WEBVTT 00:00:28.400 --> 00:00:31.490 Good Morning. [applause] 00:00:33.280 --> 00:00:35.997 Good morning. It's fantastic uh to be in 00:00:36.080 --> 00:00:41.037 Madrid uh and in Spain and uh to be in a theater. In fact, I was just thinking of 00:00:41.120 --> 00:00:42.797 the last time I was in a theater like 00:00:42.880 --> 00:00:46.397 this this big was uh growing up in India 00:00:46.480 --> 00:00:51.037 watching some Bollywood flick uh as a middle schooler and being you know 00:00:51.120 --> 00:00:53.197 completely taken aback by the size of 00:00:53.280 --> 00:00:55.997 the theater. Man, this is huge. Uh so 00:00:56.080 --> 00:00:58.797 it's fantastic uh to be with all of you 00:00:58.880 --> 00:01:01.037 um especially at a time like this right 00:01:01.120 --> 00:01:04.077 when there is so much you know talking 00:01:04.160 --> 00:01:07.117 about you know um popular culture who 00:01:07.200 --> 00:01:09.357 would have thought uh that you pick up 00:01:09.440 --> 00:01:12.477 any paper you pick up any news feed and 00:01:12.560 --> 00:01:16.077 all you hear about is AI everywhere um 00:01:16.160 --> 00:01:19.117 and it's exciting uh and so what I 00:01:19.200 --> 00:01:21.597 wanted to do today was to spend some 00:01:21.680 --> 00:01:25.197 time perhaps trying to give our point of 00:01:25.280 --> 00:01:28.157 view on what I think collectively what 00:01:28.240 --> 00:01:30.397 this technology means to us and what we 00:01:30.480 --> 00:01:33.197 can do with it. Uh to us it always 00:01:33.280 --> 00:01:36.717 starts with our mission. Uh our mission 00:01:36.800 --> 00:01:39.597 as a company is about empowering people 00:01:39.680 --> 00:01:44.477 and organizations to achieve more. Right? That's where we 00:01:44.560 --> 00:01:46.397 start from. And whenever there's been a 00:01:46.480 --> 00:01:49.037 new technology, however powerful, we 00:01:49.120 --> 00:01:51.597 have always tried to focus on can that 00:01:51.680 --> 00:01:54.637 technology realize that mission. uh 00:01:54.720 --> 00:01:56.477 which is can we put it in the hands of 00:01:56.560 --> 00:01:59.837 people and in the institutions that they 00:01:59.920 --> 00:02:03.117 build and can that organization that 00:02:03.200 --> 00:02:06.397 person uh really change the frontier of 00:02:06.480 --> 00:02:10.477 what they do right in fact we have this entire description of what a frontier 00:02:10.560 --> 00:02:12.717 firm is right because if you think about 00:02:12.800 --> 00:02:16.157 it right any technology is an input uh 00:02:16.240 --> 00:02:18.397 ultimately what we try to do with it is 00:02:18.480 --> 00:02:20.877 change some output um and that's what 00:02:20.960 --> 00:02:23.597 the movement of the frontier is uh and 00:02:23.680 --> 00:02:26.957 in practice practical terms for us u you 00:02:27.040 --> 00:02:28.717 know in the past we talked about digital 00:02:28.800 --> 00:02:30.957 transformation and now we're thinking 00:02:31.040 --> 00:02:34.397 and talking about uh AI transformation 00:02:34.480 --> 00:02:38.237 how can an organization use AI to change 00:02:38.320 --> 00:02:40.557 the customer experience the employee 00:02:40.640 --> 00:02:43.917 experience the internal operations bend 00:02:44.000 --> 00:02:45.917 the curve of innovation right whatever 00:02:46.000 --> 00:02:48.637 goods or services you're producing can 00:02:48.720 --> 00:02:52.077 you now do it with more throughput a 00:02:52.160 --> 00:02:55.117 much more expans with much better speed. 00:02:55.200 --> 00:02:58.397 Uh that really is the bending of the uh 00:02:58.480 --> 00:03:01.597 frontier. Now the interesting thing with 00:03:01.680 --> 00:03:06.077 AI is that you now have to perhaps you 00:03:06.160 --> 00:03:08.077 know take a step back and think a little 00:03:08.160 --> 00:03:13.037 bit on how to structure yourself for 00:03:13.120 --> 00:03:16.317 this new world. And I think it starts 00:03:16.400 --> 00:03:20.477 with a mind uh mindset shift because AI 00:03:20.560 --> 00:03:23.997 in some sense is a little bit of a horizontal general purpose technology 00:03:24.080 --> 00:03:27.037 that cuts across roles, cuts across 00:03:27.120 --> 00:03:30.397 functions. Uh so you want to think end 00:03:30.480 --> 00:03:35.757 toend workflows, end-to-end process, end-to-end task trajectories, right? 00:03:35.840 --> 00:03:39.037 That mindset is slightly different because we've built up all of what we 00:03:39.120 --> 00:03:40.877 have done in our institution in our 00:03:40.960 --> 00:03:44.077 organizations over the decades on a 00:03:44.160 --> 00:03:48.237 particular workflow. In fact, the last time you could say the workflow really 00:03:48.320 --> 00:03:50.317 changed especially in the enterprise or 00:03:50.400 --> 00:03:53.517 in the in the at the work setting uh was 00:03:53.600 --> 00:03:55.757 when the PCs became standard issue. 00:03:55.840 --> 00:03:57.837 Right? Sometimes I always describe this. 00:03:57.920 --> 00:04:03.037 Imagine in the 80s uh you were a company trying to do a forecast. Uh how would 00:04:03.120 --> 00:04:06.557 you do it? Right? you would sort of have faxes going around and you would have 00:04:06.640 --> 00:04:09.917 inter office memos and then uh and then 00:04:10.000 --> 00:04:13.757 ultimately there would be some kind of a paper forecast and then suddenly email 00:04:13.840 --> 00:04:18.717 became standard issues. Excel was there and you would send an Excel spreadsheet 00:04:18.800 --> 00:04:21.997 and email and people would enter numbers and you would have a forecast, right? 00:04:22.080 --> 00:04:24.877 They just changed the work the workflow 00:04:24.960 --> 00:04:26.637 and the work artifact all 00:04:26.720 --> 00:04:29.357 simultaneously. And that's what I mean 00:04:29.440 --> 00:04:34.157 by really rethinking how the work workflow and the work artifact change 00:04:34.240 --> 00:04:35.917 and the mindset shift. So that means you 00:04:36.000 --> 00:04:38.157 need to be able to think about new 00:04:38.240 --> 00:04:41.197 skills that you need uh to do that new 00:04:41.280 --> 00:04:43.597 tools to do that just like you know 00:04:43.680 --> 00:04:46.237 standard issue in the past uh was all 00:04:46.320 --> 00:04:49.037 these office tools right I mean I one of 00:04:49.120 --> 00:04:53.277 the other things I think about is if somebody had come to us in the 80s and 00:04:53.360 --> 00:04:57.277 said you know three billion people are going to get up or four billion people 00:04:57.360 --> 00:05:01.277 are going to get up every morning and start typing and we would have said why 00:05:01.360 --> 00:05:04.077 like why would we need so many typists 00:05:04.160 --> 00:05:08.637 uh except That's what we do and we describe it as knowledge work. uh right 00:05:08.720 --> 00:05:11.037 and that's what now the new skill set 00:05:11.120 --> 00:05:13.437 with the new tools is going to allow us 00:05:13.520 --> 00:05:17.277 to do and one thing that I think we all 00:05:17.360 --> 00:05:21.117 are going to learn is how to measure the 00:05:21.200 --> 00:05:24.397 quality of the outputs we create right 00:05:24.480 --> 00:05:28.797 so this eval is something that AI people talk about right whenever you're 00:05:28.880 --> 00:05:32.477 building a new model what you're doing is hill climbing on an eval or an 00:05:32.560 --> 00:05:35.757 evaluation criteria and all of us so 00:05:35.840 --> 00:05:37.837 this is not just for labs or the hypers 00:05:37.920 --> 00:05:40.157 scalers or product makers but it's for 00:05:40.240 --> 00:05:42.877 every one of us in some sense to have 00:05:42.960 --> 00:05:46.637 that high taste high bar for what is the 00:05:46.720 --> 00:05:50.157 evaluation of what we are creating as an 00:05:50.240 --> 00:05:53.037 organization and to get clear about it 00:05:53.120 --> 00:05:56.157 and then use AI to transform that eval 00:05:56.240 --> 00:05:58.877 to a new frontier or hill climb. So 00:05:58.960 --> 00:06:02.157 that's a bit of how we think uh one has 00:06:02.240 --> 00:06:05.517 to think about AI transformation. 00:06:05.600 --> 00:06:07.357 And to that end, what we're doing is 00:06:07.440 --> 00:06:09.437 building a full stack. And I'm going to 00:06:09.520 --> 00:06:13.197 spend the rest of my time uh today just 00:06:13.280 --> 00:06:16.637 walking through uh this stack. Um but 00:06:16.720 --> 00:06:19.437 before I do that, I wanted to mention 00:06:19.520 --> 00:06:22.637 that perhaps it's important to think 00:06:22.720 --> 00:06:27.757 about two very important criteria. 00:06:27.840 --> 00:06:30.877 uh when you think about AI since it's a 00:06:30.960 --> 00:06:34.957 learning system of course it means it's 00:06:35.040 --> 00:06:37.677 about intelligence and the intelligence 00:06:37.760 --> 00:06:40.797 that you as a company you as a person 00:06:40.880 --> 00:06:43.437 are going to build with AI right so 00:06:43.520 --> 00:06:46.317 that's clearly uh going to be the case 00:06:46.400 --> 00:06:48.077 but one of the things that you want to 00:06:48.160 --> 00:06:51.997 make sure is top of mind for all of us 00:06:52.080 --> 00:06:56.797 is that it's our intelligence whether 00:06:56.880 --> 00:07:01.597 it's as an individual level or it's an organization. Right? This is not about 00:07:01.680 --> 00:07:03.677 celebrating some intelligence that's 00:07:03.760 --> 00:07:05.197 outside. 00:07:05.280 --> 00:07:08.957 It is about us being able to take the 00:07:09.040 --> 00:07:14.157 latest model but turning that into more 00:07:14.240 --> 00:07:18.077 tacit knowledge, more knowhow inside the 00:07:18.160 --> 00:07:21.917 company. In fact, I'll talk about sovereignty at the very end. Of course, 00:07:22.000 --> 00:07:23.757 everybody's talking about sovereignty at 00:07:23.840 --> 00:07:26.877 the comp country level, but I would say 00:07:26.960 --> 00:07:29.597 one of the most important topics is 00:07:29.680 --> 00:07:32.477 perhaps sovereignty of a firm. Right? 00:07:32.560 --> 00:07:34.397 After all, what is a firm? A firm is 00:07:34.480 --> 00:07:38.477 about the tacet knowledge a company has 00:07:38.560 --> 00:07:43.117 thanks to the human capital that is there uh and even the digital systems. 00:07:43.200 --> 00:07:45.117 And so in the AI age, you want to think 00:07:45.200 --> 00:07:47.597 even more carefully that you are 00:07:47.680 --> 00:07:50.797 building that tacet knowledge and even 00:07:50.880 --> 00:07:54.877 embedding that tacet knowledge and perhaps a set of weights [snorts] that 00:07:54.960 --> 00:07:59.117 you control, a set of agents or models that you control. Right? So it's not 00:07:59.200 --> 00:08:01.757 about celebrating anybody else's model. 00:08:01.840 --> 00:08:04.797 It is about your model that you have in 00:08:04.880 --> 00:08:07.037 your firm that makes you a unique 00:08:07.120 --> 00:08:09.437 participant in the economy. Right? 00:08:09.520 --> 00:08:11.917 That's at the end of the day uh the real 00:08:12.000 --> 00:08:15.117 currency for all of us and that comes 00:08:15.200 --> 00:08:19.197 with this second consideration that means trust. Trust everywhere. There's a 00:08:19.280 --> 00:08:22.557 technical aspect to it, right? For example, how do you make sure that 00:08:22.640 --> 00:08:24.157 there's real governance around your 00:08:24.240 --> 00:08:27.277 data? Uh how do you make sure that uh 00:08:27.360 --> 00:08:29.917 you have security around anything that's 00:08:30.000 --> 00:08:32.717 happening which agents are accessing. Uh 00:08:32.800 --> 00:08:35.517 but the biggest trust is going to be a 00:08:35.600 --> 00:08:39.677 trust in ensuring that your enterprise 00:08:39.760 --> 00:08:41.917 value that you have built over the 00:08:42.000 --> 00:08:45.037 decades inside of your firm doesn't leak 00:08:45.120 --> 00:08:49.517 out uh to somebody else's uh model or or 00:08:49.600 --> 00:08:52.957 somebody else's business and that I think is going to be the two things that 00:08:53.040 --> 00:08:56.077 are going to become very very important as you think through your AI 00:08:56.160 --> 00:08:58.717 transformation journey. So with that as 00:08:58.800 --> 00:09:01.677 a bit of a backdrop I want to start with 00:09:01.760 --> 00:09:04.877 uh the first uh part of our sort of AI 00:09:04.960 --> 00:09:07.597 stack at the experience layer. In fact 00:09:07.680 --> 00:09:11.517 if you look at the best way for us to 00:09:11.600 --> 00:09:14.957 introduce these powerful new AI tools is 00:09:15.040 --> 00:09:18.477 to put them where you already are 00:09:18.560 --> 00:09:22.637 working and then transform that work and 00:09:22.720 --> 00:09:26.237 that workflow. Right? So that's the approach we have taken with all of this 00:09:26.320 --> 00:09:28.477 co-pilot layer. I describe copilot as 00:09:28.560 --> 00:09:31.917 the UI for AI. Um, it's about really 00:09:32.000 --> 00:09:35.197 taking the human and the human agency 00:09:35.280 --> 00:09:37.517 and the human ambition 00:09:37.600 --> 00:09:41.197 and really amplifying it uh by stitching 00:09:41.280 --> 00:09:44.877 in AI uh and then letting you then use 00:09:44.960 --> 00:09:47.357 AI to do things that you never imagined 00:09:47.440 --> 00:09:50.077 even that you could do before, right? Um 00:09:50.160 --> 00:09:52.237 in fact the the way I think about 00:09:52.320 --> 00:09:54.557 sometimes is we've had these very 00:09:54.640 --> 00:09:57.197 powerful metaphors uh that have been 00:09:57.280 --> 00:09:59.597 there with computing right Jobs had this 00:09:59.680 --> 00:10:01.917 of the bicycle for the mind which I I 00:10:02.000 --> 00:10:07.117 think was a very powerful one. Bill in the mid '9s came up with this metaphor 00:10:07.200 --> 00:10:10.877 where he talks about he talked about information at your fingertips. Uh that 00:10:10.960 --> 00:10:13.597 was another very sort of approachable 00:10:13.680 --> 00:10:18.237 way to think about computers especially with the internet. uh and now I feel 00:10:18.320 --> 00:10:22.237 like we're at that a you know the beginning of a new era where the 00:10:22.320 --> 00:10:24.557 metaphor could be that we have infinite 00:10:24.640 --> 00:10:27.437 minds with all these agents that each of 00:10:27.520 --> 00:10:31.277 us can use to shape what we're doing the 00:10:31.360 --> 00:10:34.077 artifacts we create the work we do and 00:10:34.160 --> 00:10:38.557 so on and so that's what we are trying to get done whether it's for you know 00:10:38.640 --> 00:10:43.117 information work whether it's in coding whether it's in security whether it's 00:10:43.200 --> 00:10:45.117 even in science and so that's sort of 00:10:45.200 --> 00:10:47.117 the co-pilot so If you take information 00:10:47.200 --> 00:10:50.397 work, what we've done is built this 00:10:50.480 --> 00:10:53.117 co-pilot layer right into Outlook into 00:10:53.200 --> 00:10:56.077 teams into word, excel, powerpoint and 00:10:56.160 --> 00:10:58.237 also built the copilot app, right? 00:10:58.320 --> 00:11:02.557 Effectively think of it as a new module even and this is a graph that's 00:11:02.640 --> 00:11:04.797 connected. It's stateful. So that means 00:11:04.880 --> 00:11:07.597 you can start one place and go to 00:11:07.680 --> 00:11:12.717 another place and that intelligence layer continues to compound build uh so 00:11:12.800 --> 00:11:17.277 that it helps you with what you're going to do next. Uh right it doesn't stop in 00:11:17.360 --> 00:11:21.277 any one app. It's not captive. In fact that's perhaps the biggest sort of 00:11:21.360 --> 00:11:23.117 misnomer. It's not about any one 00:11:23.200 --> 00:11:26.477 application. It is intelligence that is 00:11:26.560 --> 00:11:29.197 sort of ambient across all applications. 00:11:29.280 --> 00:11:32.397 Right? In fact, you can even think about like one of the metaphors we're playing 00:11:32.480 --> 00:11:35.357 with is think of it as a composer that 00:11:35.440 --> 00:11:38.717 sits above all apps uh right and it's 00:11:38.800 --> 00:11:43.117 there everywhere and that's really literally what at least copilot uh is in 00:11:43.200 --> 00:11:45.197 its first version of it right so you can 00:11:45.280 --> 00:11:47.837 imagine uh where the the direction of 00:11:47.920 --> 00:11:51.677 travel there is so and and this these 00:11:51.760 --> 00:11:55.197 agentic workflows are already there right we started with chat right the 00:11:55.280 --> 00:11:59.277 first couple of years of uh copilot was all about chat chat chat but Now 00:11:59.360 --> 00:12:01.117 increasingly if I look at even my own 00:12:01.200 --> 00:12:04.717 daily usage it's got many agents that 00:12:04.800 --> 00:12:08.637 I'm using all the time and the most popular one that at least all of us I 00:12:08.720 --> 00:12:09.997 think have gotten used to in the last 00:12:10.080 --> 00:12:12.717 what 12 months or so is researcher. So 00:12:12.800 --> 00:12:15.677 that ambient video shows you I go t you 00:12:15.760 --> 00:12:17.917 know essentially assign a research task 00:12:18.000 --> 00:12:20.077 in this case I think it's some kind of a 00:12:20.160 --> 00:12:22.717 product strategy uh document I want the 00:12:22.800 --> 00:12:25.197 researcher to create it goes off and by 00:12:25.280 --> 00:12:29.997 the way it's not just doing reasoning over all the data but it's also using 00:12:30.080 --> 00:12:31.837 the computer in fact it has a computer 00:12:31.920 --> 00:12:35.997 use agent um that it's using as a tool 00:12:36.080 --> 00:12:39.677 uh in order to complete the task I assigned it right so that's kind of an 00:12:39.760 --> 00:12:42.157 example uh of researcher we built 00:12:42.240 --> 00:12:44.317 Another agent called analyst, right? 00:12:44.400 --> 00:12:46.717 This is like having a data science 00:12:46.800 --> 00:12:50.717 analyst with you 24 by7. Um, in fact, 00:12:50.800 --> 00:12:52.797 this one I think is doing um some kind 00:12:52.880 --> 00:12:55.837 of a campaign that I'm u analyzing and I 00:12:55.920 --> 00:12:59.997 load up a bunch of spreadsheets and it goes and does all of the analysis that 00:13:00.080 --> 00:13:02.397 is needed and gives me insights uh like 00:13:02.480 --> 00:13:04.957 a data scientist would do. Uh and the 00:13:05.040 --> 00:13:08.877 third one is uh the Excel uh agent. Uh 00:13:08.960 --> 00:13:11.117 so we've not only uh built these agents 00:13:11.200 --> 00:13:13.677 in copilot but we have also built agent 00:13:13.760 --> 00:13:16.557 mode into word, excel, powerpoint so 00:13:16.640 --> 00:13:19.517 that you can create these new artifacts. 00:13:19.600 --> 00:13:23.917 In fact you know artifact creation 00:13:24.000 --> 00:13:26.557 has become just easier because you can 00:13:26.640 --> 00:13:29.677 single shot it. But the more interesting 00:13:29.760 --> 00:13:33.117 thing is what you do with the artifact 00:13:33.200 --> 00:13:36.397 by using this agent mode and iterating 00:13:36.480 --> 00:13:38.797 on it continuously refining it. Right? 00:13:38.880 --> 00:13:40.797 when I talked about that eval right you 00:13:40.880 --> 00:13:44.157 want like like a great DCF uh you want a 00:13:44.240 --> 00:13:46.877 spreadsheet that does a fantastic cash 00:13:46.960 --> 00:13:49.437 flow uh or what have you and you can get 00:13:49.520 --> 00:13:53.997 that out and then iterate on it refine on uh refine it and that's sort of the 00:13:54.080 --> 00:13:57.917 new way to think about even the workflow because if you you you know previously 00:13:58.000 --> 00:14:02.157 spent a lot of the time building the spreadsheet you now spend a lot more 00:14:02.240 --> 00:14:06.637 time doing the analysis with the spreadsheet uh with agent mode and so 00:14:06.720 --> 00:14:08.637 this is the transformation information. 00:14:08.720 --> 00:14:11.357 And to show you some of this at work, I 00:14:11.440 --> 00:14:13.437 wanted to invite up on stage uh my 00:14:13.520 --> 00:14:17.317 colleague Anne. Anne, take it away. 00:14:17.400 --> 00:14:19.420 [applause] 00:14:19.520 --> 00:14:22.477 >> Thank you, Satia. Hi, everyone. I'm so 00:14:22.560 --> 00:14:24.397 excited to show you how the entire 00:14:24.480 --> 00:14:29.757 Copilot stack is transforming the way people work. I start my day in Copilot, 00:14:29.840 --> 00:14:33.037 and the first thing I'm going to do is look at all of these models that I have 00:14:33.120 --> 00:14:35.117 to choose from. I'm going to pick GPT 00:14:35.200 --> 00:14:37.597 5.2 into quick response because I've got 00:14:37.680 --> 00:14:40.877 a live demo and a lot to show you. Then 00:14:40.960 --> 00:14:44.157 I'll ask Co-Pilot to prep me for my day. 00:14:44.240 --> 00:14:48.717 Co-pilot uses Work IQ to look at my meetings, emails, and chats. And it 00:14:48.800 --> 00:14:52.157 looks at its previous history of interaction with me to know what I 00:14:52.240 --> 00:14:54.557 prioritize the most and come out with a 00:14:54.640 --> 00:14:56.877 customized tailored response. So, let's 00:14:56.960 --> 00:15:02.077 see what's going on today. First up, I have a time block for this demo here. 00:15:02.160 --> 00:15:06.317 So, I'm glad I'm in the right place with you all. And then if we keep going, we 00:15:06.400 --> 00:15:08.397 see at the top of my inbox, I have an 00:15:08.480 --> 00:15:11.117 ongoing customer conversation about a 00:15:11.200 --> 00:15:16.157 pending contract. And part of my job as an accounts receivable manager is to 00:15:16.240 --> 00:15:19.917 make sure our customers pay their bills on time. And to incentivize them to do 00:15:20.000 --> 00:15:22.637 so, I need to offer potentially a cash 00:15:22.720 --> 00:15:26.157 discount for early payments. And again, 00:15:26.240 --> 00:15:30.077 I have lots of options I can offer for such a cash discount, but I want to use 00:15:30.160 --> 00:15:34.957 co-pilot to make sure that I'm following a policy compliant approach. So, I'll 00:15:35.040 --> 00:15:39.517 ask Copilot, what are my options to offer this customer a discount? Now, 00:15:39.600 --> 00:15:44.637 this is where the power of work IQ really shines. Copilot's looking at our 00:15:44.720 --> 00:15:48.637 policy documentation. It's looking at the context and history of that customer 00:15:48.720 --> 00:15:52.077 and this particular contract. And it's applying my role as an accounts 00:15:52.160 --> 00:15:54.477 receivable manager to drive what it is 00:15:54.560 --> 00:15:59.117 I'm specifically entitled to author. And here we go. We see how all of these 00:15:59.200 --> 00:16:01.357 things come through in the response. 00:16:01.440 --> 00:16:06.077 First up, I can see that we did ini indeed find that policy documentation. I 00:16:06.160 --> 00:16:10.157 can verify that the source is our finance SharePoint. So, we're accurately 00:16:10.240 --> 00:16:14.477 pulling those details. And then if I look through copilot's given me this 00:16:14.560 --> 00:16:18.397 great summary of all the different options, I can provide anything that 00:16:18.480 --> 00:16:23.197 requires additional approval beyond my manager level remit. And it ultimately 00:16:23.280 --> 00:16:25.197 summarizes we should start with our 00:16:25.280 --> 00:16:27.357 standard 2% cash discount for any 00:16:27.440 --> 00:16:30.237 payments made within 10 days of invoice. 00:16:30.320 --> 00:16:34.637 Now, that makes sense to me, but before I go off and add this to the contract, I 00:16:34.720 --> 00:16:38.557 need to assess the impact to our profitability. And what better place to 00:16:38.640 --> 00:16:42.797 do that than Excel? Here in Excel, I have a workbook that I've already built 00:16:42.880 --> 00:16:45.197 with agent mode that models the revenue 00:16:45.280 --> 00:16:47.597 on the contract as it currently is 00:16:47.680 --> 00:16:52.317 stated. And I want to add this nuance of how a discount impacts it. Right from 00:16:52.400 --> 00:16:56.477 here, I'll go to co-pilot and turn on agent mode. So, I don't have to do all 00:16:56.560 --> 00:17:00.797 this work manually. I'll have the agent take care of it for me. I'll ask it to 00:17:00.880 --> 00:17:04.237 create a new sheet that models the impact of offering that 210 cash 00:17:04.320 --> 00:17:08.797 discount. And because the agent's grounded in the same work context I am, 00:17:08.880 --> 00:17:12.077 we'll give it access to that policy documentation file. And it can 00:17:12.160 --> 00:17:14.077 extrapolate additional details from that 00:17:14.160 --> 00:17:17.277 context. Then we'll kick it off. What 00:17:17.360 --> 00:17:19.757 the agent is going to do here is work 00:17:19.840 --> 00:17:24.317 directly inside my workbook just like I might or a co-orker might. Unlike us, it 00:17:24.400 --> 00:17:27.917 goes quite a bit faster, but it still takes a little bit of time, which means 00:17:28.000 --> 00:17:31.677 now I have time I can use for other activities. 00:17:31.760 --> 00:17:35.437 All these insights that are helping drive business decision making, whether 00:17:35.520 --> 00:17:39.917 it's in Copilot, in Outlook or Excel, they come from our underlying financial 00:17:40.000 --> 00:17:43.917 data estate. So for our business insights to be valuable, our data needs 00:17:44.000 --> 00:17:48.477 to be accurate and up-to-date. Dynamics 365 is our platform for intelligent 00:17:48.560 --> 00:17:53.197 business applications, which enables us to do just that. Here in Dynamics, when 00:17:53.280 --> 00:17:57.037 I log in, I see I have a notification from the account reconciliation agent. 00:17:57.120 --> 00:18:00.557 That's a pre-built agent that comes out of the box in Dynamics, and it's letting 00:18:00.640 --> 00:18:04.077 me know there's a couple discrepancies in our books that need my attention. So, 00:18:04.160 --> 00:18:08.717 let's take a look. For each of our reconciliation exceptions, I get a 00:18:08.800 --> 00:18:12.157 natural language summary that immediately tells me exactly what's 00:18:12.240 --> 00:18:16.477 going on. In this case, a transaction was posted directly to our ledger, and I 00:18:16.560 --> 00:18:20.557 can see those posting details here, and it missed our subleddger. And if you're 00:18:20.640 --> 00:18:24.957 not a finance person, don't worry. Just know this is not ideal. We need 00:18:25.040 --> 00:18:29.277 transactions in both places. And it's not just insight that the agent is 00:18:29.360 --> 00:18:33.517 giving me, but it's also recommending actions. It knows finance best 00:18:33.600 --> 00:18:35.917 practices. It knows our best practices. 00:18:36.000 --> 00:18:39.357 And in this case, it recommends we reverse the transaction so we can post 00:18:39.440 --> 00:18:42.637 it properly through the subleddger. 00:18:42.720 --> 00:18:47.277 From there, all I have to do is confirm and commit the transaction. And I want 00:18:47.360 --> 00:18:51.197 to highlight what's happening here because it's so important. This fix is 00:18:51.280 --> 00:18:53.997 being logged directly in our financials. 00:18:54.080 --> 00:18:57.837 So our source of truth can remain just that. There we go. We got a confirmation 00:18:57.920 --> 00:19:02.317 the reversal was completed. And for any of my audit folks out there today, 00:19:02.400 --> 00:19:06.077 everything the agent does, everything I do gets logged in our audit trail. So we 00:19:06.160 --> 00:19:09.837 have that critical accuracy and accountability we need in finance 00:19:09.920 --> 00:19:13.597 processes. Okay, we took care of that discrepancy. 00:19:13.680 --> 00:19:15.357 Let's go back and check on our co-worker 00:19:15.440 --> 00:19:18.077 agent in Excel. 00:19:18.160 --> 00:19:21.037 And what we can see already is how we 00:19:21.120 --> 00:19:25.757 are on a brand new sheet that the agent created and it's making edits live right 00:19:25.840 --> 00:19:27.917 in front of us. We can even follow its 00:19:28.000 --> 00:19:30.317 thought process as it goes step by step 00:19:30.400 --> 00:19:35.437 in the pane over here. It looks like it's reassessing. And because we're 00:19:35.520 --> 00:19:39.037 doing a live demo and we've only got so much time left, I'm going to jump ahead 00:19:39.120 --> 00:19:43.117 to the same prompt that I ran earlier just to show you that output of what it 00:19:43.200 --> 00:19:45.917 looks like. So, we can see we have our 00:19:46.000 --> 00:19:48.317 discount situation broken out. It's 00:19:48.400 --> 00:19:53.197 pulled that detail of the 2% 10 days from the prompt and additional context 00:19:53.280 --> 00:19:55.277 from the document that I attached to it. 00:19:55.360 --> 00:19:57.197 It's looked at our monthly impact to 00:19:57.280 --> 00:20:02.397 revenue and ultimately we should have 00:20:02.480 --> 00:20:06.717 the chart I asked for as well over here on the side. What I want to stress here, 00:20:06.800 --> 00:20:10.557 it's not just putting static text in cells. These are dynamic formulas and 00:20:10.640 --> 00:20:12.637 visuals that work with me and the agent 00:20:12.720 --> 00:20:14.717 as we continue to work together and 00:20:14.800 --> 00:20:17.837 refine. And most importantly over here I 00:20:17.920 --> 00:20:21.917 have full visibility into all the data sources, calculations and assumptions 00:20:22.000 --> 00:20:24.157 the agent used to build out this 00:20:24.240 --> 00:20:27.437 context. So to recap, what did we just 00:20:27.520 --> 00:20:31.997 see? We saw how co-pilot with work IQ helps me work not just more efficiently 00:20:32.080 --> 00:20:36.637 but more effectively. We saw how agent mode in Excel is helping me very quickly 00:20:36.720 --> 00:20:40.957 strategize and make decisions that drive profitability of a contract. And we saw 00:20:41.040 --> 00:20:45.837 how agents in Dynamics 365 are helping us create a financial data estate that 00:20:45.920 --> 00:20:49.837 enables the whole business to make great decisions. And this is not just for 00:20:49.920 --> 00:20:54.317 finance personas. Everything I showed you today is 100% available. I hope you 00:20:54.400 --> 00:20:56.557 go home and try it out in Copilot, in 00:20:56.640 --> 00:21:03.160 Excel, and in Dynamics 365. Thanks. And back to you, Satia. 00:21:05.470 --> 00:21:08.877 >> [applause] >> you know, I was uh when I was seeing the 00:21:08.960 --> 00:21:13.117 demo, I was thinking, God, I started, you know, my career at Microsoft doing a 00:21:13.200 --> 00:21:14.797 lot of Excel demos. I had a lot more 00:21:14.880 --> 00:21:17.917 hair then and here I am um you know, in 00:21:18.000 --> 00:21:20.557 2026 still looking at Excel demos. You 00:21:20.640 --> 00:21:24.077 know, one thing that hopefully you capt, 00:21:24.160 --> 00:21:26.317 you know, you got out of that what an 00:21:26.400 --> 00:21:29.837 showed was, you know, the design of AI 00:21:29.920 --> 00:21:34.717 systems, I think requires a particular thought that it doesn't add more 00:21:34.800 --> 00:21:38.637 cognitive load to you, right? I mean, one of the things that could easily 00:21:38.720 --> 00:21:41.037 happen is because you have this powerful 00:21:41.120 --> 00:21:43.357 tool that can do things like create an 00:21:43.440 --> 00:21:45.917 unbelievable spreadsheet or go off and 00:21:46.000 --> 00:21:48.077 reason for hours and come back with some 00:21:48.160 --> 00:21:51.997 answer. And if all it does is cause you 00:21:52.080 --> 00:21:54.477 to then spend more time understanding 00:21:54.560 --> 00:21:56.637 what it did, then that's not great 00:21:56.720 --> 00:21:59.757 design. So one of the subtle things that 00:21:59.840 --> 00:22:03.917 we are trying to make sure happens uh is 00:22:04.000 --> 00:22:06.557 as we build out this co-pilot layer, how 00:22:06.640 --> 00:22:09.357 it works with work IQ and how you 00:22:09.440 --> 00:22:12.077 traverse uh your job and your task 00:22:12.160 --> 00:22:16.957 trajectory, the cognitive load has to come down. uh even the aesthetic like 00:22:17.040 --> 00:22:18.557 one of the things that an was showing in 00:22:18.640 --> 00:22:22.157 Excel in agent mode uh right 00:22:22.240 --> 00:22:24.477 spreadsheets the reason why spreadsheets 00:22:24.560 --> 00:22:27.997 have st stood the test of time uh from 00:22:28.080 --> 00:22:29.757 whatever 40 I think we just celebrated 00:22:29.840 --> 00:22:33.117 the 40th year of Excel uh is because it 00:22:33.200 --> 00:22:36.957 has it communicates a lot cognitively 00:22:37.040 --> 00:22:38.877 right the the fact that you have a 00:22:38.960 --> 00:22:42.157 tabular data form um you can sort of 00:22:42.240 --> 00:22:46.157 look at the formulas in fact I so I remember the first time I sat next to 00:22:46.240 --> 00:22:50.157 somebody you know one of these Excel ma maestros was like sitting next to a 00:22:50.240 --> 00:22:52.477 piano player uh because the way they 00:22:52.560 --> 00:22:55.517 built the model uh it made sense and 00:22:55.600 --> 00:22:58.157 when you see the output it made sense to 00:22:58.240 --> 00:23:02.717 you as a consumer of that model uh and that needs to be true even when it comes 00:23:02.800 --> 00:23:06.637 to AI right so it can't be I single shot at a model that nobody else can 00:23:06.720 --> 00:23:10.877 understand but it has the answer and then you're looking for hours as to how 00:23:10.960 --> 00:23:12.797 the heck did they find the answer and so 00:23:12.880 --> 00:23:17.997 that's the type of stuff uh that matters in how AI gets designed into 00:23:18.080 --> 00:23:21.917 experiences. So we it's great to see the diffusion of 00:23:22.000 --> 00:23:25.677 all of this technology right here in Spain. I had a chance to meet with many 00:23:25.760 --> 00:23:27.437 customers who are deploying you know 00:23:27.520 --> 00:23:30.797 co-pilot uh using C-pilot. Um I had a 00:23:30.880 --> 00:23:32.877 chance to meet this one you know uh 00:23:32.960 --> 00:23:36.157 partner of ours uh illunion uh which is 00:23:36.240 --> 00:23:38.637 you know doing it's a doing a fantastic 00:23:38.720 --> 00:23:40.877 job of not only using this technology 00:23:40.960 --> 00:23:43.757 but the social good uh you know how they 00:23:43.840 --> 00:23:48.237 employ people with disabilities in their workforce and empower them uh with 00:23:48.320 --> 00:23:52.477 technology is fantastic to see they built an agent in fact using an agent 00:23:52.560 --> 00:23:55.037 builder inside a co-pilot uh for their 00:23:55.120 --> 00:23:57.357 the law department to be able to reason 00:23:57.440 --> 00:24:00.477 over complex PDFs Uh in fact that's kind 00:24:00.560 --> 00:24:05.277 of one of the most favorite features of copilot is man PDFs are tough to parse 00:24:05.360 --> 00:24:09.677 uh and you can go you know assign them to copilot and it can come back uh and 00:24:09.760 --> 00:24:14.237 extract the right sort of uh reasoning on top of it and give you information. 00:24:14.320 --> 00:24:16.237 It's just great to see uh this kind of 00:24:16.320 --> 00:24:19.357 diffusion. So that was the first layer 00:24:19.440 --> 00:24:22.477 of the stack which is about building in 00:24:22.560 --> 00:24:26.957 AI experience into the flow uh of human 00:24:27.040 --> 00:24:29.117 activity and then really helping us 00:24:29.200 --> 00:24:32.637 raise the human ambition. Uh the next 00:24:32.720 --> 00:24:35.037 piece of the stack is the agent 00:24:35.120 --> 00:24:37.757 platform. Um and when it comes to agent 00:24:37.840 --> 00:24:42.877 platform the first place to start is the 00:24:42.960 --> 00:24:45.437 data layer but more importantly not just 00:24:45.520 --> 00:24:48.797 raw data but uplifting the semantics of 00:24:48.880 --> 00:24:51.437 this data into what we describe as the 00:24:51.520 --> 00:24:54.237 IQ layer right and an talked about this 00:24:54.320 --> 00:24:56.477 right when when she was using co-pilot 00:24:56.560 --> 00:24:59.437 she was essentially interfacing with 00:24:59.520 --> 00:25:03.037 this thing that we describe as work IQ 00:25:03.120 --> 00:25:05.197 uh and and and I would claim that one of 00:25:05.280 --> 00:25:07.997 the most important databases inside of a 00:25:08.080 --> 00:25:11.517 commercial entity is the information 00:25:11.600 --> 00:25:14.077 underneath their Microsoft 365 tenant. I 00:25:14.160 --> 00:25:16.237 mean think about it right inside of 00:25:16.320 --> 00:25:19.997 Microsoft 365 you have all this very 00:25:20.080 --> 00:25:22.717 rich information people their 00:25:22.800 --> 00:25:24.877 relationships to other people like your 00:25:24.960 --> 00:25:28.397 org chart uh not just org chart in its 00:25:28.480 --> 00:25:32.557 formality but even who is working with whom on what projects right I mean think 00:25:32.640 --> 00:25:37.597 about how important that is for any company uh you have all their calendars 00:25:37.680 --> 00:25:41.437 you have all of their communications in teams and outlook you have all their 00:25:41.520 --> 00:25:43.517 documents in one drive and sharepoint 00:25:43.600 --> 00:25:46.797 you have projects and project schedules. 00:25:46.880 --> 00:25:49.437 So that is a very rich set of 00:25:49.520 --> 00:25:52.557 relationship and entities uh that now 00:25:52.640 --> 00:25:55.837 you can in fact use AI models to reason 00:25:55.920 --> 00:25:58.157 over. That's what work IQ is. It's a 00:25:58.240 --> 00:26:01.277 stateful reasoning agent essentially on 00:26:01.360 --> 00:26:04.957 top of a very rich data model uh which 00:26:05.040 --> 00:26:07.277 captures all this tacet knowledge of an 00:26:07.360 --> 00:26:09.997 organization. The other place where 00:26:10.080 --> 00:26:12.157 we've done in fact you all have created 00:26:12.240 --> 00:26:14.957 a tremendous amount of semantic data 00:26:15.040 --> 00:26:19.517 semantic information is PowerBI right I mean if you think about the number of 00:26:19.600 --> 00:26:21.917 dashboards over the years that have been 00:26:22.000 --> 00:26:24.957 created someone took the pains to make 00:26:25.040 --> 00:26:27.677 sure that all that data was annotated 00:26:27.760 --> 00:26:30.877 with the right semantic um annotation so 00:26:30.960 --> 00:26:35.677 that you could make sense of it right uh this is some sales aggregate information 00:26:35.760 --> 00:26:37.837 it is some forecast information it's 00:26:37.920 --> 00:26:40.797 some supply chain information. Now all 00:26:40.880 --> 00:26:43.917 that can be brought together uh in 00:26:44.000 --> 00:26:48.877 fabric IQ. Uh and then foundry IQ is another service we have built in to be 00:26:48.960 --> 00:26:52.317 able to in fact have an outer orchestrator of all these IQs. 00:26:52.400 --> 00:26:54.877 Essentially you can have a plan that 00:26:54.960 --> 00:26:59.757 gets created that orchestrates work IQ, fabric IQ and any other document 00:26:59.840 --> 00:27:04.477 database or any other thing that you have. Uh and now you can build 00:27:04.560 --> 00:27:09.037 essentially agents on top of all of this IQ layer. Right? So data is the best 00:27:09.120 --> 00:27:11.117 bootstrap but more importantly not just 00:27:11.200 --> 00:27:13.117 raw data that you build a forward index 00:27:13.200 --> 00:27:16.397 on or you do a search on but literally 00:27:16.480 --> 00:27:19.357 have a reasoning layer plus data 00:27:19.440 --> 00:27:22.397 available for any agentic system. So 00:27:22.480 --> 00:27:24.797 that's sort of the first layer of the uh 00:27:24.880 --> 00:27:27.757 the agent platform. Then once you have 00:27:27.840 --> 00:27:32.077 this, the thing that you need is an app 00:27:32.160 --> 00:27:34.077 server, right? I mean with any 00:27:34.160 --> 00:27:37.437 generation of technology, 00:27:37.520 --> 00:27:41.677 you always built a middle tier app server so that you could build 00:27:41.760 --> 00:27:43.757 applications on that technology, right? 00:27:43.840 --> 00:27:45.757 It happened with the web, right? We had 00:27:45.840 --> 00:27:48.877 web servers. Uh we had it with.NET and 00:27:48.960 --> 00:27:51.917 Java. Uh we had that with cloud, right? 00:27:52.000 --> 00:27:54.157 We had cloudnative app servers. So this 00:27:54.240 --> 00:27:56.877 time around again for us it's foundry 00:27:56.960 --> 00:27:59.837 where you have all the services to build 00:27:59.920 --> 00:28:02.317 agent and multi- aent systems starting 00:28:02.400 --> 00:28:05.117 with frameworks uh with you know 00:28:05.200 --> 00:28:09.117 services for AI safety like guard rails 00:28:09.200 --> 00:28:12.077 services for doing these eval services 00:28:12.160 --> 00:28:14.237 to do fine-tuning what have you. So 00:28:14.320 --> 00:28:17.837 these are a complete list of runtime 00:28:17.920 --> 00:28:20.157 services that you need in order to build 00:28:20.240 --> 00:28:23.037 agents and multi- aent systems. And on 00:28:23.120 --> 00:28:26.237 top of that, you have models. In fact, 00:28:26.320 --> 00:28:30.157 Foundry comes with 11,000 models. And you would say, "Wow, why do I need 00:28:30.240 --> 00:28:33.277 11,000 models?" In fact, it comes with all the popular ones, right? You have 00:28:33.360 --> 00:28:37.117 all the GPT family, you have the Claude family, you have grot, you, you know, 00:28:37.200 --> 00:28:39.037 you pick your, you know, frontier model 00:28:39.120 --> 00:28:40.957 that you want to use. It's there. But it 00:28:41.040 --> 00:28:44.397 also has in fact all the models in all 00:28:44.480 --> 00:28:48.717 the domains that you want to be able to access and use as part of your 00:28:48.800 --> 00:28:51.917 applications. And in fact just to kind of give you an intuition for why the 00:28:52.000 --> 00:28:54.477 world needs more models not just one 00:28:54.560 --> 00:28:57.037 model it, you know, we launched this 00:28:57.120 --> 00:28:58.957 thing called gigat time I think a few 00:28:59.040 --> 00:29:03.197 months back. Um and you know it was done 00:29:03.280 --> 00:29:06.557 in in cooperation with Providence um and 00:29:06.640 --> 00:29:10.957 a couple of other partners of our University of Washington. Um and the 00:29:11.040 --> 00:29:13.517 idea was if you look at cancer treatment 00:29:13.600 --> 00:29:18.237 and imunotherapy, one of the challenges is you never know if a particular 00:29:18.320 --> 00:29:22.237 imunotherapy would work on a particular tumor. And so in order to figure that 00:29:22.320 --> 00:29:26.717 out, there's a test that one does. And that test, you know, costs a lot of 00:29:26.800 --> 00:29:29.037 money, takes a lot of time. And so the 00:29:29.120 --> 00:29:32.077 researchers put together essentially a 00:29:32.160 --> 00:29:35.517 spatial proteomics um solution. So 00:29:35.600 --> 00:29:40.397 essentially using generative AI to simulate that test, right? So instead of 00:29:40.480 --> 00:29:44.237 doing the test, what if we could do a simulation of that test and that way 00:29:44.320 --> 00:29:47.837 then you can reduce the cost of it all and make it available to every secondary 00:29:47.920 --> 00:29:50.237 hospital in every uh country so that 00:29:50.320 --> 00:29:55.037 they can now do that same test and you know make a difference in oncology. So 00:29:55.120 --> 00:29:56.717 to give you a flavor for this, let's 00:29:56.800 --> 00:29:59.597 roll the video. 00:29:59.680 --> 00:30:02.157 Imunotherapy is really the most 00:30:02.240 --> 00:30:04.557 promising direction for taming cancer 00:30:04.640 --> 00:30:09.757 once and for all. The whole idea is using the immune system to fight cancer. 00:30:09.840 --> 00:30:12.797 >> The big challenges we have in the field right now is to understand which 00:30:12.880 --> 00:30:16.877 patients will respond. >> Facial proteomics can simultaneously 00:30:16.960 --> 00:30:19.357 measure multiple protein and create 00:30:19.440 --> 00:30:22.477 multiplex imunofllororescent images. It 00:30:22.560 --> 00:30:26.637 can tell you whether the patient might respond to imunotherapy or not. But 00:30:26.720 --> 00:30:28.797 unfortunately to generate this kind of 00:30:28.880 --> 00:30:34.760 images takes days and cost thousands [music] of dollars. 00:30:35.520 --> 00:30:39.437 In collaboration with Providence and University of Washington, we developed 00:30:39.520 --> 00:30:42.077 this multimodel AI called gigat time to 00:30:42.160 --> 00:30:45.277 generate spatial proteomics at scale. So 00:30:45.360 --> 00:30:46.877 we have trained the model using large 00:30:46.960 --> 00:30:50.157 scale data analyzing millions of cells. 00:30:50.240 --> 00:30:52.957 The input image has limited data. We 00:30:53.040 --> 00:30:55.597 don't know where the protein is active. 00:30:55.680 --> 00:30:57.437 Gigat time can convert [music] this 00:30:57.520 --> 00:31:00.957 image to a colorful multiplex imunosence 00:31:01.040 --> 00:31:03.437 image. The color is the protein 00:31:03.520 --> 00:31:08.317 activation. They are the key factor drive diseases especially in cancer. 00:31:08.400 --> 00:31:12.637 This can help us to predict patient drug response. We now taking the first 00:31:12.720 --> 00:31:16.397 meaningful step to be able to generate this kind of data for everyone in the 00:31:16.480 --> 00:31:18.797 globe. 00:31:18.880 --> 00:31:23.677 The gigat time model is now available to everyone through Microsoft Foundry. Our 00:31:23.760 --> 00:31:25.677 hope is to [music] empower every 00:31:25.760 --> 00:31:28.317 secondary hospital in every tertiary 00:31:28.400 --> 00:31:32.717 city to be able to now conduct their own research. 00:31:32.800 --> 00:31:37.037 >> We look forward to a future where we can actually bring this to the clinic, 00:31:37.120 --> 00:31:38.877 impact patients lives and bring the next 00:31:38.960 --> 00:31:40.877 generation of therapies to places where 00:31:40.960 --> 00:31:44.557 this would be otherwise impossible. 00:31:44.640 --> 00:31:48.077 You know, it's a great example of um how 00:31:48.160 --> 00:31:50.957 you translate sort of unique knowledge 00:31:51.040 --> 00:31:55.837 that a institution in this case the combination of quite frankly Providence 00:31:55.920 --> 00:31:59.437 and University of Washington had into something that now of course is going to 00:31:59.520 --> 00:32:03.517 be a powerful tool and a powerful model for the rest of the world to use. Uh but 00:32:03.600 --> 00:32:08.237 when I talked earlier about that intelligence in every company, that's 00:32:08.320 --> 00:32:11.597 what it is. In fact, if some someone asks me like how many models should 00:32:11.680 --> 00:32:15.997 there be in the world, I would say the number of the companies out there are in 00:32:16.080 --> 00:32:17.997 the world because after all, every 00:32:18.080 --> 00:32:21.277 company is effectively a model of some 00:32:21.360 --> 00:32:25.277 unique tacet knowledge. Uh sometimes it's proprietary. You don't want to put 00:32:25.360 --> 00:32:29.037 it in foundry, but it'll be in your company. But sometimes it may make sense 00:32:29.120 --> 00:32:33.117 for you to then expose it to even the world. And that to me is uh how to think 00:32:33.200 --> 00:32:36.477 about models. is not about just one model but it's about sort of the models 00:32:36.560 --> 00:32:40.797 all of us build uh that truly increase not only the intelligence in your 00:32:40.880 --> 00:32:43.037 company but the collective intelligence. 00:32:43.120 --> 00:32:45.277 So now we talked about two things, 00:32:45.360 --> 00:32:47.597 right? We talked about that IQ layer. We 00:32:47.680 --> 00:32:50.157 talked about the app server with 00:32:50.240 --> 00:32:54.077 Foundry. And now it's time to build uh 00:32:54.160 --> 00:32:57.517 agents. And to do that, we have a a full 00:32:57.600 --> 00:33:01.757 tool chain, right? Starting with App Builder. Uh for those of you who used 00:33:01.840 --> 00:33:03.997 things like Power Apps before, App 00:33:04.080 --> 00:33:06.957 Builder is the AI sort of era version of 00:33:07.040 --> 00:33:09.437 it where I can now go in and just use 00:33:09.520 --> 00:33:11.757 natural language to essentially build a 00:33:11.840 --> 00:33:14.877 fullscale app. Um and in this case, I 00:33:14.960 --> 00:33:18.957 think it's sort of um an app that was builting building for a person a 00:33:19.040 --> 00:33:21.677 personalized career path uh application. 00:33:21.760 --> 00:33:25.277 Um and so they were able to just simply use natural language to build a full 00:33:25.360 --> 00:33:28.317 stack application that was generated. Um 00:33:28.400 --> 00:33:31.597 a second uh solution we have is copilot 00:33:31.680 --> 00:33:33.517 studio right so copilot studio allows 00:33:33.600 --> 00:33:36.477 you to build agents uh so you can again 00:33:36.560 --> 00:33:39.517 uh take any workflow and make it agentic 00:33:39.600 --> 00:33:42.157 by grounding it in some knowledge uh and 00:33:42.240 --> 00:33:45.037 again using uh this builder tool uh to 00:33:45.120 --> 00:33:47.197 build a pretty sophisticated um 00:33:47.280 --> 00:33:49.437 application for employee onboarding for 00:33:49.520 --> 00:33:53.517 example um and so this is a second tool 00:33:53.600 --> 00:33:55.837 that's available uh for us and of course 00:33:55.920 --> 00:34:00.717 if you're a professional developer We now have GitHub and Visual Studio and uh 00:34:00.800 --> 00:34:04.477 you're able to use the sophestication of those tools to build agent and 00:34:04.560 --> 00:34:07.277 multi-agent systems. Uh in fact, it's 00:34:07.360 --> 00:34:12.077 fantastic to see the growth of GitHub as a community in uh uh is Spain. We have 00:34:12.160 --> 00:34:13.837 two and a half million developers on 00:34:13.920 --> 00:34:17.677 GitHub. It's grown 22% over year. Um and 00:34:17.760 --> 00:34:21.357 it's just unbelievable like what's happening uh in coding and the number of 00:34:21.440 --> 00:34:25.517 repos that are getting created in public and of course a lot more in private. uh 00:34:25.600 --> 00:34:29.677 just given the power of these coding agents and so uh it's great to see even 00:34:29.760 --> 00:34:31.997 uh Spain participate in it. So one of 00:34:32.080 --> 00:34:34.397 the core goals we have with GitHub is 00:34:34.480 --> 00:34:37.037 not just to have your repo or your CI/CD 00:34:37.120 --> 00:34:40.157 pipelines and all the other core 00:34:40.240 --> 00:34:42.557 primitives of GitHub but it's also 00:34:42.640 --> 00:34:45.197 becomes the organizing layer for all of 00:34:45.280 --> 00:34:48.237 the coding agents. So we're designing in 00:34:48.320 --> 00:34:51.117 uh right into GitHub uh you know 00:34:51.200 --> 00:34:55.357 whatever your favorite coding agent is is available right there. In fact, you 00:34:55.440 --> 00:35:00.077 can go to your mobile app in the morning and you essentially have a new inbox uh 00:35:00.160 --> 00:35:01.837 which is your coding agents and their 00:35:01.920 --> 00:35:05.357 work uh and you can go and keep track of 00:35:05.440 --> 00:35:08.477 what's happening on your repo uh you 00:35:08.560 --> 00:35:11.357 know by these coding agents. Uh so 00:35:11.440 --> 00:35:14.477 that's what we describe as agent HQ. Now 00:35:14.560 --> 00:35:19.357 the other thing that we've also sort of um made sure is we build all the 00:35:19.440 --> 00:35:23.677 different form factors right because if you look at how coding has changed uh 00:35:23.760 --> 00:35:27.997 the IDE still remains a very important one in fact if you look back uh the 00:35:28.080 --> 00:35:30.397 evolution the first thing where we used 00:35:30.480 --> 00:35:32.877 AI was in the next that it suggests 00:35:32.960 --> 00:35:35.437 right so you're in an IDE it's doing 00:35:35.520 --> 00:35:38.797 essentially uh completions uh then we 00:35:38.880 --> 00:35:41.597 put chat uh so that you could stay in 00:35:41.680 --> 00:35:43.837 flow and you didn't have to go outside 00:35:43.920 --> 00:35:48.797 to get answers and you could continue to code. Uh then we had agent mode where 00:35:48.880 --> 00:35:53.277 you could assign the agent a set of tasks. Um and it could complete that. Uh 00:35:53.360 --> 00:35:57.677 and then you have these fully autonomous agents as well. Now so in fact putting 00:35:57.760 --> 00:36:01.677 all of these form factors together, right? The CLI or the command line 00:36:01.760 --> 00:36:04.477 matters, the IDE matters, the autonomous 00:36:04.560 --> 00:36:09.437 agent matters. The other thing we are also making sure we do is all the models 00:36:09.520 --> 00:36:11.117 are available to you. Right? It's not 00:36:11.200 --> 00:36:15.117 it's multi-form factor multimodel uh and 00:36:15.200 --> 00:36:17.917 in fact many many sub aents uh that you 00:36:18.000 --> 00:36:19.757 may build for a variety of reasons for 00:36:19.840 --> 00:36:23.757 profiling for security all are part of 00:36:23.840 --> 00:36:28.237 the GitHub copilot ecosystem and to show you all of this in action I wanted to 00:36:28.320 --> 00:36:30.237 invite up on stage my colleague Marlene 00:36:30.320 --> 00:36:33.537 Marlene take it away 00:36:33.620 --> 00:36:35.437 [applause] 00:36:35.520 --> 00:36:39.277 >> thanks Aia and hi everyone let's say I'm 00:36:39.360 --> 00:36:44.077 a developer working at a toy company called Tailspin Toys. A few days ago, I 00:36:44.160 --> 00:36:48.157 got an email from the search product management team asking me to add some 00:36:48.240 --> 00:36:52.557 new search and filter features to the site. They've given me a short checklist 00:36:52.640 --> 00:36:56.797 of what I should build, including a search bar with text search for simple 00:36:56.880 --> 00:36:59.037 searches and Azure search for more 00:36:59.120 --> 00:37:01.277 complex ones. They've also asked me to 00:37:01.360 --> 00:37:03.997 add in a sidebar so customers can filter 00:37:04.080 --> 00:37:09.037 by both category and price. I'd like for Copilot to help me with this task. And 00:37:09.120 --> 00:37:11.917 so to do that, I'm going to head to my terminal 00:37:12.000 --> 00:37:13.677 and I'm going to run the command to get 00:37:13.760 --> 00:37:17.757 GitHub Copilot CLI started. With Copilot 00:37:17.840 --> 00:37:21.117 CLI, I can now kick off any agentic task 00:37:21.200 --> 00:37:23.997 directly in my terminal. 00:37:24.080 --> 00:37:28.477 For C-pilot to know how it should help me with my task, it needs to somehow be 00:37:28.560 --> 00:37:30.317 able to get the information from the 00:37:30.400 --> 00:37:32.797 email we got from the PM team and bring 00:37:32.880 --> 00:37:35.597 that information into the terminal. 00:37:35.680 --> 00:37:37.997 Earlier, Satia talked about work IQ, 00:37:38.080 --> 00:37:40.557 which allows us to connect to the M365 00:37:40.640 --> 00:37:43.757 suite and to bring that information here 00:37:43.840 --> 00:37:49.197 into the CLI so Copilot can work with it. You'll see at the top here. It's 00:37:49.280 --> 00:37:50.637 doing a lot of thinking, but you'll see 00:37:50.720 --> 00:37:52.957 that Copilot has already connected to 00:37:53.040 --> 00:37:56.237 the work IQ MCP server. In my prompt, 00:37:56.320 --> 00:37:58.397 I've asked Copilot to look for the PM 00:37:58.480 --> 00:38:01.757 team's email and then to return a list 00:38:01.840 --> 00:38:03.837 of the things we need to build based off 00:38:03.920 --> 00:38:05.997 of that email. You'll also see that 00:38:06.080 --> 00:38:09.437 copilot has invoked the work IQ skill 00:38:09.520 --> 00:38:12.557 which under the hood uses the graph API 00:38:12.640 --> 00:38:14.957 to be able to triage through my emails. 00:38:15.040 --> 00:38:17.757 It's going to look through and hopefully 00:38:17.840 --> 00:38:20.237 if the demo gods are on my side this 00:38:20.320 --> 00:38:22.797 morning, it's going to find the email 00:38:22.880 --> 00:38:25.997 that the PM team sent us and then we'll 00:38:26.080 --> 00:38:29.197 list out the things we need to build. 00:38:29.280 --> 00:38:31.037 While it's doing that, I'll also point 00:38:31.120 --> 00:38:33.757 out that I'm currently connected to 00:38:33.840 --> 00:38:38.557 Claude Opus 4.6. So, this is one of the 00:38:38.640 --> 00:38:40.717 most popular coding models that we have 00:38:40.800 --> 00:38:43.997 available, but we have a couple more 00:38:44.080 --> 00:38:46.957 um as well. Okay, great. I can see that 00:38:47.040 --> 00:38:49.757 Copilot has used Worth Q to find the 00:38:49.840 --> 00:38:52.077 email and it's list out listed out the 00:38:52.160 --> 00:38:54.797 checklist that was in the email we saw 00:38:54.880 --> 00:38:57.117 earlier before. So, we're almost ready 00:38:57.200 --> 00:38:59.997 to build. And like I said, I'm connected 00:39:00.080 --> 00:39:03.277 to Claude Opus 4.6, but I'm not sure if 00:39:03.360 --> 00:39:08.077 this is the right model for the job. So, let's see what options we have. With a 00:39:08.160 --> 00:39:10.237 single GitHub copilot subscription, I 00:39:10.320 --> 00:39:12.237 now have access to all the latest 00:39:12.320 --> 00:39:15.917 models, including like I can see here, 00:39:16.000 --> 00:39:18.877 GPT 5.3 codecs. Now, I'm not sure how 00:39:18.960 --> 00:39:21.117 many of you are chronically on social 00:39:21.200 --> 00:39:23.837 media online like I am, but if you are, 00:39:23.920 --> 00:39:26.317 you might have already heard about this 00:39:26.400 --> 00:39:28.637 model. It just was released a few days 00:39:28.720 --> 00:39:30.957 ago and it's already available for us to 00:39:31.040 --> 00:39:33.837 use in GitHub Copilot today. So, this is 00:39:33.920 --> 00:39:35.757 super exciting. I'm going to choose this 00:39:35.840 --> 00:39:38.957 model and then I'm going to send through 00:39:39.040 --> 00:39:41.117 and then I'm going to send through a new 00:39:41.200 --> 00:39:44.797 prompt to be able to ask co-pilots to go 00:39:44.880 --> 00:39:47.677 ahead and get started with building that 00:39:47.760 --> 00:39:52.077 search and filter feature. You'll see that the model begins by 00:39:52.160 --> 00:39:54.237 thinking. It's going to analyze both my 00:39:54.320 --> 00:39:57.597 prompt and the codebase around it. And 00:39:57.680 --> 00:40:01.837 after that, it's going to create a plan and then implement on that plan. This 00:40:01.920 --> 00:40:03.517 process is going to take a few minutes. 00:40:03.600 --> 00:40:07.197 And so while Copilot works in the background, let's switch over to a new 00:40:07.280 --> 00:40:09.757 terminal. Earlier today, I gave Copilot 00:40:09.840 --> 00:40:12.477 the same prompt asking you to build that 00:40:12.560 --> 00:40:14.557 search and filter feature, but I asked 00:40:14.640 --> 00:40:17.757 it to add the changes to a PR. You'll 00:40:17.840 --> 00:40:20.477 see that we are also connected to the 00:40:20.560 --> 00:40:23.677 Playright MCP server. I'd like for 00:40:23.760 --> 00:40:26.877 Copilot to use this MCP server to test 00:40:26.960 --> 00:40:29.677 the search bar and the filter just like 00:40:29.760 --> 00:40:32.397 a human user would. 00:40:32.480 --> 00:40:34.717 Playright is an open-source testing 00:40:34.800 --> 00:40:36.797 framework built by Microsoft that allows 00:40:36.880 --> 00:40:39.997 us to be able to automate browser 00:40:40.080 --> 00:40:42.797 interactions. So to be able to test an 00:40:42.880 --> 00:40:46.237 app, Playright will begin by opening up 00:40:46.320 --> 00:40:49.997 a browser and it will start running an 00:40:50.080 --> 00:40:54.397 automated script. It's going to start typing in inputs, navigating to 00:40:54.480 --> 00:40:56.877 different pages, and all along the way 00:40:56.960 --> 00:40:59.277 it's going to take screenshots. You can 00:40:59.360 --> 00:41:01.757 see my hands are not on the keyboard. 00:41:01.840 --> 00:41:04.717 And this is all co-pilot and playright 00:41:04.800 --> 00:41:07.197 doing this work for us and testing that 00:41:07.280 --> 00:41:09.837 our features are working as expected. We 00:41:09.920 --> 00:41:12.717 just saw Copilot literally working 00:41:12.800 --> 00:41:17.437 through four different functional tests of our application. We can see that our 00:41:17.520 --> 00:41:19.117 search bar and our filters are working 00:41:19.200 --> 00:41:21.517 as expected. And this is perfect. Our 00:41:21.600 --> 00:41:24.717 tests have passed. Now that we know that 00:41:24.800 --> 00:41:27.117 the app is working, we're almost ready 00:41:27.200 --> 00:41:29.997 to merge the PR. But before we do, let's 00:41:30.080 --> 00:41:34.477 add in some documentation and head over to VS Code for that. 00:41:34.560 --> 00:41:36.957 For developers, good documentation has 00:41:37.040 --> 00:41:39.437 always been important, but I would say 00:41:39.520 --> 00:41:42.157 it's even more so now that we can 00:41:42.240 --> 00:41:45.517 generate code faster than ever with AI. 00:41:45.600 --> 00:41:49.757 To be able to automate the process of creating better documentation, I've 00:41:49.840 --> 00:41:52.157 created a PR visualizer app and I've 00:41:52.240 --> 00:41:55.037 used the GitHub Copilot SDK to do this. 00:41:55.120 --> 00:41:57.757 Copilot SDK gives developers the power 00:41:57.840 --> 00:42:00.557 of GitHub copilot directly in their 00:42:00.640 --> 00:42:02.877 apps. The Python code that you can see 00:42:02.960 --> 00:42:05.757 on the screen creates a co-pilot agent 00:42:05.840 --> 00:42:08.077 session. And in this session, I'm 00:42:08.160 --> 00:42:11.437 passing through this PR data tool set. 00:42:11.520 --> 00:42:15.837 And that set will be used to pull in data from the PR we saw uh in the 00:42:15.920 --> 00:42:18.077 terminal. And then we'll pass that data, 00:42:18.160 --> 00:42:20.717 the key points to Excaliraw, which we've 00:42:20.800 --> 00:42:23.517 connected to Copilot using MCP. And then 00:42:23.600 --> 00:42:25.917 Excaliraw is going to generate a diagram 00:42:26.000 --> 00:42:30.637 for us that will show us which changes have been made to our code in that PR. 00:42:30.720 --> 00:42:34.477 So I have the app running locally in simple browser and I'm going to pass 00:42:34.560 --> 00:42:36.237 through the link to the PR that I got 00:42:36.320 --> 00:42:40.877 from Copilot and it's going to start off 00:42:40.960 --> 00:42:45.597 by calling that tool we passed through in the code and we'll use that tool to 00:42:45.680 --> 00:42:49.197 pull in the information and then it's going to combine that with the excal 00:42:49.280 --> 00:42:52.477 draw format to understand how to then 00:42:52.560 --> 00:42:57.677 generate a diagram like you can see on the screen there to show us what has 00:42:57.760 --> 00:43:01.357 been added to our code. So we can see which parts the search bar the filter 00:43:01.440 --> 00:43:05.037 have been added to the code and which parts of our code base have been 00:43:05.120 --> 00:43:07.677 modified as well. So in the past few 00:43:07.760 --> 00:43:11.277 minutes we've seen Copilot with work IQ 00:43:11.360 --> 00:43:13.677 pull in a spec from an email and build 00:43:13.760 --> 00:43:16.237 it. We've also seen Copilot work with 00:43:16.320 --> 00:43:18.797 Playright to be able to test an app 00:43:18.880 --> 00:43:20.957 functionally. And finally we've just 00:43:21.040 --> 00:43:24.957 seen Copilot SDK with MCP generate an 00:43:25.040 --> 00:43:27.517 architecture diagram for us. So whether 00:43:27.600 --> 00:43:30.637 you're working in the CLI, in the IDE, 00:43:30.720 --> 00:43:33.597 or adding AI to your applications, give 00:43:33.680 --> 00:43:36.397 GitHub copilot a try today. Over to you, 00:43:36.480 --> 00:43:40.920 Charles. Thanks. 00:43:42.240 --> 00:43:47.437 >> No, thanks. Thanks, Marlene. And um you know, that kind of gives you a flavor 00:43:47.520 --> 00:43:49.037 for 00:43:49.120 --> 00:43:53.277 um you know, what's possible uh with 00:43:53.360 --> 00:43:57.037 GitHub Copilot. And if you look at the, you know, who would have thought again, 00:43:57.120 --> 00:43:59.357 you know, you're back at the terminal 00:43:59.440 --> 00:44:01.917 doing all these powerful things. And 00:44:02.000 --> 00:44:06.637 that's because you can through natural language get your tasks done, right? I 00:44:06.720 --> 00:44:08.397 mean, you're not just writing code, but 00:44:08.480 --> 00:44:10.877 you're able to literally reference a 00:44:10.960 --> 00:44:13.757 team's meeting, your email discussions, 00:44:13.840 --> 00:44:17.437 and have work IQ surface the plan that 00:44:17.520 --> 00:44:21.357 you then go execute through the coding agent and change your repo. I mean 00:44:21.440 --> 00:44:23.997 that's the power uh of some of these AI 00:44:24.080 --> 00:44:27.277 capabilities all compounding. Uh so work 00:44:27.360 --> 00:44:30.957 IQ doesn't just remain in uh Office 365 00:44:31.040 --> 00:44:34.317 or Microsoft 365. It now is available to 00:44:34.400 --> 00:44:39.597 you even when uh you're coding and that I think is a great example. The other 00:44:39.680 --> 00:44:43.117 thing that Marlene showed which I think is another very powerful thing for 00:44:43.200 --> 00:44:46.397 developers is the entire platform or the 00:44:46.480 --> 00:44:49.037 runtime underneath GitHub copilot is now 00:44:49.120 --> 00:44:51.197 available for anybody to build their own 00:44:51.280 --> 00:44:54.157 agentic system. Uh so you can now take 00:44:54.240 --> 00:44:56.237 that access to multiple models. Right? 00:44:56.320 --> 00:45:01.357 So when she did the PR visualizer that was built using that SDK uh where 00:45:01.440 --> 00:45:04.877 literally you have access to all the models that GitHub copilot has. So your 00:45:04.960 --> 00:45:07.837 app has that the skills uh and its 00:45:07.920 --> 00:45:12.637 ability to call MCP servers. It has that. So that way you can now build 00:45:12.720 --> 00:45:16.877 applications that are you know maybe in core banking or in insurance or in 00:45:16.960 --> 00:45:21.357 healthcare that literally are also going to be like GitHub copilot with variety 00:45:21.440 --> 00:45:25.677 of different form factors multiple models and access to all the MCP servers 00:45:25.760 --> 00:45:30.477 and skills. So that's just a great example. So that was the stanza where we 00:45:30.560 --> 00:45:35.117 now said we can you can build you can build all these agents and multi- aent 00:45:35.200 --> 00:45:37.437 systems. In fact you can take what is 00:45:37.520 --> 00:45:39.597 underneath even GitHub copilot and build 00:45:39.680 --> 00:45:42.717 your own version of it. Now if you have 00:45:42.800 --> 00:45:45.677 all these agents, the next big issue 00:45:45.760 --> 00:45:48.157 that one has to deal with is how do you 00:45:48.240 --> 00:45:51.597 govern them? And really to ensure that 00:45:51.680 --> 00:45:55.357 you have a robust solution uh for both 00:45:55.440 --> 00:45:58.877 observability and governance of AI 00:45:58.960 --> 00:46:01.517 agents, we built agent 365. And to talk 00:46:01.600 --> 00:46:06.637 about this very important initiative, uh I wanted to introduce up on stage my 00:46:06.720 --> 00:46:09.947 colleague Charles Lammana. Charles, come on up. 00:46:10.030 --> 00:46:12.050 >> [applause] 00:46:12.640 --> 00:46:15.437 >> Thanks. 00:46:15.520 --> 00:46:18.237 And like Satia said, we're going to have 00:46:18.320 --> 00:46:22.557 a lot of agents moving forward. What we're going to see is a need to have a 00:46:22.640 --> 00:46:26.957 control plane to go manage and sit on top of all of these different agents. 00:46:27.040 --> 00:46:29.677 That way, we're going to be able to get the right level of visibility and 00:46:29.760 --> 00:46:33.437 understanding about what people are creating and how those agents are 00:46:33.520 --> 00:46:37.197 interacting with other people, with data, and all the different systems 00:46:37.280 --> 00:46:39.837 inside the enterprise. And this control 00:46:39.920 --> 00:46:42.717 plane really enables three core things 00:46:42.800 --> 00:46:45.917 across all of your agents. First, it 00:46:46.000 --> 00:46:50.077 helps you with governance. You can create policies and workflows so you can 00:46:50.160 --> 00:46:52.557 manage exactly who creates your agents, 00:46:52.640 --> 00:46:54.717 who consumes them, how they're shared, 00:46:54.800 --> 00:46:57.597 and what systems they can work with. 00:46:57.680 --> 00:47:01.757 Second, you can have perfect observability. So, you can understand 00:47:01.840 --> 00:47:06.397 what these agents are doing on a regular basis. That way, if there's a risk or a 00:47:06.480 --> 00:47:10.557 potential security problem or just errors or cost issues, you can start to 00:47:10.640 --> 00:47:14.877 have an understanding across all of your different agents. And last but not 00:47:14.960 --> 00:47:18.797 least, you can enforce security policies. So no matter what these agents 00:47:18.880 --> 00:47:23.437 are doing in terms of connecting or authenticating or authorizing or even 00:47:23.520 --> 00:47:28.077 having real-time protection for things like jailbreaks and XPIA, you're able to 00:47:28.160 --> 00:47:31.997 protect these agents across your entire estate. And as all of this starts to 00:47:32.080 --> 00:47:35.437 come together, we need to have a really integrated experience. So, I'm going to 00:47:35.520 --> 00:47:39.757 walk you through some of the new Agent 365 capabilities that we've been 00:47:39.840 --> 00:47:43.917 releasing. If we look at what Agent 365 enables you 00:47:44.000 --> 00:47:47.677 to do is it's a one-stop shop to get a view of all of those agents that are 00:47:47.760 --> 00:47:51.917 created, it's available right inside the Microsoft Admin Center. So the same 00:47:52.000 --> 00:47:56.477 place where you already go today to do things like manage your users or devices 00:47:56.560 --> 00:47:58.877 or even your co-pilot rollout, you're 00:47:58.960 --> 00:48:01.117 now able to see all of the different 00:48:01.200 --> 00:48:04.877 agents that are active. And if I look at 00:48:04.960 --> 00:48:07.437 some of this top level dashboard view, 00:48:07.520 --> 00:48:09.357 you can get an idea of how many agents 00:48:09.440 --> 00:48:12.877 there are. In this example, 128,000, as 00:48:12.960 --> 00:48:16.797 well as how many users are interacting with those agents, in this case, over 00:48:16.880 --> 00:48:19.677 58,000. And we chose these numbers for 00:48:19.760 --> 00:48:21.517 the demo to really highlight that ratio 00:48:21.600 --> 00:48:24.077 that we expect. There are going to be a 00:48:24.160 --> 00:48:28.797 lot of agents and every user is going to need to interact with many agents to get 00:48:28.880 --> 00:48:32.717 their job done most efficiently. Whether it's inside of GitHub copilot or agents 00:48:32.800 --> 00:48:37.757 that your IT team or dev teams have created or even from third parties. And 00:48:37.840 --> 00:48:39.277 that's one of the important parts of 00:48:39.360 --> 00:48:43.517 agent 365 is it's not just for Microsoft 00:48:43.600 --> 00:48:47.917 platforms or Microsoft agents. We of course have registrations for Copilot 00:48:48.000 --> 00:48:49.997 Studio and Foundry, but if your agents 00:48:50.080 --> 00:48:54.957 are built and integrated with Workday or Google or Service Now or Manis, you're 00:48:55.040 --> 00:48:58.957 able to see these agents and what they're doing. And you can track things 00:48:59.040 --> 00:49:03.997 like trends as well as potential risks as they show up. And the risk part is an 00:49:04.080 --> 00:49:07.197 important item because if you have over a 100,000 agents, you don't want to 00:49:07.280 --> 00:49:10.077 click through each one of them to understand what the problems are. So, 00:49:10.160 --> 00:49:13.997 we're able to give you the top actions as an admin. So, you can go and manage 00:49:14.080 --> 00:49:17.757 things like which agents require review or approval, where you have those 00:49:17.840 --> 00:49:20.717 potential problems or agents that have been orphaned and no longer have an 00:49:20.800 --> 00:49:25.277 owner. All available inside that console. And if I say click through into 00:49:25.360 --> 00:49:28.557 agent risk, you can see here are some of them that that have been identified. And 00:49:28.640 --> 00:49:32.557 if I look at the comm's agent, I can start to get a quick view of what its 00:49:32.640 --> 00:49:34.317 problems are. I can see there are two 00:49:34.400 --> 00:49:35.997 security risks that require my 00:49:36.080 --> 00:49:39.757 attention. And if I go and look at it, 00:49:39.840 --> 00:49:43.437 you can start to see that the agents are integrating with the same systems we 00:49:43.520 --> 00:49:46.797 already use to secure those users and those devices. In this case, it uses 00:49:46.880 --> 00:49:50.877 Entra to let me know that the agent has an abnormal sign-in frequency. So that's 00:49:50.960 --> 00:49:53.197 based on things like location, IP range, 00:49:53.280 --> 00:49:57.997 how often it's logging in, it's been flagged for further review. Or one of 00:49:58.080 --> 00:50:00.957 the reasons you want to have your agents and your users managed in the same 00:50:01.040 --> 00:50:04.957 system is we can look and identify a risky user who may have been compromised 00:50:05.040 --> 00:50:09.117 is interacting with this agent in a way that isn't normal. This can help us 00:50:09.200 --> 00:50:13.357 elevate and flag across people and agents and devices where that risk is 00:50:13.440 --> 00:50:18.317 starting to show up. And in addition to looking at the agents, you have to have 00:50:18.400 --> 00:50:22.557 a process by which these agents get deployed. So we have an experience where 00:50:22.640 --> 00:50:26.397 you can approve each agent as it's getting set up. So if I look at this 00:50:26.480 --> 00:50:30.317 staffing agent which someone has requested me to go provision, I can 00:50:30.400 --> 00:50:33.677 start to publish it out to users and I can manage do I want to do it to 00:50:33.760 --> 00:50:37.277 everybody or do I want to do it to specific users or groups? I can manage 00:50:37.360 --> 00:50:41.997 who has access to it and importantly I have this idea of a template and these 00:50:42.080 --> 00:50:45.997 templates have built-in policies around what the agent can do. So I can just 00:50:46.080 --> 00:50:49.357 choose for example the staffing access governance template to make sure the 00:50:49.440 --> 00:50:54.237 agent's going to behave the way I expect it to. And if I look here, I can also 00:50:54.320 --> 00:50:57.197 review the permissions and the access that this agent is going to take 00:50:57.280 --> 00:51:00.957 advantage of. So I can very quickly click through and get a good idea of 00:51:01.040 --> 00:51:04.637 what these agents are doing across my organization and start to scale that 00:51:04.720 --> 00:51:09.597 agent practice. But it's not just about the agents, it's also about the 00:51:09.680 --> 00:51:14.317 intelligence, the systems it connects to. So if I look here inside of the 00:51:14.400 --> 00:51:19.277 agent 365 experience, we have an IQ page. And this is important because the 00:51:19.360 --> 00:51:23.437 thing that makes you unique is your data, not the model. And here I can 00:51:23.520 --> 00:51:26.237 understand things like how many emails 00:51:26.320 --> 00:51:30.477 or how much emails or meeting contexts are being used by agents on a regular 00:51:30.560 --> 00:51:34.957 basis. And this lets you know how that IQ is being put to work to go drive that 00:51:35.040 --> 00:51:36.877 business outcome and business process 00:51:36.960 --> 00:51:39.197 value. And this goes across all of the 00:51:39.280 --> 00:51:42.797 important work IQ data and more. So what 00:51:42.880 --> 00:51:45.917 agent 365 really starts to enable is 00:51:46.000 --> 00:51:50.637 this idea that you can empower users to create agents. You can have IT teams 00:51:50.720 --> 00:51:54.317 that build their agents and you can also work with third parties and other 00:51:54.400 --> 00:51:56.397 companies to get those agents deployed 00:51:56.480 --> 00:51:59.837 at work. This creates an incredible 00:51:59.920 --> 00:52:02.557 agent ecosystem. We've already worked 00:52:02.640 --> 00:52:06.477 with dozens of other companies like SAP, 00:52:06.560 --> 00:52:08.877 Service Now, Workday, as well as 00:52:08.960 --> 00:52:11.677 startups like Genpark and Manis or even 00:52:11.760 --> 00:52:14.717 the AI labs like Anthropic and OpenAI to 00:52:14.800 --> 00:52:17.677 integrate with agent 365. That way, as 00:52:17.760 --> 00:52:21.837 you get these things deployed, you can make sure you have a single simple way 00:52:21.920 --> 00:52:24.317 to govern and control them across the 00:52:24.400 --> 00:52:29.197 entire company. So, with that, I hope you give a 365 a try. It's available in 00:52:29.280 --> 00:52:31.117 the Microsoft admin center today and 00:52:31.200 --> 00:52:34.437 I'll hand it back over to you Satia. 00:52:34.520 --> 00:52:38.237 [applause] >> Thanks. 00:52:38.320 --> 00:52:40.557 >> Yeah, that it's it's a very important 00:52:40.640 --> 00:52:43.677 sort of aspect I think of how the roll 00:52:43.760 --> 00:52:45.357 out of these agents will happen. I mean 00:52:45.440 --> 00:52:48.557 that thing that Charles showed the ratio 00:52:48.640 --> 00:52:51.597 of people to agents by definition is 00:52:51.680 --> 00:52:53.677 going to be enormous. uh it may be what 00:52:53.760 --> 00:52:58.317 2:1 3:1 who knows uh but you really want 00:52:58.400 --> 00:53:01.437 to have this governance system uh this 00:53:01.520 --> 00:53:03.517 observability system and the assurance 00:53:03.600 --> 00:53:06.317 of security so that you can confidently 00:53:06.400 --> 00:53:10.797 deploy these agents inside of the enterprise. You know the good news in 00:53:10.880 --> 00:53:15.997 all of this again is there's fantastic traction in building out these agentic 00:53:16.080 --> 00:53:20.397 systems right here in Spain. And I had this morning a chance to meet uh you 00:53:20.480 --> 00:53:23.757 know a bunch of developers. Uh I had a chance to meet the developers at 00:53:23.840 --> 00:53:27.357 Factorial who are building out a full HR 00:53:27.440 --> 00:53:29.997 agentic system. Uh it's fantastic to see 00:53:30.080 --> 00:53:32.397 the success they're having uh their 00:53:32.480 --> 00:53:34.637 design uh of how they're building the 00:53:34.720 --> 00:53:39.517 system out. Um also had a chance to meet with Ferovial. In fact obviously they're 00:53:39.600 --> 00:53:42.797 some of the largest infrastructure builders. They're building in fact our 00:53:42.880 --> 00:53:45.757 Madrid data center expansion as well. 00:53:45.840 --> 00:53:47.597 Um, and they were showing me how they've 00:53:47.680 --> 00:53:50.077 built an agentic system that literally 00:53:50.160 --> 00:53:53.597 allows you to effectively apply uh, 00:53:53.680 --> 00:53:56.877 intelligence in real time to any project 00:53:56.960 --> 00:53:58.957 of theirs, whether it's a data center 00:53:59.040 --> 00:54:01.997 buildout, a highway, an airport. Um, had 00:54:02.080 --> 00:54:04.157 a chance to meet with the Repol folks 00:54:04.240 --> 00:54:07.357 who are deploying um, uh, these IT 00:54:07.440 --> 00:54:11.837 workflows. Uh in fact they showed me an agent that was doing automatic conver I 00:54:11.920 --> 00:54:13.917 mean basically was able to connect a 00:54:14.000 --> 00:54:16.477 front end using ABBA to their SAP and 00:54:16.560 --> 00:54:19.757 doing the automation uh to of a workflow 00:54:19.840 --> 00:54:22.317 and now they're scaling it beyond it. um 00:54:22.400 --> 00:54:24.957 had a chance to meet with uh Sanitas 00:54:25.040 --> 00:54:28.237 which is really built a voice system um 00:54:28.320 --> 00:54:32.237 the which is a front end to how their co 00:54:32.320 --> 00:54:35.597 customers are interfacing uh with uh 00:54:35.680 --> 00:54:37.917 their uh their entire sort of mobile 00:54:38.000 --> 00:54:40.797 application. Uh and so it's great to see 00:54:40.880 --> 00:54:45.437 uh the level of innovation and the diffusion of all of these technologies 00:54:45.520 --> 00:54:49.117 uh and the developers right here in Spain uh not only building for Spain but 00:54:49.200 --> 00:54:51.677 for the world uh and uh and it just 00:54:51.760 --> 00:54:54.157 showcases the power uh of the human 00:54:54.240 --> 00:54:57.277 capital uh here and then the and in 00:54:57.360 --> 00:54:59.197 their hands these tools get translated 00:54:59.280 --> 00:55:03.117 into an unbelievable uh value. The last 00:55:03.200 --> 00:55:06.317 sort of part of the talk is going to be 00:55:06.400 --> 00:55:08.237 about the token factory. So, we talked 00:55:08.320 --> 00:55:12.717 about AI um the experience layer, we 00:55:12.800 --> 00:55:16.477 talked about the agent platform, and of course, if you're going to have lots of 00:55:16.560 --> 00:55:19.197 agents, you need infrastructure that 00:55:19.280 --> 00:55:22.397 generates tokens most efficiently. Um 00:55:22.480 --> 00:55:24.157 and it's all about this particular 00:55:24.240 --> 00:55:26.557 formula, right? It's about performance 00:55:26.640 --> 00:55:29.837 uh of tokens per dollar per watt. Um and 00:55:29.920 --> 00:55:34.557 the goal is how do you really make sure um uh you have the most efficient system 00:55:34.640 --> 00:55:36.477 for it? and we're building out Azure as 00:55:36.560 --> 00:55:40.077 the world's computer. Um we have uh 70 00:55:40.160 --> 00:55:42.797 plus data center regions, 400 plus data 00:55:42.880 --> 00:55:45.197 centers. We continuing to expand that. 00:55:45.280 --> 00:55:50.317 Uh right here in Spain, we have two regions uh both in central and north. Uh 00:55:50.400 --> 00:55:54.877 the other thing that we have remained very committed to is renewable. Uh and 00:55:54.960 --> 00:55:56.877 we are 100% in fact we've achieved that 00:55:56.960 --> 00:55:59.837 milestone where all of our data centers 00:55:59.920 --> 00:56:02.637 have 100% of renewable power. In fact, 00:56:02.720 --> 00:56:05.277 you can even think about our spend in 00:56:05.360 --> 00:56:08.557 capital and our procurement and PPAs are 00:56:08.640 --> 00:56:11.597 a great stimulant uh for essentially any 00:56:11.680 --> 00:56:14.237 renewable financing uh and project 00:56:14.320 --> 00:56:18.957 financing. And so it's great to see this uh continue to be a focus for us and 00:56:19.040 --> 00:56:20.397 everyone else who are our partners 00:56:20.480 --> 00:56:25.437 there. The other aspect uh is the the 00:56:25.520 --> 00:56:29.197 systems themselves. In fact, we are at a fascinating time, right? We are at a 00:56:29.280 --> 00:56:32.397 time where you are starting to see the 00:56:32.480 --> 00:56:35.357 workloads somewhat get stable, right? 00:56:35.440 --> 00:56:40.557 These atcale training workloads or inference workloads. They're unlike 00:56:40.640 --> 00:56:41.917 anything we've seen in the past, right? 00:56:42.000 --> 00:56:45.037 These are large sync data parallel 00:56:45.120 --> 00:56:49.117 synchronous um you know GPU workloads. 00:56:49.200 --> 00:56:51.357 Uh but you're now learning that for 00:56:51.440 --> 00:56:54.237 example take even an inference workload 00:56:54.320 --> 00:56:56.957 that you can optimize it. uh you can 00:56:57.040 --> 00:57:00.477 disagregate the prefill and the decode. 00:57:00.560 --> 00:57:04.637 um you could in fact take even the how 00:57:04.720 --> 00:57:08.957 you access memory whether it's high bandwidth memory or SRAMM how do you 00:57:09.040 --> 00:57:11.357 think about uh the dist like you can 00:57:11.440 --> 00:57:14.237 have a draft model completely in SRAMM 00:57:14.320 --> 00:57:18.717 so there's a complete renaissance of systems in fact so you know starting in 00:57:18.800 --> 00:57:20.637 December to now there has been a 00:57:20.720 --> 00:57:22.957 renaissance uh in how you think about 00:57:23.040 --> 00:57:25.917 even the systems for inference uh and so 00:57:26.000 --> 00:57:28.237 the Microsoft approach is going to be to 00:57:28.320 --> 00:57:31.437 have a a fleet that takes advantage of 00:57:31.520 --> 00:57:35.997 the best technology and uses the power of software to make it available to so 00:57:36.080 --> 00:57:38.637 that you can have the best token factory. Right? We're going to partner 00:57:38.720 --> 00:57:42.637 with Nvidia. We're going to partner with AMD. We're building our own systems. In 00:57:42.720 --> 00:57:45.197 fact, Maya 200, which we just disclosed, 00:57:45.280 --> 00:57:47.677 which is taking real traffic today, uh 00:57:47.760 --> 00:57:50.397 is around 30 plus% better than any 00:57:50.480 --> 00:57:53.197 leading node in any leading system uh in 00:57:53.280 --> 00:57:57.517 Azure today. So we're going to continue to innovate uh and really take advantage 00:57:57.600 --> 00:57:59.597 of the ecosystem innovation so that we 00:57:59.680 --> 00:58:02.637 can build out the best token factories. 00:58:02.720 --> 00:58:07.277 Now we in addition to doing all this work around building out the data 00:58:07.360 --> 00:58:09.277 centers making sure that they're you 00:58:09.360 --> 00:58:12.157 know got renewable sources of energy 00:58:12.240 --> 00:58:14.317 innovation in the system side and the 00:58:14.400 --> 00:58:17.837 silicon side. Uh the other aspect is 00:58:17.920 --> 00:58:20.637 making sure we're committed uh to 00:58:20.720 --> 00:58:24.077 Europe. Um and we made in April of last 00:58:24.160 --> 00:58:26.557 year, we had these European commitments 00:58:26.640 --> 00:58:28.957 uh around strengthening the digital 00:58:29.040 --> 00:58:31.357 ecosystems here with local partnerships 00:58:31.440 --> 00:58:35.117 and local innovation, privacy, security 00:58:35.200 --> 00:58:39.597 and continuity uh and digital resilience. Uh and these are commitments 00:58:39.680 --> 00:58:44.077 we stand by and we've continued to work and one of the aspects of this is 00:58:44.160 --> 00:58:48.477 sovereignty. uh after all sovereignty is a top consideration 00:58:48.560 --> 00:58:50.957 uh for any country and uh in order to 00:58:51.040 --> 00:58:53.277 address that what we've done is really 00:58:53.360 --> 00:58:55.837 built a portfolio of solutions. In fact, 00:58:55.920 --> 00:58:57.997 if you think about sovereignty as 00:58:58.080 --> 00:59:01.037 essentially managing even risk uh the 00:59:01.120 --> 00:59:05.997 way you manage risk in any other domain is to have a portfolio. Um and that is 00:59:06.080 --> 00:59:07.597 essentially what you should even think 00:59:07.680 --> 00:59:09.757 of when it comes to sovereignty which is 00:59:09.840 --> 00:59:13.197 some workloads using sovereign services 00:59:13.280 --> 00:59:16.077 in public cloud. You can for example 00:59:16.160 --> 00:59:18.157 encrypt all your data. You can manage 00:59:18.240 --> 00:59:23.117 the keys that do the encryption. You can use confidential computing uh in order 00:59:23.200 --> 00:59:27.197 to ensure that the data that's being used even in a public cloud is never 00:59:27.280 --> 00:59:29.997 outside a confidential chamber. Right? 00:59:30.080 --> 00:59:33.597 So effectively it is fully sovereign. Uh 00:59:33.680 --> 00:59:37.437 so that's sort of one solution. On top of that, we have a solution which is 00:59:37.520 --> 00:59:40.557 fully local. Um and so when it comes to 00:59:40.640 --> 00:59:45.357 uh we already have Microsoft 365 uh local, we have Azure local and we're 00:59:45.440 --> 00:59:49.037 very thrilled this week to announce Foundry local. Right? So when we talked 00:59:49.120 --> 00:59:50.957 about Foundry with all its models, 00:59:51.040 --> 00:59:53.757 they'll be available now even locally uh 00:59:53.840 --> 00:59:58.477 for you to be able to use in a private uh setting. Uh and not only that, we 00:59:58.560 --> 01:00:00.557 were also thrilled this week to announce 01:00:00.640 --> 01:00:03.197 that we've taken Azure local which up to 01:00:03.280 --> 01:00:07.917 now had a control plane uh which was in the cloud but even the local with local 01:00:08.000 --> 01:00:10.157 operations but now you can have fully 01:00:10.240 --> 01:00:13.197 disconnected uh um control plane. So 01:00:13.280 --> 01:00:16.157 that means you have a local cloud with a 01:00:16.240 --> 01:00:18.797 local operation control plane versus 01:00:18.880 --> 01:00:21.917 being in any hybrid uh mode uh at all. 01:00:22.000 --> 01:00:25.277 And so these are all again um options 01:00:25.360 --> 01:00:29.037 for you to think about the different workloads and where you would like to 01:00:29.120 --> 01:00:31.677 put them in order to have the best uh 01:00:31.760 --> 01:00:34.317 efficient use of the cloud resources and 01:00:34.400 --> 01:00:38.717 we also have many sovereign partners locally uh who are helping us ensure 01:00:38.800 --> 01:00:43.597 that all these portfolio of solutions are available uh to all of you. The one 01:00:43.680 --> 01:00:45.597 aspect I wanted to uh talk about though 01:00:45.680 --> 01:00:47.437 at the very end of it is as you think 01:00:47.520 --> 01:00:50.877 about sovereignty and risk, it's also 01:00:50.960 --> 01:00:55.357 important especially for policy makers to think about even cyber and cyber 01:00:55.440 --> 01:00:59.357 security and cyber resilience because if you think about cyber uh it's 01:00:59.440 --> 01:01:03.277 fundamentally an intelligence game, right? It's a global intelligence game, 01:01:03.360 --> 01:01:06.317 right? you're only as resilient on cyber 01:01:06.400 --> 01:01:09.037 to be able to sort of make tap into the 01:01:09.120 --> 01:01:12.077 full intelligence the world has to 01:01:12.160 --> 01:01:17.437 protect yourself. Uh because no cyber attack respects borders. Uh and that I 01:01:17.520 --> 01:01:21.677 think is a very important thing. So when you think about a sovereign service or a 01:01:21.760 --> 01:01:26.317 sovereign cloud and if for what whatever reason it doesn't have the best cyber 01:01:26.400 --> 01:01:30.397 protection that sovereignty unfortunately could get compromised and 01:01:30.480 --> 01:01:32.797 so therefore you really want to think 01:01:32.880 --> 01:01:35.597 about sovereignty and cyber as two parts 01:01:35.680 --> 01:01:38.237 of the risk uh spectrum that you want to 01:01:38.320 --> 01:01:43.437 manage uh simultaneously and that's what uh we have a commitment to ensuring we 01:01:43.520 --> 01:01:47.437 are partnering with in fact local providers uh in bringing that global 01:01:47.520 --> 01:01:49.597 intelligence to all of these portfolio 01:01:49.680 --> 01:01:52.477 solutions. Um, so I want to close out 01:01:52.560 --> 01:01:55.757 where I started. Ultimately, to us, it's 01:01:55.840 --> 01:01:58.397 about taking all of what we talked about 01:01:58.480 --> 01:02:00.557 this power of this new technology, but 01:02:00.640 --> 01:02:03.837 putting it in your hands so that you can 01:02:03.920 --> 01:02:08.077 start to transform your organizations, 01:02:08.160 --> 01:02:12.477 uh, your communities and all of Spain, uh, through the power of these 01:02:12.560 --> 01:02:14.877 technologies and really your ingenuity. 01:02:14.960 --> 01:02:18.797 So, thank you all very very much for your partnership and I look forward to 01:02:18.880 --> 01:02:22.627 sort of seeing what you all build going forward. Thank you and have a great day. 01:02:22.710 --> 01:02:24.730 [applause]