1 00:00:00,000 --> 00:00:02,077 All right, everybody. We're thrilled to 2 00:00:02,160 --> 00:00:04,797 have the one, the only Tata Nadella 3 00:00:04,880 --> 00:00:08,397 here, the third CEO of Microsoft for a 4 00:00:08,480 --> 00:00:11,117 uh impromptu fireside chat with David 5 00:00:11,200 --> 00:00:17,277 Saxs, Arzar of AI and crypto. Satia 6 00:00:17,360 --> 00:00:21,597 third, CEO of Microsoft, born in India. 7 00:00:21,680 --> 00:00:24,157 What an incredible story. came here 8 00:00:24,240 --> 00:00:27,597 right after college and uh you had a 9 00:00:27,680 --> 00:00:29,997 little round trip to pick up your wife 10 00:00:30,080 --> 00:00:31,837 in your book to to bring her here. Tell 11 00:00:31,920 --> 00:00:35,197 everybody briefly uh how that occurred. 12 00:00:35,280 --> 00:00:38,877 >> Well um uh you know so that's that's a 13 00:00:38,960 --> 00:00:42,557 great story of uh the uh the labyrinth 14 00:00:42,640 --> 00:00:45,437 that is the immigration policies of the 15 00:00:45,520 --> 00:00:49,437 United States I think. Um I my wife and 16 00:00:49,520 --> 00:00:51,357 I went to college together in India. I 17 00:00:51,440 --> 00:00:53,677 came here for grad school. we then got 18 00:00:53,760 --> 00:00:58,477 married. Uh I got my green card um and 19 00:00:58,560 --> 00:01:00,557 she couldn't come join because we got 20 00:01:00,640 --> 00:01:04,317 married. So the story goes basically I 21 00:01:04,400 --> 00:01:07,837 had to give up my green card. So the funny thing is I went to the American 22 00:01:07,920 --> 00:01:11,197 embassy in Delhi >> and I said where's the line to give up 23 00:01:11,280 --> 00:01:14,717 my green card and they said there is no such line. Um 24 00:01:14,800 --> 00:01:18,797 >> that would be a crazy thing to do in the '9s. So it was a strange thing to give 25 00:01:18,880 --> 00:01:22,957 up your green card, get an H1 so that she could join, but it all worked out. 26 00:01:23,040 --> 00:01:25,437 So um you know it's a long lost memory 27 00:01:25,520 --> 00:01:27,837 but it was you know a way to work around 28 00:01:27,920 --> 00:01:33,997 it. Um I wanted to ask you uh having 29 00:01:34,080 --> 00:01:37,517 launched a co-pilot first with GitHub 30 00:01:37,600 --> 00:01:39,277 then having a co-pilot on the desktop. 31 00:01:39,360 --> 00:01:43,037 You made a very bold move for Microsoft to put that in the Windows product, 32 00:01:43,120 --> 00:01:47,597 which I use every day, on the desktop, but you did that before it really could 33 00:01:47,680 --> 00:01:52,077 recognize the file system and interact with applications. Got a little bit of a 34 00:01:52,160 --> 00:01:56,317 lukewarm reception, but now you've been doubling down, doubling down, and there 35 00:01:56,400 --> 00:01:58,717 seems to be, in my estimation, three 36 00:01:58,800 --> 00:02:01,597 modalities for knowledge workers. 37 00:02:01,680 --> 00:02:04,397 Elon's building at XAI, 38 00:02:04,480 --> 00:02:08,797 what they're calling a human emulator, if you saw that leak this week. Yeah. 39 00:02:08,880 --> 00:02:12,157 uh where they're just building employees and just putting them into their chat 40 00:02:12,240 --> 00:02:16,557 rooms and email. Then you have Claude came out with co-work this week. 41 00:02:16,640 --> 00:02:19,997 Incredibly powerful. People are kind of losing their minds over it. I've been 42 00:02:20,080 --> 00:02:21,597 playing with it for the last 40 hours. 43 00:02:21,680 --> 00:02:26,637 Truly impressive. What's your vision for Microsoft and how 44 00:02:26,720 --> 00:02:30,077 knowledge workers will actually put this to use because there seems to be a gap 45 00:02:30,160 --> 00:02:33,997 between you know playing around with chatbt and getting some interesting 46 00:02:34,080 --> 00:02:36,477 results and getting business results. 47 00:02:36,560 --> 00:02:39,117 >> Yeah. So I think it one of the most um 48 00:02:39,200 --> 00:02:43,357 perhaps illustrative examples um of 49 00:02:43,440 --> 00:02:46,637 trying to understand these various form factors is looking at coding which is 50 00:02:46,720 --> 00:02:49,837 obviously a form of knowledge work or uh 51 00:02:49,920 --> 00:02:53,837 probably the best example of knowledge work and if you think about the journey 52 00:02:53,920 --> 00:02:56,557 coding has been it started with u 53 00:02:56,640 --> 00:03:00,637 essentially uh uh the next edit suggest 54 00:03:00,720 --> 00:03:02,557 right that was the first time in fact my 55 00:03:02,640 --> 00:03:05,597 own belief in this entire uh generation 56 00:03:05,680 --> 00:03:10,317 of tech really sort of got formulated where I started seeing I think it was G 57 00:03:10,400 --> 00:03:11,997 you know there's a codeex model back in 58 00:03:12,080 --> 00:03:14,637 the day it was preGPT35 59 00:03:14,720 --> 00:03:17,357 uh that's when next edits started 60 00:03:17,440 --> 00:03:19,677 working with some real accuracy then we 61 00:03:19,760 --> 00:03:22,957 went to chat then we went to actions and 62 00:03:23,040 --> 00:03:25,197 now to full autonomous agents and then 63 00:03:25,280 --> 00:03:28,397 the autonomous agents can be both 64 00:03:28,480 --> 00:03:32,877 foreground background in the cloud or local right so that's all the form 65 00:03:32,960 --> 00:03:36,157 factors that exist today when you're coding and interestingly If you look at 66 00:03:36,240 --> 00:03:40,397 it, you use all of them, right? It's not like there's only one form factor. So 67 00:03:40,480 --> 00:03:43,197 that's I think probably one of the other lessons. So for example, when I'm in a 68 00:03:43,280 --> 00:03:47,197 CLI, I can you go a foreground agent, 69 00:03:47,280 --> 00:03:51,997 background agent, and then just literally go edit in VS Code, right? 70 00:03:52,080 --> 00:03:53,837 There all happening in parallel. Right? 71 00:03:53,920 --> 00:03:56,637 So that sort of shows how these form 72 00:03:56,720 --> 00:03:59,757 factors even composed. So then you bring 73 00:03:59,840 --> 00:04:02,237 that to knowledge work to your point. We 74 00:04:02,320 --> 00:04:06,157 started with chat. chat with reasoning 75 00:04:06,240 --> 00:04:08,237 sort of goes beyond just request 76 00:04:08,320 --> 00:04:10,157 response because you now have that chain 77 00:04:10,240 --> 00:04:13,517 of thought uh where you can see it work 78 00:04:13,600 --> 00:04:15,677 now they're actions right essentially 79 00:04:15,760 --> 00:04:18,877 either through computer use or through 80 00:04:18,960 --> 00:04:21,837 uh AP you know basically skills uh and 81 00:04:21,920 --> 00:04:24,237 agent calls so you can do actions so 82 00:04:24,320 --> 00:04:26,157 that's kind of the state of the copilot 83 00:04:26,240 --> 00:04:31,117 today now there is a way to think about 84 00:04:31,200 --> 00:04:34,477 you know the the theory of the mind evolution Right? Because you need like 85 00:04:34,560 --> 00:04:38,397 if you remember you know Jobs had the best line I would say for PCs or 86 00:04:38,480 --> 00:04:41,357 computers was to say if you it's a 87 00:04:41,440 --> 00:04:45,837 bicycle for the mind. Bill had a line which I liked as well which was it's 88 00:04:45,920 --> 00:04:48,077 information at your fingertips. We kind 89 00:04:48,160 --> 00:04:54,397 of need now a new concept metaphor for how we use computers in the AI age. And 90 00:04:54,480 --> 00:04:56,637 you have one >> and the one I like actually came from 91 00:04:56,720 --> 00:05:00,077 the CEO of notion which I you know that manager of >> incredible product. 92 00:05:00,160 --> 00:05:02,477 >> Yeah. You haven't bought it yet? I'm not 93 00:05:02,560 --> 00:05:06,237 not that uh but the it's both management 94 00:05:06,320 --> 00:05:10,957 of you know basically a manager of infinite minds. That's a 95 00:05:11,040 --> 00:05:14,237 nice way to think about it right when you sort of really look at all the 96 00:05:14,320 --> 00:05:19,277 agents that you are working with you kind of need to understand what I in 97 00:05:19,360 --> 00:05:22,877 fact the other term I like is we macrodellegate 98 00:05:22,960 --> 00:05:25,917 and micro steer in fact you kind of need 99 00:05:26,000 --> 00:05:30,077 that in in in coding you kind of have it right. So you do a macro delegation and 100 00:05:30,160 --> 00:05:34,717 then I can in parallel give it instructions while it is doing work. So 101 00:05:34,800 --> 00:05:39,277 that's sort of the state even today of co-pilot or what have you. you bring up 102 00:05:39,360 --> 00:05:42,557 a little bit of what one of one of the form factors I'm very excited about and 103 00:05:42,640 --> 00:05:45,037 you'll see us even in the next week even 104 00:05:45,120 --> 00:05:48,237 uh do things is while I'm sitting in 105 00:05:48,320 --> 00:05:52,637 GitHub copilot what it's not as if software developers 106 00:05:52,720 --> 00:05:54,717 sit in isolation right it's not like the 107 00:05:54,800 --> 00:05:57,517 only thing I work on is my repo I attend 108 00:05:57,600 --> 00:06:01,357 meetings um I write specs or others have 109 00:06:01,440 --> 00:06:04,077 written specs that I'm implementing uh I 110 00:06:04,160 --> 00:06:06,397 need to have my repo be consistent with 111 00:06:06,480 --> 00:06:08,397 that so that means using either a 112 00:06:08,480 --> 00:06:11,197 straightforward MCP server or a skill I 113 00:06:11,280 --> 00:06:13,917 want to be able to call into my work IQ 114 00:06:14,000 --> 00:06:16,557 which is the co-pilot bring that in 115 00:06:16,640 --> 00:06:21,677 that's the type of composition uh of knowledge work that'll happen same thing 116 00:06:21,760 --> 00:06:25,597 with security say you're a security professional you have lots of logs uh 117 00:06:25,680 --> 00:06:29,837 how do you sort of really analyze them you drop them into a file system then 118 00:06:29,920 --> 00:06:33,757 write code on top of it create a dashboard what have you those are the 119 00:06:33,840 --> 00:06:38,077 types of knowledge work that we can enable there I think you bring up One 120 00:06:38,160 --> 00:06:41,837 more thing which is can you create quote unquote digital employees, digital 121 00:06:41,920 --> 00:06:45,997 co-workers or what have you and it's all about credentials right so the I today 122 00:06:46,080 --> 00:06:48,637 you could like you can literally assign >> are you working on that as well 123 00:06:48,720 --> 00:06:53,357 >> yeah so in fact we introduced something called agent 365 as a way to give 124 00:06:53,440 --> 00:06:57,997 identities in fact extending the identities we have for humans today uh 125 00:06:58,080 --> 00:07:02,157 and the endpoint protection we have for their compute devices to agents so 126 00:07:02,240 --> 00:07:06,397 >> so you might clone me working in the HR department or working in the marketing 127 00:07:06,480 --> 00:07:09,677 department and have a virtual correct >> version of me inside of office. 128 00:07:09,760 --> 00:07:12,877 >> That's correct. So, so there are two sort of modalities there. One is you 129 00:07:12,960 --> 00:07:15,277 give every knowledge worker infinite 130 00:07:15,360 --> 00:07:18,237 minds. That's kind of one and then you 131 00:07:18,320 --> 00:07:22,797 create even infinite minds independent of the your identity because the 132 00:07:22,880 --> 00:07:26,557 identity is one of the key things you got to get right even for it to work 133 00:07:26,640 --> 00:07:29,757 right. So permissions >> and decision making 134 00:07:29,840 --> 00:07:31,517 >> permissions decision making and like one 135 00:07:31,600 --> 00:07:36,637 of the key in things is who did what to whom is sort of the most important query 136 00:07:36,720 --> 00:07:40,317 in an organization right at the end of the day the organization needs to 137 00:07:40,400 --> 00:07:43,597 understand what work got done h and 138 00:07:43,680 --> 00:07:47,677 what's the provenence of that work uh and how do you trace it back right so 139 00:07:47,760 --> 00:07:49,677 therefore you kind of want either what 140 00:07:49,760 --> 00:07:52,637 if it's a human with a lot of agents 141 00:07:52,720 --> 00:07:55,117 then it's really macro delegation micro 142 00:07:55,200 --> 00:08:00,317 steering by the human whose identity was passed on. So it's delegation versus a 143 00:08:00,400 --> 00:08:03,117 separate identity >> and that was done by a level of 144 00:08:03,200 --> 00:08:07,437 management, product management that you've eliminated, that Alphabets 145 00:08:07,520 --> 00:08:11,517 eliminated. Meta has started to eliminate in their organization four 146 00:08:11,600 --> 00:08:13,597 years ago. You had the same number of 147 00:08:13,680 --> 00:08:15,837 employees you have at Microsoft now, but 148 00:08:15,920 --> 00:08:18,797 you put a $90 billion onto the top line 149 00:08:18,880 --> 00:08:20,637 of the revenue in that time and you 150 00:08:20,720 --> 00:08:24,717 doubled your income during that time. So 151 00:08:24,800 --> 00:08:28,957 how did that happen? Is that automation of those jobs? Is it you were a little 152 00:08:29,040 --> 00:08:32,957 bit overstaffed? >> Unpack. >> I think it's it's actually you're 153 00:08:33,040 --> 00:08:36,957 pulling on a very interesting thread which is at some level what's the big 154 00:08:37,040 --> 00:08:38,877 structural change that needs to happen? 155 00:08:38,960 --> 00:08:41,197 In fact, I would say this is probably 156 00:08:41,280 --> 00:08:46,557 the biggest change in knowledge work since PCs. I mean I always you know 157 00:08:46,640 --> 00:08:50,397 think about like how did work happen precise right? I mean think about a 158 00:08:50,480 --> 00:08:52,077 multinational company like ours trying 159 00:08:52,160 --> 00:08:55,037 to do a forecast. uh right faxes went 160 00:08:55,120 --> 00:08:57,997 around, inter office memos got sent and 161 00:08:58,080 --> 00:09:02,077 then you kind of created a you know uh a forecast then suddenly you know PCs 162 00:09:02,160 --> 00:09:05,997 became standard issue uh you put an Excel spreadsheet put some numbers sent 163 00:09:06,080 --> 00:09:10,077 it in email everybody entered numbers and you had a forecast so the work the 164 00:09:10,160 --> 00:09:12,717 work artifact and the workflow all 165 00:09:12,800 --> 00:09:15,117 changed that's what's happening so for 166 00:09:15,200 --> 00:09:18,077 example I'll give you uh at LinkedIn 167 00:09:18,160 --> 00:09:20,877 uh we used to have product managers we 168 00:09:20,960 --> 00:09:25,437 had designers uh we had front-end engineers and then we had CIS backend 169 00:09:25,520 --> 00:09:29,997 engineers and so on. So what we did is we sort of took those first four roles 170 00:09:30,080 --> 00:09:32,397 and combined them in fact increased 171 00:09:32,480 --> 00:09:34,557 scope and said let's they're all full 172 00:09:34,640 --> 00:09:37,277 stack builders. So I like that because 173 00:09:37,360 --> 00:09:39,757 that's a structural change that allows 174 00:09:39,840 --> 00:09:43,117 for us to increase the change both the 175 00:09:43,200 --> 00:09:46,797 work and the workflow between these functions and I would assume the 176 00:09:46,880 --> 00:09:50,077 velocity because you don't have four people communicating and that throughput 177 00:09:50,160 --> 00:09:53,437 of ideas it's just one person and vibe coding. Exactly. And there's a new 178 00:09:53,520 --> 00:09:57,277 workflow. So what at the same time as you can imagine if to build an AI 179 00:09:57,360 --> 00:09:59,197 product today, there's a complete new 180 00:09:59,280 --> 00:10:01,277 workflow, right? It starts with eval 181 00:10:01,360 --> 00:10:04,957 right. So basically there's this eval to 182 00:10:05,040 --> 00:10:08,717 science to infrastructure. And so eval 183 00:10:08,800 --> 00:10:12,237 are done by these full stack builders and what have you and product managers 184 00:10:12,320 --> 00:10:15,997 in the new form. The infrastructure is built by the systems engineers at the 185 00:10:16,080 --> 00:10:19,357 back end because they support the science that supports the product. So in 186 00:10:19,440 --> 00:10:22,397 some sense there's a new loop uh and you 187 00:10:22,480 --> 00:10:24,797 have to structurally change and so a lot 188 00:10:24,880 --> 00:10:27,757 of what is happening inside a tech is 189 00:10:27,840 --> 00:10:32,317 that change uh which is I think going to be pretty massive uh and at the same 190 00:10:32,400 --> 00:10:36,077 time a company like ours I have to do everything it's not like I can just so 191 00:10:36,160 --> 00:10:39,757 go live in the future I have to make sure we're doing a fantastic job of 192 00:10:39,840 --> 00:10:44,637 doing hot patching on Windows is done with quality uh while at the same time 193 00:10:44,720 --> 00:10:48,397 building the evals that are improving co-pilot quality right and so both of 194 00:10:48,480 --> 00:10:51,837 those have to be first class. >> I assume this is the most challenging 195 00:10:51,920 --> 00:10:54,637 moment of your career because Microsoft 196 00:10:54,720 --> 00:10:58,157 was so dominant duopoly in some spaces. 197 00:10:58,240 --> 00:11:00,717 Um, but you really weren't up against 198 00:11:00,800 --> 00:11:05,197 the competition level. You're up against now. I was talking to Elon, you know, 199 00:11:05,280 --> 00:11:08,797 and he was sort of saying, well, building cars was pretty easy, uh, 200 00:11:08,880 --> 00:11:13,437 because I was up against the legacy car makers. And now I'm up against 201 00:11:13,520 --> 00:11:15,757 just look at the set you're up against. 202 00:11:15,840 --> 00:11:18,237 Yeah, it's it's a pretty intense time. I 203 00:11:18,320 --> 00:11:20,477 mean, so the way I I I always think is 204 00:11:20,560 --> 00:11:23,517 it's always helpful uh when you have a 205 00:11:23,600 --> 00:11:27,997 complete new set of competitors every decade because that keeps you fit. Uh if 206 00:11:28,080 --> 00:11:29,757 you think about it, I joined uh 207 00:11:29,840 --> 00:11:32,957 Microsoft in ' 92 when I had Noel as the 208 00:11:33,040 --> 00:11:36,317 big existential competitor we had. Um 209 00:11:36,400 --> 00:11:39,837 and here we are in 2026. Uh and you're 210 00:11:39,920 --> 00:11:43,997 absolutely right. It's a pretty intense time. I'm glad there's the competition. 211 00:11:44,080 --> 00:11:47,197 uh it's it's quite honestly at the end of the day when I look at it right as a 212 00:11:47,280 --> 00:11:49,997 percentage of GDP 5 years from now where 213 00:11:50,080 --> 00:11:53,277 will tech be right uh it will be higher 214 00:11:53,360 --> 00:11:56,077 so we're blessed to be in this industry 215 00:11:56,160 --> 00:12:02,157 it's lot of intense competition but it's not so zero sum as some people make it out 216 00:12:02,240 --> 00:12:05,117 >> it's getting much bigger >> much the TAM and the you know just the 217 00:12:05,200 --> 00:12:09,757 impact of this tech is going to be so massive um the question then of course 218 00:12:09,840 --> 00:12:14,397 is what is like I I always go back to what's the brand identity Microsoft has 219 00:12:14,480 --> 00:12:16,317 brand permission. We have what do 220 00:12:16,400 --> 00:12:18,957 customers expect from us. It's sometimes 221 00:12:19,040 --> 00:12:21,997 we kind of overthink somehow that every 222 00:12:22,080 --> 00:12:26,157 customer wants the same thing from all of the competitors and finding that out. 223 00:12:26,240 --> 00:12:29,357 Right. It's kind of a different take on the Peter Theal thing which is you got 224 00:12:29,440 --> 00:12:32,077 to avoid competition by really 225 00:12:32,160 --> 00:12:35,197 understanding what customers really want 226 00:12:35,280 --> 00:12:39,197 from you uh versus thinking everybody's a competitor. 227 00:12:39,280 --> 00:12:41,357 >> David. >> Yeah. So there are a lot of heads of 228 00:12:41,440 --> 00:12:46,157 state here obviously at Davos as well as CEOs of Fortune 500 companies and I 229 00:12:46,240 --> 00:12:49,597 think you got asked a question last night at the dinner about how they 230 00:12:49,680 --> 00:12:51,837 should think about AI and how to be 231 00:12:51,920 --> 00:12:56,717 successful and I recall they used the word diffusion and I wondering if you 232 00:12:56,800 --> 00:13:00,157 could expand on those remarks because that really resonated with some of the 233 00:13:00,240 --> 00:13:03,117 policy work I've been doing. No, absolutely. In fact, uh what you all 234 00:13:03,200 --> 00:13:05,837 have been doing to make sure in in this 235 00:13:05,920 --> 00:13:09,117 context of the American tech stack um is 236 00:13:09,200 --> 00:13:11,997 broadly used around the world and is 237 00:13:12,080 --> 00:13:14,397 trusted around the world because I think 238 00:13:14,480 --> 00:13:18,637 uh when I look back David to me um uh at 239 00:13:18,720 --> 00:13:21,437 the end of the day you create the 240 00:13:21,520 --> 00:13:27,117 technology but really the benefits 241 00:13:27,200 --> 00:13:29,997 come only by intense use. In fact, one 242 00:13:30,080 --> 00:13:34,637 of my favorite studies has always been this work that uh an economist I think 243 00:13:34,720 --> 00:13:39,597 out of Dartmouth did uh his name is Diego Kman where he studied uh basically 244 00:13:39,680 --> 00:13:43,597 what happened during the industrial revolution um how did countries get 245 00:13:43,680 --> 00:13:46,317 ahead. Um and the simple sort of 246 00:13:46,400 --> 00:13:49,357 takeaway from that was any country uh 247 00:13:49,440 --> 00:13:52,157 that brought the latest technology into 248 00:13:52,240 --> 00:13:55,597 their uh country and then did value add 249 00:13:55,680 --> 00:13:59,517 technology on top of it. Right? So it's like don't reinvent the wheel, bring the 250 00:13:59,600 --> 00:14:02,077 latest and then build on top of it. 251 00:14:02,160 --> 00:14:05,677 That's to me what happens, you know, when you have diffusion. So especially 252 00:14:05,760 --> 00:14:09,917 with general purpose technologies like AI, it needs to spread like right in 253 00:14:10,000 --> 00:14:13,837 our, you know, in our own country in the United States, we now need we have the 254 00:14:13,920 --> 00:14:18,317 tech. The question is, is it being used in healthcare? Is it being used um in 255 00:14:18,400 --> 00:14:22,237 financial services? Is it being used in every sector of the economy by large 256 00:14:22,320 --> 00:14:23,517 businesses, small business, public 257 00:14:23,600 --> 00:14:27,357 sector? Um so to me un unless and until 258 00:14:27,440 --> 00:14:31,277 we see that diffusion and intense use uh 259 00:14:31,360 --> 00:14:35,837 we're not going to have the success. Um uh and so that's the phase we are in 260 00:14:35,920 --> 00:14:38,237 it's f you know it's diffusing faster um 261 00:14:38,320 --> 00:14:40,237 and so some of the work policy work you 262 00:14:40,320 --> 00:14:43,677 have done um and in general all you know 263 00:14:43,760 --> 00:14:45,437 the good news here is the technologies 264 00:14:45,520 --> 00:14:48,557 there the rails around cloud and mobile 265 00:14:48,640 --> 00:14:52,717 that were laid out make it possible for this thing to spread right it's not you 266 00:14:52,800 --> 00:14:56,717 know impossible to get the tokens the question is what are the use cases how 267 00:14:56,800 --> 00:15:00,797 do and how do you manage the change in all of that um you know like one of the 268 00:15:00,880 --> 00:15:04,797 questions at least in Davos is it's one for the west and the developed nations. 269 00:15:04,880 --> 00:15:07,197 What about the global south? Um I think 270 00:15:07,280 --> 00:15:09,197 global south has a huge opportunity too 271 00:15:09,280 --> 00:15:11,837 quite frankly because to me like let's 272 00:15:11,920 --> 00:15:15,037 say you know 40% 50% of the GDP of most 273 00:15:15,120 --> 00:15:17,757 global south countries is public sector. 274 00:15:17,840 --> 00:15:21,037 So just imagine this tech making a 275 00:15:21,120 --> 00:15:23,597 difference in how the governments uh 276 00:15:23,680 --> 00:15:28,077 really parlay the taxpayer money into services for citizens and there's if 277 00:15:28,160 --> 00:15:32,077 there's efficiency gains that's probably couple of points for GDP growth right 278 00:15:32,160 --> 00:15:34,957 there and so I'm very optimistic that 279 00:15:35,040 --> 00:15:37,597 there's going to be a pull uh and that 280 00:15:37,680 --> 00:15:39,917 we should as the United States given the 281 00:15:40,000 --> 00:15:43,197 technology stack we have uh in Europe in 282 00:15:43,280 --> 00:15:47,757 Asia in you know in South America in Africa and everywhere 283 00:15:47,840 --> 00:15:49,917 get it to be broadly deployed. 284 00:15:50,000 --> 00:15:51,437 >> You one of the questions I get asked a 285 00:15:51,520 --> 00:15:55,117 lot about the AI race is how do you know 286 00:15:55,200 --> 00:15:59,357 if you're winning or how do you know if the United States is ahead of its global 287 00:15:59,440 --> 00:16:02,877 competitors? And the answer I give is market share. You [clears throat] know, 288 00:16:02,960 --> 00:16:05,517 if we look around the world in 5 years 289 00:16:05,600 --> 00:16:08,557 and we see that American companies, 290 00:16:08,640 --> 00:16:12,557 American technology has say 80% market share, it means we did a good job. If we 291 00:16:12,640 --> 00:16:16,717 look around the world in 5 years and see that it's say Chinese chips and Chinese 292 00:16:16,800 --> 00:16:20,957 models that are being used all over the world, well means we probably lost. So 293 00:16:21,040 --> 00:16:25,757 you know ultimately usage is the proof of the pudding is in the eating of it. I 294 00:16:25,840 --> 00:16:27,837 mean the in this case the way that you 295 00:16:27,920 --> 00:16:31,357 know that you're succeeding is through 296 00:16:31,440 --> 00:16:35,757 market shares through usage. and and I I would agree with that. But David, since 297 00:16:35,840 --> 00:16:40,237 you even worked at Microsoft for a few years, um you I you know, one of the 298 00:16:40,320 --> 00:16:44,317 things that I'm very grounded on is always uh that Bill Gates line of a 299 00:16:44,400 --> 00:16:48,477 platform, right? So, one of the things that I always think about is it's market 300 00:16:48,560 --> 00:16:52,397 share, but it's also ecosystem effects, right? See, what the United States 301 00:16:52,480 --> 00:16:55,037 always has done is not just about our 302 00:16:55,120 --> 00:16:58,717 market share or even um the revenues to 303 00:16:58,800 --> 00:17:02,557 US companies. In fact, one of the things I learned at Microsoft is whenever I did 304 00:17:02,640 --> 00:17:04,717 a country visit, the data I would first 305 00:17:04,800 --> 00:17:07,997 study is in let's say in the UK or in 306 00:17:08,080 --> 00:17:09,757 Switzerland or what have you is what is 307 00:17:09,840 --> 00:17:12,637 the total employment created in 308 00:17:12,720 --> 00:17:15,597 Switzerland uh in our channel that used 309 00:17:15,680 --> 00:17:20,237 to be like the number one thing uh in our country reports right and the total number 310 00:17:20,320 --> 00:17:22,957 >> that be like the number of IT workers the number 311 00:17:23,040 --> 00:17:24,957 >> exactly we used to have office workers 312 00:17:25,040 --> 00:17:27,757 channel so channel partners we ISVS so 313 00:17:27,840 --> 00:17:30,957 number of ISVS uh who were there. So we 314 00:17:31,040 --> 00:17:33,117 used to have a complete marker of how 315 00:17:33,200 --> 00:17:35,517 did the ecosystem around the platform 316 00:17:35,600 --> 00:17:38,397 get built one country at a time and that 317 00:17:38,480 --> 00:17:40,157 is what the United States has always 318 00:17:40,240 --> 00:17:42,877 done. In fact the US tech stack 319 00:17:42,960 --> 00:17:46,157 including in China got built because 320 00:17:46,240 --> 00:17:50,237 others built around our tech stack. The same thing is going to happen. So that's 321 00:17:50,320 --> 00:17:52,877 why I think the work you're doing around diffusion, >> right, 322 00:17:52,960 --> 00:17:55,757 >> is about really increasing the size of 323 00:17:55,840 --> 00:18:01,357 the pie, the trust in the platform so that there is true economic opportunity 324 00:18:01,440 --> 00:18:03,837 quite frankly. >> Well, you're right and and I remember 325 00:18:03,920 --> 00:18:07,997 actually you brought back some memories from this is about a decade ago when my 326 00:18:08,080 --> 00:18:11,437 company Yammer was acquired by Microsoft. We were part of uh the 327 00:18:11,520 --> 00:18:14,077 SharePoint group and I remember that the 328 00:18:14,160 --> 00:18:15,917 um the product managers there were very 329 00:18:16,000 --> 00:18:19,037 proud of the fact that the revenue from 330 00:18:19,120 --> 00:18:21,597 the SharePoint ecosystem meaning non- 331 00:18:21,680 --> 00:18:24,637 Microsoft the um the consulting 332 00:18:24,720 --> 00:18:28,637 community the implementers who would go into companies implement SharePoint I 333 00:18:28,720 --> 00:18:32,717 think their revenue is something like seven times greater than Microsoft's own 334 00:18:32,800 --> 00:18:35,117 software revenue and I think >> in aggregate 335 00:18:35,200 --> 00:18:39,117 >> in aggregate and I think and I think Bill had a line about you're not an 336 00:18:39,200 --> 00:18:43,677 ecosystem platform until the revenue on top of your platform 337 00:18:43,760 --> 00:18:48,477 >> is some, you know, factor of your own revenue. And and I and I think that I 338 00:18:48,560 --> 00:18:52,317 think what's really important about this is when we talk about diffusion and 339 00:18:52,400 --> 00:18:56,557 obviously want the United States to have this leading position, it doesn't mean 340 00:18:56,640 --> 00:18:59,757 it's bad for the rest of the world because they're able to build on top of 341 00:18:59,840 --> 00:19:02,477 those platforms and create even more value. 342 00:19:02,560 --> 00:19:04,477 >> 100%. In fact, that's sort of the most 343 00:19:04,560 --> 00:19:06,237 important point, right? So this is not 344 00:19:06,320 --> 00:19:11,597 that this is not about uh American tech um and America revenues to the United 345 00:19:11,680 --> 00:19:15,037 States. It's actually creating opportunity using a new platform 346 00:19:15,120 --> 00:19:17,357 everywhere. And in fact you know the you 347 00:19:17,440 --> 00:19:19,517 know like I remember I worked on our 348 00:19:19,600 --> 00:19:22,317 database products uh in the '90s u you 349 00:19:22,400 --> 00:19:25,677 know with SAP in fact the combination of 350 00:19:25,760 --> 00:19:29,197 uh SQL Server and R3 were successful on 351 00:19:29,280 --> 00:19:32,717 both sides. There's a lot talked about Intel and Microsoft, but one of the 352 00:19:32,800 --> 00:19:36,477 other things that I grew up in which has sort of been foundational in how I look 353 00:19:36,560 --> 00:19:40,797 at the world is what we did with a European software company that is still 354 00:19:40,880 --> 00:19:45,197 uh you know a giant. And so that you know who knows what the next big AI app 355 00:19:45,280 --> 00:19:47,277 will be and where and what will happen. 356 00:19:47,360 --> 00:19:49,837 But uh I I sort of go in with the 357 00:19:49,920 --> 00:19:54,637 attitude that there will be tech companies uh maybe even top five tech 358 00:19:54,720 --> 00:19:56,797 companies that could emerge everywhere 359 00:19:56,880 --> 00:20:02,077 with even the American tech stack. you have um done some amazing acquisitions 360 00:20:02,160 --> 00:20:05,197 and you're quite a dealmaker on top of being a technologist. It's probably the 361 00:20:05,280 --> 00:20:08,157 least reported aspect of your 362 00:20:08,240 --> 00:20:09,997 spectacular tenure and the massive 363 00:20:10,080 --> 00:20:12,397 growth you've had. 364 00:20:12,480 --> 00:20:15,837 But you did a deal with OpenAI and 365 00:20:15,920 --> 00:20:18,077 probably one of the most savvy 366 00:20:18,160 --> 00:20:19,917 slashcontroversial dealmakers of all 367 00:20:20,000 --> 00:20:23,037 time, Sam Alman. That deal was looked at 368 00:20:23,120 --> 00:20:27,517 as you you you're you're set up to get a 369 00:20:27,600 --> 00:20:31,517 windfall in cash which you don't need as Microsoft always nice I'm guessing if 370 00:20:31,600 --> 00:20:34,237 they IPO but did you create potentially 371 00:20:34,320 --> 00:20:37,197 and this was the criticism of it 372 00:20:37,280 --> 00:20:40,157 an ultimate competitor to Microsoft and 373 00:20:40,240 --> 00:20:44,557 how do you think about that and how can Microsoft which missed Steve Balmer's 374 00:20:44,640 --> 00:20:46,157 biggest regret missing the mobile 375 00:20:46,240 --> 00:20:50,317 revolution how can you not have a Gemini 376 00:20:50,400 --> 00:20:53,917 an XAI a claude that is your own or in 377 00:20:54,000 --> 00:20:57,597 your mind do you have that because you have the source code of open AI? 378 00:20:57,680 --> 00:21:00,797 >> Yeah, I think that that's right. So when when people say uh where is your 379 00:21:00,880 --> 00:21:04,637 foundation model I mean at the end of the day we do have the IP but that said 380 00:21:04,720 --> 00:21:08,797 I think you bring up a couple different things right one is to us the most 381 00:21:08,880 --> 00:21:10,877 important thing when I look at what is 382 00:21:10,960 --> 00:21:14,637 Microsoft's uh strategy today one is we 383 00:21:14,720 --> 00:21:18,397 want to build token factories right so our biggest business today is Azure 384 00:21:18,480 --> 00:21:20,557 business and the Azure business the TAM 385 00:21:20,640 --> 00:21:22,877 given what's going to happen is is so 386 00:21:22,960 --> 00:21:25,437 huge that we now need to be fantastic uh 387 00:21:25,520 --> 00:21:29,917 at building these token factories and um that's means a heterogeneous fleet of 388 00:21:30,000 --> 00:21:33,997 infrastructure and that every hyperscaler has always done which is use 389 00:21:34,080 --> 00:21:36,637 software to make maximum use of it and 390 00:21:36,720 --> 00:21:41,197 for TCO and utilization. So that's one side of it. Then there's the app server 391 00:21:41,280 --> 00:21:44,877 business right which is everybody we you talked about like if everyone's going to 392 00:21:44,960 --> 00:21:49,997 be building agents have infinite minds have these RL gyms have eval what have 393 00:21:50,080 --> 00:21:54,077 you there's an entire just like every platform has had an app server this one 394 00:21:54,160 --> 00:21:57,117 has an app server that's what we're doing with foundry and what have you 395 00:21:57,200 --> 00:22:01,197 right so there's an app server business in that app server one of the things 396 00:22:01,280 --> 00:22:03,917 that structurally now is pretty clear is 397 00:22:04,000 --> 00:22:08,717 anyone building any application or any company is going to use not one model 398 00:22:08,800 --> 00:22:12,557 but all the models Right? Why would I not? Right? Which is in fact I will 399 00:22:12,640 --> 00:22:17,677 orchestrate for any given task even multiple models. Right? There's this one 400 00:22:17,760 --> 00:22:21,197 nice thing that we came out in our healthcare practice called the decision 401 00:22:21,280 --> 00:22:24,477 orchestrator. What it proves is that by 402 00:22:24,560 --> 00:22:27,517 assigning roles, right? So investigator, 403 00:22:27,600 --> 00:22:30,317 data analyst, domain expert, just giving 404 00:22:30,400 --> 00:22:33,437 even prompted roles to models and then 405 00:22:33,520 --> 00:22:35,917 orchestrating them gets better results 406 00:22:36,000 --> 00:22:38,397 than any one single frontier model. Am I 407 00:22:38,480 --> 00:22:43,037 right to read into that then that you're bullish on the open- source models and 408 00:22:43,120 --> 00:22:47,117 think large language models will largely be commoditized and that's not where the 409 00:22:47,200 --> 00:22:49,437 value will occur. >> In fact, the way I think about it is 410 00:22:49,520 --> 00:22:52,477 that just like Apple thinks that too. By the way, 411 00:22:52,560 --> 00:22:54,317 >> by the way, what the way you think about 412 00:22:54,400 --> 00:22:57,837 what happened in the database market, right? You know, I used to be like 413 00:22:57,920 --> 00:23:01,597 everything is just a SQL database until it was not, right? There was I mean, 414 00:23:01,680 --> 00:23:05,597 think about it. There doc databases, there is no SQL databases. the 415 00:23:05,680 --> 00:23:09,597 proliferation of database right who would have thought that the database 416 00:23:09,680 --> 00:23:13,597 market would have such a richness to it >> or that it could ever be open source 417 00:23:13,680 --> 00:23:16,957 that was that's I mean talk about Postgress or what has happened even with 418 00:23:17,040 --> 00:23:20,637 which is open but there are even companies that have backed it and so so 419 00:23:20,720 --> 00:23:24,877 to me that's what's going to happen it to me a model is like the database 420 00:23:24,960 --> 00:23:28,637 market you know it's it's got it's got differences but I sort of somehow think 421 00:23:28,720 --> 00:23:32,077 that uh it's not there are definitely going to be frontier models that are 422 00:23:32,160 --> 00:23:35,757 closed source you know there going to be open source models that are going to be 423 00:23:35,840 --> 00:23:38,317 uh uh frontier class. In fact, if 424 00:23:38,400 --> 00:23:40,717 anything, I think in this next year, 425 00:23:40,800 --> 00:23:43,357 what will be probably a big part of the 426 00:23:43,440 --> 00:23:46,077 discussion is what's the future of a 427 00:23:46,160 --> 00:23:50,717 firm? A firm should be able to take the 428 00:23:50,800 --> 00:23:54,797 tacit knowledge it has and embed it 429 00:23:54,880 --> 00:23:59,677 inside a weights in a model that they control. Right? So, when somebody asks 430 00:23:59,760 --> 00:24:03,357 me how many models should be there, I'll say as many models as firms in the 431 00:24:03,440 --> 00:24:04,797 world. Right? That's sort of the an 432 00:24:04,880 --> 00:24:07,837 extreme way uh because because to me 433 00:24:07,920 --> 00:24:10,397 that's how I think this you know this 434 00:24:10,480 --> 00:24:12,557 knowledge economy becomes an AI economy. 435 00:24:12,640 --> 00:24:15,037 >> Are you secretly and you can say it here 436 00:24:15,120 --> 00:24:18,557 since we're on allin working on an LLM 437 00:24:18,640 --> 00:24:22,237 to exist on the Windows desktop because that you are 438 00:24:22,320 --> 00:24:23,837 >> you have it like today there's a five 439 00:24:23,920 --> 00:24:26,397 silica model which is completely 440 00:24:26,480 --> 00:24:29,757 resident using NPUs and of course using 441 00:24:29,840 --> 00:24:34,397 uh GPUs. in fact the largest installation um of high power. In fact, 442 00:24:34,480 --> 00:24:37,357 it's one of the fascinating the workstation is back. I'm one of the most 443 00:24:37,440 --> 00:24:42,077 if you went to see each great for Microsoft because you you have a nice 444 00:24:42,160 --> 00:24:44,957 desktop business. >> Absolutely. And so we and in fact we 445 00:24:45,040 --> 00:24:47,437 think that that form factor especially I 446 00:24:47,520 --> 00:24:52,477 mean I I always say this which is u you know I started my career on a command 447 00:24:52,560 --> 00:24:55,037 line. Who knows I may just end it in a command line. 448 00:24:55,120 --> 00:24:59,277 >> Well you started at Sun which was the original 5 $10,000 workstation. Do you 449 00:24:59,360 --> 00:25:01,437 see a time where you'll be meeting with 450 00:25:01,520 --> 00:25:03,757 your customers here and advocating a 10 451 00:25:03,840 --> 00:25:07,117 $20,000 desktop machine that has an LLM 452 00:25:07,200 --> 00:25:09,357 and the hardware? You can you can put a 453 00:25:09,440 --> 00:25:14,717 DGX card and you can have like just a fantastic machine and the models I and 454 00:25:14,800 --> 00:25:18,877 by the way you know we are one architecture tweak away from even having 455 00:25:18,960 --> 00:25:22,877 some kind of a distributed model architecture right even ane architecture 456 00:25:22,960 --> 00:25:27,197 that shows knows how to really distribute itself right that's the type 457 00:25:27,280 --> 00:25:29,357 of breakthrough that can completely 458 00:25:29,440 --> 00:25:32,317 change uh what hybrid AI may look like 459 00:25:32,400 --> 00:25:34,557 but we're absolutely committed and 460 00:25:34,640 --> 00:25:37,437 focused on making the PC a great place 461 00:25:37,520 --> 00:25:42,397 for local models uh and local models that then do even a lot of the prompt 462 00:25:42,480 --> 00:25:45,517 processing and call into the cloud, right? So there's a whole lot of work 463 00:25:45,600 --> 00:25:48,557 that can happen and that's sort of definitely something that's underway. 464 00:25:48,640 --> 00:25:50,877 >> Yeah, I think that the cloud co-work has 465 00:25:50,960 --> 00:25:53,997 kind of shown the power of tapping into the local file drive and be able to use 466 00:25:54,080 --> 00:25:56,077 that. That that brings up another point. 467 00:25:56,160 --> 00:25:59,357 You you got me thinking about Yammer and for people who don't know um you know 468 00:25:59,440 --> 00:26:03,437 Yammer's claim to fame, this is about 15 years ago, was that it pioneered a lot 469 00:26:03,520 --> 00:26:07,437 of um well it used a lot of consumer growth tactics to attack enterprise 470 00:26:07,520 --> 00:26:09,357 software. I'm wondering as you think 471 00:26:09,440 --> 00:26:13,117 about enterprise adoption of AI. How do 472 00:26:13,200 --> 00:26:16,477 you think it's going to spread over the next year? It feels like we're at sort 473 00:26:16,560 --> 00:26:21,117 of a a a critical point. Do you think it's going to be top down? Is it going 474 00:26:21,200 --> 00:26:25,917 to come from the CEO directing a team, giving them a strategic transformation 475 00:26:26,000 --> 00:26:29,837 project, and they're going to do an RFP, or do you think it's going to spread 476 00:26:29,920 --> 00:26:32,637 bottom up in the enterprise through AI 477 00:26:32,720 --> 00:26:37,517 native employees who are adaptable, who are using the tools in their own lives 478 00:26:37,600 --> 00:26:41,277 and they start to bring these things to work and start accomplishing amazing 479 00:26:41,360 --> 00:26:43,357 things. >> Yeah. No, I think, you know, like all 480 00:26:43,440 --> 00:26:47,277 things, David, I think it's both the top down, bottom up, right? uh the that the 481 00:26:47,360 --> 00:26:49,757 reason I say that top down is if I look 482 00:26:49,840 --> 00:26:53,277 at the ROI uh of uh applying AI in 483 00:26:53,360 --> 00:26:56,317 customer service uh or in supply chain 484 00:26:56,400 --> 00:26:59,677 or in HR self-service those are the easy 485 00:26:59,760 --> 00:27:03,917 projects where uh IT and CXOs can make 486 00:27:04,000 --> 00:27:05,917 calls and that's where you'll see the 487 00:27:06,000 --> 00:27:09,437 first drop of uh real AI adoption but 488 00:27:09,520 --> 00:27:13,917 the bottom up is what ultimately will happen right I mean with even with the 489 00:27:14,000 --> 00:27:16,077 PCs in fact if you think back at the 490 00:27:16,160 --> 00:27:18,797 lawyers brought word in and then finance 491 00:27:18,880 --> 00:27:23,197 bought Excel in and then email came and then it became standard issue. That's 492 00:27:23,280 --> 00:27:27,197 what's happening right now. So for example, these agents when I sort of 493 00:27:27,280 --> 00:27:29,117 talk about everybody's building agents, 494 00:27:29,200 --> 00:27:33,197 they are figuring out a way to go create 495 00:27:33,280 --> 00:27:37,437 these things that are changing workflow and removing drudgery in their work. 496 00:27:37,520 --> 00:27:40,157 Right? That's sort of the beginning of 497 00:27:40,240 --> 00:27:43,117 what is a bottomup transformation. Um I 498 00:27:43,200 --> 00:27:47,197 you know you I was in fact the thing that I'm most excited about is this 499 00:27:47,280 --> 00:27:48,877 bottomup change even at Microsoft for 500 00:27:48,960 --> 00:27:52,237 example we manage something like 500 odd 501 00:27:52,320 --> 00:27:56,797 fiber operators around the world in in Azure today and by the way I had not 502 00:27:56,880 --> 00:28:00,477 myself realized it a lot of it you know it's called DevOps but it's a it's a 503 00:28:00,560 --> 00:28:04,957 physical asset things get cut and when you sort of say DevOps that means you 504 00:28:05,040 --> 00:28:08,397 literally are emailing people and saying hey what happened to that fiber cut how 505 00:28:08,480 --> 00:28:10,557 do we repair it so there's a lot of back 506 00:28:10,640 --> 00:28:15,597 and forth so this network the the person who runs our global network basically 507 00:28:15,680 --> 00:28:19,357 has built to your point about these person they're just digital employees 508 00:28:19,440 --> 00:28:21,997 essentially that are doing all of that 509 00:28:22,080 --> 00:28:25,277 devops uh and so that's and there's a 510 00:28:25,360 --> 00:28:29,357 completely bottoms up uh where you see the tools it's kind of like hey I have 511 00:28:29,440 --> 00:28:31,997 the new way to build agents it's there 512 00:28:32,080 --> 00:28:34,237 I'm going to use it to create levels of 513 00:28:34,320 --> 00:28:36,717 automation uh that remove drudgery 514 00:28:36,800 --> 00:28:39,197 improve efficiency improve quality and 515 00:28:39,280 --> 00:28:41,757 that ultimately is a skilling thing 516 00:28:41,840 --> 00:28:46,237 which is sort of the big issue which is u and skilling is not mystical it's just 517 00:28:46,320 --> 00:28:48,157 by doing right so it's not like I go to 518 00:28:48,240 --> 00:28:50,397 a class per se it's like the diffusion 519 00:28:50,480 --> 00:28:55,517 of the tools uh and using the tools and that I think is what's really going to 520 00:28:55,600 --> 00:28:57,437 be happening >> and and we're in a very interesting 521 00:28:57,520 --> 00:29:00,957 moment empowering an existing employee 522 00:29:01,040 --> 00:29:03,917 with these tools is so much easier than 523 00:29:04,000 --> 00:29:08,077 hiring and mentoring and bringing up the next generation so it feels like we're 524 00:29:08,160 --> 00:29:13,117 in a little bit of an indigestion moment at Microsoft, do you think who's going 525 00:29:13,200 --> 00:29:14,797 to have my job 526 00:29:14,880 --> 00:29:17,677 >> in 30 or 40 years if the company stays 527 00:29:17,760 --> 00:29:20,957 the same size? Because given your 528 00:29:21,040 --> 00:29:24,717 technology first approach, there's really no reason to ever add another 529 00:29:24,800 --> 00:29:26,717 Microsoft employee at the pace this is 530 00:29:26,800 --> 00:29:31,117 going and you haven't for years. So, how 531 00:29:31,200 --> 00:29:34,637 you may have swapped some in and out and changed the texture of it. So, how do 532 00:29:34,720 --> 00:29:38,717 you think about maybe this next generation? What advice would you have 533 00:29:38,800 --> 00:29:42,077 for these college graduates who maybe don't have an offer from Microsoft right 534 00:29:42,160 --> 00:29:46,317 now? And you used to spend a lot of time on that building that group, but maybe 535 00:29:46,400 --> 00:29:49,437 you don't have that luxury now. Do you think about it ever? 536 00:29:49,520 --> 00:29:52,797 >> I I it's a great question. I you know there's a little bit of a debate what 537 00:29:52,880 --> 00:29:56,957 happens to early in career and how is college recruiting. I still am a big 538 00:29:57,040 --> 00:29:58,717 believer in uh college recruiting 539 00:29:58,800 --> 00:30:02,877 because at the end of the day um this is 540 00:30:02,960 --> 00:30:05,677 going to change the curve by which 541 00:30:05,760 --> 00:30:08,317 anyone can pick up proficiency in a 542 00:30:08,400 --> 00:30:10,317 codebase let's just takes sort of just 543 00:30:10,400 --> 00:30:14,157 regular CS hiring u what has changed is 544 00:30:14,240 --> 00:30:16,797 perhaps for someone who comes in new 545 00:30:16,880 --> 00:30:20,477 into a team and to be able to ramp up 546 00:30:20,560 --> 00:30:24,637 thanks to all of uh the markdowns the 547 00:30:24,720 --> 00:30:28,397 skills uh the fact that I can go ask the agent. 548 00:30:28,480 --> 00:30:33,117 I mean, think about it, right? It's like having an unbelievable mentor who is 549 00:30:33,200 --> 00:30:35,197 getting you onboarded onto a codebase 550 00:30:35,280 --> 00:30:36,797 faster. So, in some sense, the 551 00:30:36,880 --> 00:30:40,237 productivity curve uh of a college hire 552 00:30:40,320 --> 00:30:45,117 is going to be much steeper than it ever before. So, I think there might be a 553 00:30:45,200 --> 00:30:47,597 difference. In fact, one of the things we're experimenting with is a different 554 00:30:47,680 --> 00:30:49,597 type of apprenticeship, right? which is 555 00:30:49,680 --> 00:30:53,037 you take somebody who's an IC senior dev 556 00:30:53,120 --> 00:30:56,477 have like a cohort of college uh hires 557 00:30:56,560 --> 00:31:00,637 working with them because it's a new way of working it's like I remember like all 558 00:31:00,720 --> 00:31:04,477 you know everybody who joined Microsoft would say go how did you know whatever 559 00:31:04,560 --> 00:31:07,437 um Cutler implement Malik or what have 560 00:31:07,520 --> 00:31:09,997 you right he would go try to read uh his 561 00:31:10,080 --> 00:31:12,397 code to understand uh what great 562 00:31:12,480 --> 00:31:14,957 craftsmanship looks like nowadays I 563 00:31:15,040 --> 00:31:17,117 think that great craftsmanship uh comes 564 00:31:17,200 --> 00:31:20,157 by looking at even how the 10x 100x 565 00:31:20,240 --> 00:31:24,317 engineers use AI to build great quality 566 00:31:24,400 --> 00:31:29,277 products. uh and that is what these new college grads will learn and learn 567 00:31:29,360 --> 00:31:32,957 faster and so that's a beneficial thing for a company like us because at the end 568 00:31:33,040 --> 00:31:34,637 of the day you know until we saw 569 00:31:34,720 --> 00:31:37,837 longevity or something we need people to 570 00:31:37,920 --> 00:31:41,837 come into the workforce be successful at Microsoft so we are very committed but 571 00:31:41,920 --> 00:31:46,877 we are also making sure that the scopes of the jobs make sense for what the 572 00:31:46,960 --> 00:31:50,157 aspirations of people are going to be both who are currently in the workforce 573 00:31:50,240 --> 00:31:52,637 and people who are entering the workforce. 574 00:31:52,720 --> 00:31:58,570 Okay, on that note, Sache Nadella, >> thank you so much. [applause]