing. That's more of a business thing. So from an adding perspective, so we initially rolled out agent mode as part of the Excel labs add in. So that's an add in that exists in the, in the office store. Uh, that is where we put like early bits, like exploratory stuff. Um, it's still available there, but I would suggest using, uh, the one built into copilot, cause that's going to have the latest and greatest functionality. And, uh, the add in the Excel labs add in is not supported in windows either. It's only supported in Excel for the web. So I would suggest at this point, just, uh, you know, leverage the, leverage the copilot proper, uh, and then be patient on windows. We'll, we'll, we'll, we'll get there, uh, pretty soon here also on Mac.

[00:51:39] Speaker 3: Great. Awesome. Next question is an agent mode. Can you extract data from multiple PDFs that contain tables of data to a summarized worksheet containing data from several PDFs? There are 60 tables in this example.

[00:51:52] Speaker 2: So as we were talking about earlier, agent mode only has access to the current workbook today. Um, and web, if you enable web. Uh, so it, it doesn't have the ability currently. This is something we're working on to expand the capabilities, to pull data from the other files that you load. Uh, for that, I would suggest using M365 chat. So through M365 chat, you can actually load different PDFs, PowerPoints, Word documents, and it can extract that data for you. And then through that, then you can, you can insert the data into Excel as well. It wouldn't agentically automatically do it for you, but that's possible through M365 chat, but it's also a gap. It's a known gap that we have that we'll, you know, we'll, we'll fill it up at some point soon. Yeah. Great questions.

[00:52:37] Speaker 3: Great. Next. Um, how does it detect and review the errors? Does it detect and alert you automatically? Things need review, or do you have to prompt it using the check to check for those irregular, irregular? Good question.

[00:52:49] Speaker 2: So that kind of depends if it generates errors as the flow, right. As part of solving the problem that you ask it to solve, it will self-correct most of the time, right? It will identify a hash value, hash and a, or whatever it is. Uh, and then it can read that error. It knows the syntax. It knows the workbook. And then it's like, okay, let's correct that. In the other case, I was very specific and I asked it put a comment wherever you find an error and then it scans the data and it finds those discrepancies. So it really depends on the scenario. Good question.

[00:53:23] Speaker 3: Great. Next one. Um, does the agent mode suggest doing certain things or recommend some innovative steps? For example, how does it decide to use a conditional formatting? Should it be prompted or does it give a certain list of options that the user could explore or approve the agent to go ahead and do or implement?

[00:53:40] Speaker 2: I love that question. So. It, you don't have to necessarily note the feature that you wanted to use specifically, right? So if you say, Hey, put some color coding, some dynamic color coding on my table to highlight salient things, uh, it should be a smart enough to use conditional formatting. Now, if you know the features, if you know the name of the feature, that's part where maybe learning some Excel is useful. Uh, you can nudge it. You can be more precise, right? Um, but that's a great question. This whole idea of proactively suggesting options for the user, uh, and what features to use is one that it does do sometimes, but I think that's an area where we could, you know, we could continue to evolve the system for sure. Great question.

[00:54:22] Speaker 1: Yeah. And then, uh, so Carlos, this has been fantastic. Um, maybe just to close, what, what is the kind of future of agent mode in Excel look like? Like what are some of the things that are in the pipeline that you could share?

[00:54:33] Speaker 2: Oh my goodness. Yeah. Well, as I said, like getting into desktop is, you know, that's really, I love Excel for web is growing on me, but I live and breathe and desktop, right? So that's, uh, you know, that's just coverage, rolling it out, getting the baseline there as it is today is quite transformational. Uh, so just, that's kind of priority number one, but there's so much we can do, like both on the, you know, we have like a really awesome science team. We're doing a lot of like research on how to optimize the models, how to improve the models. And obviously the models are going to keep getting better and better. We're going to get better generations. Uh, so that's, there's a whole lot of science work to make the quality and sophistication of the output. It generates higher quality, right? There's also a whole lot of work we're doing on the UX front, like making it more seamless, making it faster.

[00:55:22] Speaker 1: And about UX, just the kind of design. Yeah.

[00:55:24] Speaker 2: The user experience, right? Like the integration between the pain and the grid. How do we explain where the agent is working? How do we explain what changes it made? How do we let users nudge agent in a more efficient way? So there's a whole lot of user interaction stuff that we're, we're exploring and a lot of new patterns that we need to invent there as well. Um, and I would say there is also like starting to expand outside, right? Like as we talked about, like there's other scenarios, other integrations with other apps in Excel, multiple workbooks, you know, there's, there's, there's just so many places we can go. Right. But part of the, the part of the, the thing that we're doing right now, as we are reaching GA is really working with customers, working with the community, understanding how people are using it. And, and that will inform, uh, quite, quite strongly where, where we take this. Right. Yeah.

[00:56:13] Speaker 1: Very, very exciting. And it's really just awesome to see where it is now. And, uh, I'm getting excited about some of these things that are coming in the future. Uh, but Carlos really appreciate you taking the time to walk through this and showcase some of these examples of how you can start using it today. Yes. Uh, and I think just based on the questions, I think there's a, it seems like a lot of enthusiasm behind it and, uh, just recommend giving a shot and try prompting, ask questions and just start using it. Uh, so yeah, thanks everyone for joining and appreciate all the questions and we'll, we'll see you in the next live stream.