1 00:00:00,120 --> 00:00:03,960 I became go from no agent to an agent within a couple hours, 2 00:00:04,440 --> 00:00:08,560 hosted, deployed in production. I don't look at code anymore. 3 00:00:10,960 --> 00:00:14,520 I literally look at the outcome. And that's a kind of wrong 4 00:00:14,600 --> 00:00:21,080 assumption is that start with the audience start like what are 5 00:00:21,080 --> 00:00:24,440 you trying to do? What is the the target audience 6 00:00:24,440 --> 00:00:28,400 of this agent? Kind of wrong assumption is that 7 00:00:28,440 --> 00:00:32,080 you will build a better orchestration layer than 8 00:00:32,080 --> 00:00:36,200 Microsoft 365 Copilot kit. Today's guest is Sebastian Nova, 9 00:00:36,360 --> 00:00:39,400 Principal Product Manager for Copilot Extensibility at 10 00:00:39,400 --> 00:00:43,520 Microsoft, where he focuses on the Pro code tools that let 11 00:00:43,560 --> 00:00:47,240 teams build deeper, more controlled agent inside. 12 00:00:47,240 --> 00:00:51,480 Microsoft 365 Copilot. This episode looks at how agent 13 00:00:51,480 --> 00:00:55,760 development changes when it become real software, not just 14 00:00:55,760 --> 00:00:59,520 quick prototype. From deciding when low code 15 00:00:59,520 --> 00:01:03,360 stops being enough, to building shared ownership, to source 16 00:01:03,360 --> 00:01:08,280 control and teamwork flows, the conversations show what it takes 17 00:01:08,400 --> 00:01:13,040 to create agents that skill. Sebastian also breaks down the 18 00:01:13,040 --> 00:01:16,960 practical benefits of pro coding agents, including richer 19 00:01:16,960 --> 00:01:21,000 connections to 3rd party systems and the ability to ship with 20 00:01:21,080 --> 00:01:24,400 proper reversioning environments and CICD. 21 00:01:25,240 --> 00:01:30,400 Now let's dive in. Hello, Sebastian, and welcome to 22 00:01:30,400 --> 00:01:33,680 Mastering AI with the Experts. Thank you so much for joining us 23 00:01:33,680 --> 00:01:36,160 today. Thanks for having me. 24 00:01:36,160 --> 00:01:40,840 Actually, it's good to to have a little chat while we're having a 25 00:01:40,840 --> 00:01:43,200 very snowy day here in Montreal. Yes, it is. 26 00:01:43,200 --> 00:01:45,640 It is very snowy and very cold as well. 27 00:01:46,280 --> 00:01:51,200 Yes, Debasi, you're a principal product manager for Copilot 28 00:01:51,200 --> 00:01:55,280 extensibility at Microsoft. Can you please start by briefly 29 00:01:55,280 --> 00:01:57,800 describing what it means actually in which part of 30 00:01:57,800 --> 00:02:00,520 extensibility of Copilot you're working on? 31 00:02:01,360 --> 00:02:05,320 Yeah. So my team is the copilot 32 00:02:05,320 --> 00:02:09,440 extensibility and development platform for Microsoft 365 33 00:02:09,440 --> 00:02:11,520 Copilot. So we are the team that is 34 00:02:11,520 --> 00:02:16,960 responsible for building the agent platform in Microsoft 365 35 00:02:16,960 --> 00:02:18,960 Copilot. So that means bringing all the 36 00:02:18,960 --> 00:02:24,480 capability that Copilot has by default and exposing them as 37 00:02:24,480 --> 00:02:28,320 capabilities in agents. When you're using for example, 38 00:02:28,320 --> 00:02:32,240 the agent builder that is built into M365 copilot, you're going 39 00:02:32,240 --> 00:02:36,160 to be able to add sites, add connectors, do code 40 00:02:36,160 --> 00:02:39,120 interpretation, do do graphic art and so on. 41 00:02:39,320 --> 00:02:42,680 All these capabilities are what our team is kind of exposing to 42 00:02:42,680 --> 00:02:44,200 that. So we have a strong partnership 43 00:02:44,200 --> 00:02:47,800 with the copilot studio team that is building the UX for 44 00:02:47,800 --> 00:02:50,320 building agents. So we have that and my role 45 00:02:50,320 --> 00:02:53,440 specifically is for another audience, another an audience 46 00:02:53,440 --> 00:02:57,160 that is bringing agents the same way but built with different 47 00:02:57,160 --> 00:03:02,400 tools really mainly for the developer audience or enterprise 48 00:03:02,400 --> 00:03:05,640 Oregon LOB developers that wants to bring a little bit more 49 00:03:05,640 --> 00:03:08,880 control over the urgent into MP-65 copilot. 50 00:03:09,360 --> 00:03:12,080 Great. So we're going to talk about Pro 51 00:03:12,080 --> 00:03:14,360 Code Tools today to extend Coopala. 52 00:03:14,480 --> 00:03:18,760 But just before we jump in, when you talk about agent 53 00:03:18,760 --> 00:03:22,200 extensibility like in Microsoft 365 Coopala today, I know it's 54 00:03:22,200 --> 00:03:24,440 not clear for everyone what an agent is. 55 00:03:25,040 --> 00:03:28,920 How can you extend Gopala with agents? 56 00:03:28,920 --> 00:03:32,440 So what does that actually mean in new concrete terms? 57 00:03:32,440 --> 00:03:36,640 And why should like both businesses and technical leader 58 00:03:37,120 --> 00:03:42,600 care about this right now? I think agents are ways to 59 00:03:42,920 --> 00:03:47,520 automate business processes. You are AI don't know an 60 00:03:47,520 --> 00:03:51,320 engineering team and you want to be able to handle the the 61 00:03:51,320 --> 00:03:55,160 tickets, the support things and you want to automate some of 62 00:03:55,160 --> 00:03:58,200 them. You want to use AI to go deeper 63 00:03:58,200 --> 00:04:01,480 into your logs and infer some context from it. 64 00:04:02,480 --> 00:04:05,600 That's what really agents are for us. 65 00:04:05,600 --> 00:04:09,440 We believe that why would you even care? 66 00:04:09,440 --> 00:04:14,320 Is it's because you want to use AI for more than just plugging 67 00:04:14,320 --> 00:04:18,200 into your productivity data? Yes, productivity, productivity 68 00:04:18,200 --> 00:04:21,920 data is amazing and it can do really, really cool things with 69 00:04:21,920 --> 00:04:24,120 that. But what if you want to connect 70 00:04:24,120 --> 00:04:29,320 to your own third party systems or to your own SAS that you're 71 00:04:29,320 --> 00:04:31,160 using or to your own internal tools? 72 00:04:31,160 --> 00:04:35,040 So that's really why we think that extensibility to bring a 73 00:04:35,440 --> 00:04:40,400 Copilot a little bit further, make it even more integrated to 74 00:04:40,400 --> 00:04:45,520 your business so it becomes your de facto like landing page. 75 00:04:45,560 --> 00:04:50,120 I go to Copilot, I click on my service ticket agent and now I 76 00:04:50,120 --> 00:04:53,120 can do everything from there. I don't need to go to my system. 77 00:04:53,120 --> 00:04:55,520 I don't. Then I can cross reference so 78 00:04:55,520 --> 00:04:58,040 many things around the house that it, it just makes sense. 79 00:04:58,040 --> 00:05:00,680 So that's really how I'm thinking about why they should 80 00:05:00,680 --> 00:05:03,040 care. Automate your business processes 81 00:05:03,320 --> 00:05:05,920 and make sure that everything fits within Copilot. 82 00:05:06,960 --> 00:05:09,080 Though it's not just about retrieving data, right? 83 00:05:09,080 --> 00:05:13,160 It's also acting on data, acting on 3rd party systems. 84 00:05:13,360 --> 00:05:16,200 You were, you were mentioning like organization using their 85 00:05:16,200 --> 00:05:18,320 own SAS, using their own third party systems. 86 00:05:18,720 --> 00:05:24,280 So it's bringing those workflow that was hysterically done by 87 00:05:24,560 --> 00:05:26,960 human being. Let's let's be real. 88 00:05:27,400 --> 00:05:32,240 And having an agent go through them so directly in M365 89 00:05:32,240 --> 00:05:33,880 Copilot, right? Why? 90 00:05:33,880 --> 00:05:38,120 What was the advantage of doing it in Copilot versus, you know, 91 00:05:38,120 --> 00:05:41,240 having a stand alone agent, even if it's built with the Copilot 92 00:05:41,240 --> 00:05:43,760 studio, for instance, or the the agent toolkit that we'll we'll 93 00:05:43,760 --> 00:05:47,520 talk a bit more later. I think the, the biggest 94 00:05:47,600 --> 00:05:52,720 advantage of bringing that to M365 Copilot and build what we 95 00:05:52,720 --> 00:05:58,200 call the Clarative agent with our basically really native M365 96 00:05:58,200 --> 00:06:01,960 Copilot agents is the proximity to your productivity data. 97 00:06:02,760 --> 00:06:06,000 And that's where it unlocks so many things. 98 00:06:06,560 --> 00:06:10,000 Your productivity, your productivity data, sorry, has so 99 00:06:10,000 --> 00:06:14,400 much potential that when you go to your SAS, you're losing all 100 00:06:14,400 --> 00:06:18,120 of that context. But now you can cross reference 101 00:06:18,320 --> 00:06:21,600 the e-mail you got, the conversations and teams you got 102 00:06:22,160 --> 00:06:26,040 with your third party SAS now, because then it can say, oh, I 103 00:06:26,040 --> 00:06:28,560 had a conversation with John yesterday about this ticket. 104 00:06:28,920 --> 00:06:31,400 He, he gave me a good way to respond. 105 00:06:31,720 --> 00:06:37,560 Can you craft me an answer for my service portal and then send 106 00:06:37,560 --> 00:06:39,040 that to the user requesting help? 107 00:06:39,360 --> 00:06:42,400 And now it just bundles everything together, sends the 108 00:06:42,680 --> 00:06:45,440 the, the update to the user, call it done. 109 00:06:45,720 --> 00:06:49,160 And I think that's really where having your productivity mashed 110 00:06:49,160 --> 00:06:53,000 with your line of business and really like business critical 111 00:06:53,000 --> 00:06:57,720 data changes everything. So being in M365 Compilot really 112 00:06:57,720 --> 00:06:59,760 changes all of that. And sometimes you don't know 113 00:06:59,760 --> 00:07:02,280 where the data is coming from. You don't remember. 114 00:07:02,600 --> 00:07:05,960 I had a chat with Samuel yesterday. 115 00:07:06,400 --> 00:07:09,560 Was it a Teams message? Was it through a comment in a 116 00:07:09,560 --> 00:07:12,880 Word doc? Was it as an e-mail? 117 00:07:12,880 --> 00:07:18,480 Like like what's in an ADO ticket, like in GitHub or in 118 00:07:18,480 --> 00:07:20,000 Azure? Like where was it? 119 00:07:20,640 --> 00:07:22,120 Well, in that case, Kapad doesn't care. 120 00:07:22,440 --> 00:07:25,680 It has access to all that productivity data and then can 121 00:07:25,680 --> 00:07:27,520 bring it to your third party data. 122 00:07:27,520 --> 00:07:30,600 That's really where I think the the big value is. 123 00:07:31,680 --> 00:07:33,080 Yeah, this is really well described. 124 00:07:33,080 --> 00:07:36,880 Actually. Fun fact, just before we we 125 00:07:36,880 --> 00:07:41,280 connected to record this episode, I was I didn't, I had 126 00:07:41,280 --> 00:07:43,320 to prepare for a demonstration for one of my customers. 127 00:07:43,320 --> 00:07:46,400 I just didn't remember what we agreed should be demonstrated. 128 00:07:46,400 --> 00:07:49,600 So I asked copilot, hey, can you tell me what's been agreed? 129 00:07:49,600 --> 00:07:53,120 And look at my e-mail, look at my recording, because we had a 130 00:07:53,240 --> 00:07:56,120 previous meeting that was transcribed and record and it 131 00:07:56,120 --> 00:08:00,200 just lists what we agreed to be demoed on that day. 132 00:08:00,840 --> 00:08:05,920 And then I, I built one of my own agents that helped me use 133 00:08:05,920 --> 00:08:09,920 some templates of demos I did in the past and readapt it based on 134 00:08:10,080 --> 00:08:12,520 the customer context and industry, etcetera. 135 00:08:12,880 --> 00:08:16,200 I've been able to call this agent afterward and getting him 136 00:08:16,200 --> 00:08:20,400 to start building draft of the demo that we'll be presenting. 137 00:08:20,400 --> 00:08:23,240 So yeah, I it's very powerful. I do agree. 138 00:08:23,760 --> 00:08:28,600 Absolutely. Agent creation in the Microsoft 139 00:08:28,600 --> 00:08:33,440 365 Copilot world, I won't say Microsoft because we also have 140 00:08:33,440 --> 00:08:38,200 Azure Foundry, but in M365 Copilot is done through Copilot 141 00:08:38,200 --> 00:08:42,600 Studio and Copilot Studio, which was previously part of Virtual 142 00:08:42,600 --> 00:08:47,280 Agent is often often positioned as a low code platform. 143 00:08:48,120 --> 00:08:53,160 You been working a lot with the Pro Code Agent Toolkit. 144 00:08:54,440 --> 00:08:56,800 So you are on the pro code side of things. 145 00:08:56,960 --> 00:09:01,840 So what is it really and what does it unlock that low code 146 00:09:01,840 --> 00:09:07,840 alone won't be able to like, especially when we're talking 147 00:09:07,840 --> 00:09:13,160 about, you know, scaling control engineering discipline. 148 00:09:13,480 --> 00:09:19,760 I, I know it's, it's a totally different way of reading agent. 149 00:09:19,800 --> 00:09:22,800 So can you explain what's the difference and what's the 150 00:09:22,800 --> 00:09:24,680 advantage of it? Absolutely. 151 00:09:24,680 --> 00:09:30,640 So I think it's more around like meeting the agent builders where 152 00:09:30,640 --> 00:09:32,720 they are. When I'm thinking about somebody 153 00:09:32,720 --> 00:09:36,000 that is an information worker, probably I don't want to bring 154 00:09:36,000 --> 00:09:40,800 them into Visual Studio Code or even using like the copilot CLI 155 00:09:40,800 --> 00:09:43,920 and bring them into a terminal. Even these days, actually, we're 156 00:09:43,920 --> 00:09:46,200 seeing more and more of these information workers being 157 00:09:46,520 --> 00:09:52,320 interested by that new way of working in an agentic way. 158 00:09:52,320 --> 00:09:54,280 I'm going to come back to that a little bit later, but. 159 00:09:54,960 --> 00:09:56,760 I see Lili. Being. 160 00:09:57,360 --> 00:10:01,320 The terminal like literally being inside like a terminal 161 00:10:01,320 --> 00:10:06,440 window that is the like the GitHub Copilot version of that. 162 00:10:06,440 --> 00:10:11,120 So like building and so on in in in our case, we want to meet the 163 00:10:11,120 --> 00:10:14,440 devs where they are and they are in their dev tool. 164 00:10:15,440 --> 00:10:21,720 They are tasked to build agentic scenarios for information 165 00:10:21,720 --> 00:10:26,440 workers so so that the the information workers can use 166 00:10:27,200 --> 00:10:31,360 corporate approved agents that are useful. 167 00:10:31,520 --> 00:10:35,400 Usually these agents are massive agents, like, OK, we want to 168 00:10:35,480 --> 00:10:40,640 build that HR agent that covers all of the scenarios, that 169 00:10:40,640 --> 00:10:46,680 connects to all of our work day data goes to our internal 170 00:10:46,720 --> 00:10:53,440 employee registry, goes to like our really simple, almost like 171 00:10:53,480 --> 00:10:56,920 access database, like vacation system. 172 00:10:57,040 --> 00:11:01,600 Like it wants to do all of that and into a coordinated single 173 00:11:01,600 --> 00:11:05,520 agent. We can ask makers, local people 174 00:11:05,520 --> 00:11:07,520 to actually do that and that is absolutely fine. 175 00:11:08,000 --> 00:11:12,360 But where we want to meet the devs is to meet them because 176 00:11:12,360 --> 00:11:14,720 they're probably building the API, they're probably building 177 00:11:15,080 --> 00:11:17,840 some of these systems. So we want to meet them where 178 00:11:17,840 --> 00:11:21,400 they are in Visual Studio Code, where they can build, publish 179 00:11:21,640 --> 00:11:24,080 and use all the engineering practices that they're used to. 180 00:11:24,600 --> 00:11:27,880 One of them is like, hey, we're six of us working on the same 181 00:11:27,880 --> 00:11:29,640 agent. How do we do that? 182 00:11:30,560 --> 00:11:36,040 Well, we use source control, we use GitHub, we use Azure DevOps 183 00:11:36,040 --> 00:11:39,080 to store the code. We can do versioning of that 184 00:11:39,080 --> 00:11:41,320 code. We can do full requests, we can 185 00:11:41,320 --> 00:11:44,840 review every single line that was changed to make sure that 186 00:11:45,240 --> 00:11:48,680 does it really bring value or that was more like a refactor 187 00:11:48,680 --> 00:11:51,720 that didn't bring any value to the actual agent. 188 00:11:52,360 --> 00:11:56,080 We also have the ability to work 6 at a time on our own version 189 00:11:56,080 --> 00:12:00,240 of the agent that are bringing more capabilities. 190 00:12:00,240 --> 00:12:03,040 I'm focused on this area, you're focused on this other area. 191 00:12:03,720 --> 00:12:07,160 So that's really kind of building that team of agent 192 00:12:07,160 --> 00:12:10,520 builders to come together and the output of that or the 193 00:12:10,520 --> 00:12:13,880 outcome of that is that agent that then you can put 194 00:12:13,920 --> 00:12:18,240 automatically through your own continuous integration process. 195 00:12:18,240 --> 00:12:22,600 So you push to your your source code, it gets built 196 00:12:22,840 --> 00:12:25,680 automatically, gets approved by an admin, then the admin 197 00:12:25,720 --> 00:12:29,120 automatically puts that into the store and becomes available to 198 00:12:29,120 --> 00:12:32,840 everybody. So while we do have the low code 199 00:12:32,840 --> 00:12:36,400 and no code ways of building these agents, we believe that 200 00:12:36,400 --> 00:12:43,160 there are key scenarios that do offer high value for enterprises 201 00:12:43,160 --> 00:12:46,520 that want to bring some of that. Something to think about also is 202 00:12:46,960 --> 00:12:51,360 the way we're we're thinking about M 365 Cobalt extensibility 203 00:12:51,800 --> 00:12:57,080 is that there's a lot that you leave behind, or I would say 204 00:12:57,080 --> 00:13:00,360 leave behind and you don't have to think about when you're 205 00:13:00,360 --> 00:13:02,680 building an agent as a dev today. 206 00:13:02,680 --> 00:13:06,200 You would one of your first reaction if, especially if 207 00:13:06,200 --> 00:13:08,880 you're in the Microsoft ecosystem, be, hey, I'm going to 208 00:13:08,880 --> 00:13:12,760 go to Foundry and I'm going to build an agent in Foundry. 209 00:13:13,760 --> 00:13:16,840 And there I'm going to have to build my own orchestrator. 210 00:13:17,520 --> 00:13:21,600 I'm going to have to build and manage my own models. 211 00:13:22,200 --> 00:13:25,760 I'm going to have to do my own function calling to 3rd party 212 00:13:25,760 --> 00:13:28,360 system. I'm going to have to build my 213 00:13:28,360 --> 00:13:33,200 own retrieval augmented generation or rag system myself. 214 00:13:33,560 --> 00:13:37,600 I'm going to and, and, and for some scenarios that totally 215 00:13:37,600 --> 00:13:41,840 makes sense, but that's one of the complexity we remove 216 00:13:42,320 --> 00:13:45,400 entirely. We're a managed solution. 217 00:13:46,560 --> 00:13:50,160 Bring your systems, we're going to connect to them, bring your 218 00:13:50,160 --> 00:13:53,160 instructions and your data sources that you want to drag on 219 00:13:53,160 --> 00:13:55,160 top of, especially in the productivity space. 220 00:13:55,600 --> 00:13:57,760 We already have it. So we're just going to cross 221 00:13:57,760 --> 00:13:59,360 everything together. We're going to do your own 222 00:13:59,360 --> 00:14:01,960 function calling. We're going to do auth for you. 223 00:14:02,160 --> 00:14:06,920 We're going to do Rai, responsible AI on top of this. 224 00:14:06,920 --> 00:14:11,360 So if there's some dodgy answers coming back from your API or 225 00:14:11,360 --> 00:14:14,360 your third party system, we're going to protect the user from 226 00:14:14,360 --> 00:14:17,400 that. All of that without you having 227 00:14:17,400 --> 00:14:21,560 to basically code anything crazy. 228 00:14:21,560 --> 00:14:30,000 You just bring your systems in a very standard way, like we are 229 00:14:30,000 --> 00:14:33,480 using the standards of the industry either through an API 230 00:14:33,480 --> 00:14:37,240 or through an MCP server, bring your systems and then the, the 231 00:14:37,800 --> 00:14:41,760 our MD's five compiled, the engine will take care of that 232 00:14:41,760 --> 00:14:43,960 automatically. So from a it's like a time to 233 00:14:43,960 --> 00:14:48,480 market that is absolutely crazy. And we can go from no agent to 234 00:14:48,480 --> 00:14:53,520 an agent within a couple hours. Hosted, deployed in production 235 00:14:55,640 --> 00:14:59,440 without any costs involved. Because it's part of the license 236 00:14:59,440 --> 00:15:03,040 that that you have either from copilot chat or from M365 237 00:15:03,040 --> 00:15:04,920 Compilot. If you're not using any 238 00:15:04,920 --> 00:15:08,080 grounding data, you can use it on copilot chat like default. 239 00:15:08,080 --> 00:15:11,040 You don't even have to have the M365 copilot license. 240 00:15:11,320 --> 00:15:14,200 You can just use copilot chat and automatically connect to 241 00:15:14,200 --> 00:15:17,840 your third party systems absolutely at no extra charge 242 00:15:17,840 --> 00:15:20,760 than your M365 regular base license. 243 00:15:22,280 --> 00:15:25,920 Is it the same architecture as the Copilot Studio build code 244 00:15:25,920 --> 00:15:28,000 tool? No, it's a different 245 00:15:28,000 --> 00:15:32,120 architecture. It's a lot more integrated 246 00:15:32,120 --> 00:15:36,080 within the M365 Copilot experience. 247 00:15:36,800 --> 00:15:39,800 When we're thinking about the agent builder capability, it's 248 00:15:39,800 --> 00:15:42,960 the same architecture as the agent builder capability, just 249 00:15:42,960 --> 00:15:45,840 that the agent builder is really, really focused on the 250 00:15:45,840 --> 00:15:48,520 information worker. So doesn't allow you to bring 251 00:15:48,520 --> 00:15:52,480 third party system, doesn't allow you to have all of the 252 00:15:52,480 --> 00:15:55,600 settings of the devs can have. So they have more levers, more 253 00:15:55,600 --> 00:15:58,360 settings, more knobs that you can turn on. 254 00:15:58,920 --> 00:16:01,560 But in the end they are the same architecture. 255 00:16:01,560 --> 00:16:05,200 When we go to Copilot Studio and build like a custom agent that 256 00:16:05,200 --> 00:16:09,520 connects to Dataverse, that connects to whatever other 257 00:16:09,520 --> 00:16:12,960 systems, that's another architecture that is more that 258 00:16:12,960 --> 00:16:16,440 is closer to the custom engine agents, which is you own 259 00:16:16,440 --> 00:16:18,800 everything, but we connect with you. 260 00:16:18,800 --> 00:16:21,320 So we make sure that the prompt gets sent to you. 261 00:16:21,480 --> 00:16:25,200 We do some checks and validation on top of that, but you're 262 00:16:25,240 --> 00:16:27,840 you're basically on your own in that case. 263 00:16:28,240 --> 00:16:31,720 Copilot Studio handles all of the heavy lifting for you, so 264 00:16:31,720 --> 00:16:35,040 you don't have to build the orchestrator, It'll have to own 265 00:16:35,160 --> 00:16:37,720 the models and so on, but it's two different architecture. 266 00:16:38,400 --> 00:16:40,240 OK, and that's really good to know. 267 00:16:40,240 --> 00:16:43,320 So I will assume that if I start creating an agent in the local 268 00:16:43,320 --> 00:16:47,200 tool or in the Procode tool, I can't I can switch to the other 269 00:16:47,200 --> 00:16:49,000 experience, right? Because it's not exactly the 270 00:16:49,000 --> 00:16:51,280 same architecture. So today you cannot. 271 00:16:51,280 --> 00:16:53,680 We have plans. We're thinking about that a lot 272 00:16:53,680 --> 00:16:58,040 on how we could expand that. So start an agent builder, 273 00:16:58,040 --> 00:17:02,320 export that to your pro dev tooling just to maybe expand on 274 00:17:02,320 --> 00:17:04,800 top of what an information worker has done. 275 00:17:05,040 --> 00:17:07,839 Then they hit limits like, OK, my, my tool cannot do more than 276 00:17:07,839 --> 00:17:10,760 that. Fine, let's give it to the deaf 277 00:17:10,760 --> 00:17:13,480 team and the deaf team will take it further or the other around 278 00:17:13,800 --> 00:17:16,920 my you know what? My pro code thing in the end is 279 00:17:16,960 --> 00:17:20,000 only a couple of grounding sources, a very complex 280 00:17:20,000 --> 00:17:22,920 instruction field, but instruction is, is instructions. 281 00:17:23,319 --> 00:17:26,760 So maybe go back to agent builder and continue there. 282 00:17:26,760 --> 00:17:30,520 So we're we have plans around that, but nothing confirmed and 283 00:17:30,520 --> 00:17:32,520 in the room now just yet. OK. 284 00:17:32,680 --> 00:17:36,440 Well, it, it might be in the plan you were mentioning Agent 285 00:17:36,440 --> 00:17:39,320 Builder just for the benefit of our audience that might not know 286 00:17:39,320 --> 00:17:42,320 what Agent Builder is specifically with the fact we're 287 00:17:42,320 --> 00:17:46,640 we've been renaming products from time to time. 288 00:17:48,480 --> 00:17:54,520 So agent Builder is when a user, an end user is creating a symbol 289 00:17:54,520 --> 00:17:59,400 agent that is basically grounded on organizational data or on the 290 00:17:59,400 --> 00:18:02,040 web. So it's it's the lightweight 291 00:18:02,040 --> 00:18:05,880 experience that is available inside of M365 Copilot, right. 292 00:18:05,880 --> 00:18:09,160 So what you're saying it's? The embedded version of that, 293 00:18:09,280 --> 00:18:11,640 it's like you don't have to go to Copilot Studio. 294 00:18:11,640 --> 00:18:14,840 It's really you stay within Copilot to build an agent that 295 00:18:14,840 --> 00:18:21,760 runs inside Copilot. So when does locode stop being 296 00:18:21,840 --> 00:18:26,120 enough like what are like clear signal, like organization, 297 00:18:26,240 --> 00:18:29,840 organizational or operational that tell a business decision 298 00:18:29,840 --> 00:18:32,760 making that it's time to move to pro code? 299 00:18:33,040 --> 00:18:36,760 I like the fact you mention source control and you know, 300 00:18:36,760 --> 00:18:40,160 multiple people working an agent because that's one of the 301 00:18:40,160 --> 00:18:42,440 challenge I have. Often when you work in Copilot 302 00:18:42,440 --> 00:18:45,280 Studio, you can't go back if you made a modification to your 303 00:18:45,280 --> 00:18:48,040 prompt, you can if you didn't save it by yourself on OneNote 304 00:18:48,040 --> 00:18:50,960 or notepad or whatever, you're kind of stuck with the with the 305 00:18:50,960 --> 00:18:54,640 new version. I'm not seeing how easily I can 306 00:18:55,440 --> 00:18:57,840 working team on Copa CGO right now. 307 00:18:57,840 --> 00:19:01,200 So I think those two points are really, really interesting, but 308 00:19:01,200 --> 00:19:04,320 what else? What What are the clear signal a 309 00:19:04,320 --> 00:19:09,040 business leader should look at then deciding which way going to 310 00:19:09,040 --> 00:19:11,320 go I. Think there's also a matter of 311 00:19:11,320 --> 00:19:14,360 ownership. Like sometimes when you're going 312 00:19:14,360 --> 00:19:17,480 to build an agent, you're going to want the user to own it 313 00:19:18,520 --> 00:19:22,480 because he's the subject matter expert on that thing. 314 00:19:22,560 --> 00:19:25,920 Then fine. That is where I think Loco Noco 315 00:19:26,160 --> 00:19:28,360 really makes sense. The user is the owner. 316 00:19:29,200 --> 00:19:32,960 Now when you bring that to the world where you're building this 317 00:19:33,040 --> 00:19:37,160 entire HR complex agent, who's the owner of that? 318 00:19:37,160 --> 00:19:41,600 It's not John, it's the HR department that is owner of 319 00:19:41,600 --> 00:19:43,760 that. The same way when you build your 320 00:19:43,760 --> 00:19:47,840 finance system, it's not one of the devs that is the owner. 321 00:19:48,280 --> 00:19:52,680 It's the finance department that is owner of that or IT depending 322 00:19:52,680 --> 00:19:55,160 on how the organization is structured. 323 00:19:55,880 --> 00:19:58,240 So I think that is also something like if you are 324 00:19:58,240 --> 00:20:03,640 looking to have shared accountability on an agent, you 325 00:20:03,640 --> 00:20:08,760 should probably build it in a way that it's possible to have 326 00:20:08,760 --> 00:20:11,560 shared accountability, shared ownership. 327 00:20:11,560 --> 00:20:14,920 So that's differently also one other thing that we're looking 328 00:20:14,920 --> 00:20:18,640 at. One extra, and I was thinking 329 00:20:18,640 --> 00:20:24,920 kind of one of the last one is you want to build for your 330 00:20:24,920 --> 00:20:27,600 organization. That's where I think Loco Noco 331 00:20:27,840 --> 00:20:31,040 absolutely makes sense if you're building for more than just your 332 00:20:31,040 --> 00:20:35,040 organization. A good example, we haven't built 333 00:20:35,320 --> 00:20:41,960 researcher in compiler studio because it it needs to be 334 00:20:41,960 --> 00:20:46,480 acquirable by all the organizations, our M365 copilot 335 00:20:46,480 --> 00:20:49,080 license holders. So when you're building that and 336 00:20:49,080 --> 00:20:51,680 there's a store for agents today that is available. 337 00:20:51,680 --> 00:20:58,840 So as a service organization that wants to build an agent 338 00:20:58,880 --> 00:21:04,720 that connects to their own SAS, think ServiceNow, ServiceNow has 339 00:21:04,720 --> 00:21:09,920 their own agent that lives inside M365 Copilot that has 340 00:21:09,920 --> 00:21:15,280 been built using our Pro Pro dev tooling now connects to all the 341 00:21:15,280 --> 00:21:16,800 customers of service now in the world. 342 00:21:17,880 --> 00:21:21,040 So that's another way, like if you want to distribute your app 343 00:21:22,000 --> 00:21:29,240 at large across organizations, you will want to have a solution 344 00:21:29,240 --> 00:21:34,400 that is built the way that the underlying system that it relies 345 00:21:34,400 --> 00:21:37,040 on service. Now in that case is built. 346 00:21:37,040 --> 00:21:40,440 ServiceNow is built by a team of devs and product manager and 347 00:21:40,440 --> 00:21:45,400 designers and scientists. You want to be able to package 348 00:21:45,400 --> 00:21:49,400 the agent the same way that the rest is also being made 349 00:21:49,400 --> 00:21:51,040 available. So that is definitely something 350 00:21:51,040 --> 00:21:53,560 that we're thinking about. Like the distribution channels 351 00:21:54,200 --> 00:21:58,360 allow for different approaches. When you're an, you're an agent 352 00:21:58,360 --> 00:21:59,640 builder. When you're an information 353 00:21:59,640 --> 00:22:04,200 worker, you want to distribute to your team or just around you, 354 00:22:04,520 --> 00:22:07,400 maybe to your department. But then if you want to make it 355 00:22:07,400 --> 00:22:10,520 a little bit broader, maybe you go to Compilot studio where you 356 00:22:10,520 --> 00:22:13,920 can have maybe like a couple departments together. 357 00:22:13,920 --> 00:22:18,080 But if you want to full org wide or you want to do a cross 358 00:22:18,080 --> 00:22:21,080 organization, I think that's another good way to do. 359 00:22:22,320 --> 00:22:24,960 That yeah, like you mentioned cross organization because I, I 360 00:22:25,000 --> 00:22:29,160 think, I mean, we're just at the beginning of this agentic AI 361 00:22:29,160 --> 00:22:33,760 wave and I've seen a lot of organization investing heavily 362 00:22:33,760 --> 00:22:36,480 on building very cool agent and very powerful agent. 363 00:22:36,480 --> 00:22:41,520 And it might even become attractive for let's say I'm a 364 00:22:41,520 --> 00:22:45,440 retailer building an agent for my own organization to then say, 365 00:22:45,440 --> 00:22:49,400 hey, maybe I want to resell this on the marketplace as well. 366 00:22:50,280 --> 00:22:53,040 But it's not not only like I'm building it for my own 367 00:22:53,040 --> 00:22:56,920 organization, right, that I can sell it afterward based on the 368 00:22:56,920 --> 00:22:58,600 work we did with our organization. 369 00:22:59,760 --> 00:23:03,920 Yeah, exactly. And I really think that when you 370 00:23:03,920 --> 00:23:07,040 decide what type of agent you want, it's one of the criteria 371 00:23:07,040 --> 00:23:08,800 you need to. OK, so that's going to drive 372 00:23:09,040 --> 00:23:12,920 which tool I want to to have. And also it it can also be like, 373 00:23:12,920 --> 00:23:18,120 OK, do I want to live on the Copilot orchestrator or do I 374 00:23:18,120 --> 00:23:20,880 want to have more levers and more knobs? 375 00:23:20,880 --> 00:23:24,080 Then I go one step further, just, OK, I'm going to host it 376 00:23:24,080 --> 00:23:27,800 in Azure and I'm just going to connect it to Compilot rather 377 00:23:27,800 --> 00:23:30,240 than make it native to Compilot. So depending on the scenarios, 378 00:23:30,240 --> 00:23:34,160 depending on the proximity of the Microsoft 365 data you need, 379 00:23:34,760 --> 00:23:36,640 you're going to have to change your approach a little bit. 380 00:23:38,080 --> 00:23:41,320 People often say you need to ship agents like software, So 381 00:23:41,320 --> 00:23:44,920 what does that actually look like in practice once once 382 00:23:44,920 --> 00:23:48,000 you've decided to go pro code, what's the workflow like? 383 00:23:48,000 --> 00:23:52,800 No versioning environment, Ticd rollback like etcetera etcetera 384 00:23:52,800 --> 00:23:58,720 in the real world. So there's the obviously that 385 00:23:58,720 --> 00:24:04,640 there are two, two approaches that are transitioning from one 386 00:24:04,640 --> 00:24:10,960 to the next right now. So the way today or yesterday, 387 00:24:10,960 --> 00:24:13,600 that's what yesterday, yesterday you were building agents. 388 00:24:13,920 --> 00:24:18,120 We have AVS code extension that is in in AVS extension that is 389 00:24:18,120 --> 00:24:23,720 called the Microsoft 365 Agents toolkit where you create your 390 00:24:23,720 --> 00:24:28,920 agent from there it creates a bunch of Jason files and these 391 00:24:28,920 --> 00:24:33,160 Jason files are the levers and knobs that you have on top of 392 00:24:33,160 --> 00:24:36,760 copilot. So you create that you configure 393 00:24:36,760 --> 00:24:40,520 that you can use GitHub copilot to help you expand change 394 00:24:40,520 --> 00:24:42,880 instructions. Really useful GitHub Copilot to 395 00:24:42,880 --> 00:24:48,280 create instructions is a master at that because it knows your 396 00:24:48,280 --> 00:24:50,240 context. It knows what are the systems 397 00:24:50,240 --> 00:24:52,320 you're connecting with. Then it comes up with the 398 00:24:52,320 --> 00:24:55,800 instructions that will be the most effective for the agent 399 00:24:55,800 --> 00:24:58,080 you're building, which is actually really cool when you're 400 00:24:58,240 --> 00:25:02,600 building in an agent take way and then you push that to your 401 00:25:02,600 --> 00:25:07,240 to your team. So we have ways to provision the 402 00:25:07,280 --> 00:25:11,400 agent directly to Microsoft 365, compile it and share with a 403 00:25:11,400 --> 00:25:13,720 couple of your teammates to test. 404 00:25:14,440 --> 00:25:19,120 Then afterwards you can broaden the scope to your department. 405 00:25:19,920 --> 00:25:21,800 Then you're like, OK, so I'm, I'm happy with this. 406 00:25:21,800 --> 00:25:23,800 Now let me put that into source control. 407 00:25:24,480 --> 00:25:26,760 So you add that to your source control and then from there you 408 00:25:26,760 --> 00:25:31,600 can do a full CICD pipeline. We have CLI, so common, common 409 00:25:31,600 --> 00:25:37,240 line interfaces, little software also built with the Agents 410 00:25:37,240 --> 00:25:41,360 toolkit that allows you to script the deployment of these 411 00:25:41,360 --> 00:25:45,160 things from OK, I want to build to my dev environment. 412 00:25:45,160 --> 00:25:46,920 So I'm going to use the dev environment. 413 00:25:47,200 --> 00:25:49,840 I'm going to build my agent, deploy to the dev environment, 414 00:25:50,320 --> 00:25:53,240 have my QA team to test it there. 415 00:25:53,520 --> 00:25:55,520 And I want to have my integration environment where I 416 00:25:55,520 --> 00:25:59,000 can now really have all of these third party systems and make 417 00:25:59,000 --> 00:26:01,400 sure that you had it there. Then we have it. 418 00:26:01,520 --> 00:26:04,920 And then after that we're going to have the the prod 419 00:26:04,920 --> 00:26:07,320 environment. So all of that through whatever 420 00:26:07,320 --> 00:26:10,040 tool you're using for that continuous integration, you're 421 00:26:10,080 --> 00:26:13,120 going to be able to build these pipelines that allow you to go 422 00:26:13,120 --> 00:26:16,560 from one to the other, can put on some approvals and and so on. 423 00:26:16,560 --> 00:26:19,960 That is not specific to the agents toolkit or to building 424 00:26:19,960 --> 00:26:22,480 agents. That's how software is being 425 00:26:22,480 --> 00:26:27,640 built now. One thing that we're seeing more 426 00:26:27,640 --> 00:26:31,320 and more is agents building software. 427 00:26:33,960 --> 00:26:39,200 So when you think about GitHub Compilot, GitHub Compilot is an 428 00:26:39,200 --> 00:26:43,560 engine in itself that can reason over your code, that can reason 429 00:26:43,560 --> 00:26:49,360 over your spec, that can reason over anything and generate the 430 00:26:49,400 --> 00:26:55,000 actual code that needs to be built for this. 431 00:26:55,000 --> 00:26:59,280 So we're now coming with a set of, of different agentic 432 00:26:59,280 --> 00:27:02,320 experiences to help devs be really powerful. 433 00:27:02,320 --> 00:27:06,720 So one of them is agent skills that basically teach the agent a 434 00:27:06,720 --> 00:27:09,880 new skill, how to build an agent, how to build a 435 00:27:09,880 --> 00:27:13,000 declarative agent. So now it knows how to start, 436 00:27:13,160 --> 00:27:17,120 how to create, how to add the web search capability, how to 437 00:27:17,120 --> 00:27:18,440 add this. And then it knows. 438 00:27:18,600 --> 00:27:23,640 So now it it, it, it becomes an expert at building declarative 439 00:27:23,640 --> 00:27:27,120 agents that live on on top of Microsoft 365 compilot. 440 00:27:27,120 --> 00:27:31,480 So we have that available now something we launched last 441 00:27:31,480 --> 00:27:33,960 Friday's timing of the of the recording is really good. 442 00:27:34,000 --> 00:27:38,920 We launched last Friday the work IQCLI the work IQ and CP server. 443 00:27:39,320 --> 00:27:41,240 What is work IQ? I'm pretty sure you, you've, 444 00:27:41,400 --> 00:27:44,560 you've heard about work IQ, but work IQ is basically the 445 00:27:44,560 --> 00:27:51,400 connective tissue that connects your like organizational data to 446 00:27:51,440 --> 00:27:54,160 your users. But if you're using that in a, 447 00:27:54,240 --> 00:27:58,280 in a, in an agentic way, now that's what we see more and more 448 00:27:58,600 --> 00:28:04,960 is people are connecting the work IQ superpower like this 449 00:28:05,280 --> 00:28:08,480 inside their coding agent. So now you can ask questions 450 00:28:08,480 --> 00:28:12,480 like last Friday we had a meeting with the engineering 451 00:28:12,480 --> 00:28:16,440 team. We made we made decisions based 452 00:28:16,440 --> 00:28:21,480 on these decisions. Can you update my agent based on 453 00:28:21,480 --> 00:28:22,840 the decisions that were made in the meeting? 454 00:28:23,080 --> 00:28:27,360 So now it's going to go look at the transcript automatically get 455 00:28:27,360 --> 00:28:29,640 all of that back and say, oh, wait a second. 456 00:28:29,640 --> 00:28:33,680 We said that we wanted to ground on a new site in SharePoint and 457 00:28:33,680 --> 00:28:36,320 it's going to add the site directly inside your SharePoint 458 00:28:36,760 --> 00:28:38,840 inside your, your configuration that says, OK, now I want to 459 00:28:38,840 --> 00:28:41,600 ground on this on top of it, or we want to integrate with the 460 00:28:41,600 --> 00:28:44,080 new third party system. OK, I'm going to go in going to 461 00:28:44,080 --> 00:28:48,640 craft the API integration with our vacation system. 462 00:28:49,200 --> 00:28:51,320 It goes on, goes to your source code. 463 00:28:51,320 --> 00:28:55,560 Wait a second, I just found the reference to the APII can craft 464 00:28:55,560 --> 00:28:58,400 what I need to do now. Boom, I stitch it all together. 465 00:28:58,760 --> 00:29:04,840 So now we're bringing the word data, organizational data inside 466 00:29:04,840 --> 00:29:09,960 the life of developers because that's that was always the 467 00:29:09,960 --> 00:29:12,920 missing piece. Like the spec is usually a word 468 00:29:12,920 --> 00:29:17,760 doc lost somewhere in a chat, but now we can actually go and 469 00:29:17,760 --> 00:29:22,160 find it, bring it reason over it, create code that respects 470 00:29:22,160 --> 00:29:26,240 the decisions and the design, got it done and then ship it to 471 00:29:26,240 --> 00:29:27,720 the software. So those are two things like 472 00:29:28,000 --> 00:29:32,720 more like the experience UI based things. 473 00:29:32,960 --> 00:29:36,840 Then there's all this new agentic way of thinking about 474 00:29:36,840 --> 00:29:42,720 building software. So I'll assume that it has 475 00:29:42,720 --> 00:29:46,680 affected or has changed the way you were coding, right? 476 00:29:46,680 --> 00:29:47,920 Like 100. Percent. 477 00:29:48,320 --> 00:29:50,880 Yeah. How, how, how do you address 478 00:29:51,240 --> 00:29:54,280 developing nowadays like if I, if you compare from three years 479 00:29:54,280 --> 00:29:56,800 ago or four years ago? We can compare to six months 480 00:29:56,800 --> 00:29:59,360 ago. It's it's that fast right now. 481 00:29:59,680 --> 00:30:02,960 I think the thing that is mesmerizing for me, like I'm 482 00:30:02,960 --> 00:30:05,400 APM, but I'm I'm coming from an engineering background. 483 00:30:05,400 --> 00:30:11,200 I used to be like a dev lead and I've I've done my my load of 484 00:30:11,200 --> 00:30:13,280 coding in the past. I still do a lot of coding but 485 00:30:13,280 --> 00:30:15,840 what I'm realizing is that I don't look at code anymore. 486 00:30:18,200 --> 00:30:20,320 I literally look at the outcome and that's it. 487 00:30:21,280 --> 00:30:24,720 Where I'm building the agent and testing the agent. 488 00:30:27,000 --> 00:30:29,560 Does that work? I'm all fine. 489 00:30:30,840 --> 00:30:37,400 So in a sense it's almost as easy as using the local tool or 490 00:30:37,400 --> 00:30:41,680 even even easier because you're chatting with an agent, get up, 491 00:30:41,680 --> 00:30:44,680 get up Coupala, for instance, an agent that is creating the code 492 00:30:44,680 --> 00:30:48,360 for you, leveraging agent Toolkit and all the tools 493 00:30:48,360 --> 00:30:50,800 available from Coupala Studio. But you don't have to worry 494 00:30:50,800 --> 00:30:55,840 about all, you know, the back end infrastructure like you 495 00:30:55,840 --> 00:30:57,680 mentioned it, orchestration, etcetera. 496 00:30:58,240 --> 00:31:00,720 Exactly. So it it really creates that 497 00:31:00,720 --> 00:31:06,440 super smooth conversation that is very similar to, Hey, let's 498 00:31:06,440 --> 00:31:10,400 get into a room and let's just figure out what we're going to 499 00:31:10,400 --> 00:31:13,560 build. Now, I'm not saying and not like 500 00:31:13,640 --> 00:31:16,040 that's where like I'm not saying don't look at the code. 501 00:31:16,680 --> 00:31:22,480 I'm saying that doing really, really rapid progress on an 502 00:31:22,480 --> 00:31:26,440 agent is really, really, really, really interesting because then 503 00:31:26,440 --> 00:31:30,040 you move really, really quick. Then afterwards you go back and 504 00:31:30,040 --> 00:31:32,760 you review, OK, now I think I have something. 505 00:31:32,760 --> 00:31:36,520 Let's review that and let's use AI to review. 506 00:31:36,520 --> 00:31:42,160 Let's use AI to OK, So this bit I want to change like this very 507 00:31:42,160 --> 00:31:45,800 specific 2 lines or three lines of instructions. 508 00:31:45,800 --> 00:31:47,280 Let's just change it in another way. 509 00:31:47,280 --> 00:31:51,160 I want to spin it differently, but in the end, if you're 510 00:31:51,320 --> 00:31:57,800 teaching your coding agent the right way to build and you're 511 00:31:57,800 --> 00:32:02,320 giving a a good steer, you're kind of more like a teammate, 512 00:32:02,320 --> 00:32:05,080 like a peer programming session between you and the coding 513 00:32:05,080 --> 00:32:08,440 agent. We're realizing that the value 514 00:32:08,440 --> 00:32:12,960 we're getting out of this is absolutely stunning. 515 00:32:13,000 --> 00:32:18,400 And with the inclusion of work IQ in in the agentic loop for 516 00:32:18,400 --> 00:32:22,280 the devs now it's not just like, oh, what did my boss say last 517 00:32:22,280 --> 00:32:23,520 week? I cannot remember exactly. 518 00:32:23,760 --> 00:32:26,280 I'm going to figure out. No, now you can just say my boss 519 00:32:26,280 --> 00:32:30,640 said something last week, bring that context back and infuse it 520 00:32:30,760 --> 00:32:32,120 in the agent that I'm trying to build. 521 00:32:33,120 --> 00:32:37,240 Yeah, Work IQ give you access to all documents you have access 522 00:32:37,240 --> 00:32:39,560 to. Previous meetings been 523 00:32:39,560 --> 00:32:43,880 transcript emails, Fairpoint site, if you grounded on it, 524 00:32:44,520 --> 00:32:47,560 etcetera, etcetera. So yeah, actually that was one 525 00:32:47,560 --> 00:32:50,280 of the challenge I had building some of my agents. 526 00:32:50,280 --> 00:32:53,400 So I'm very happy to hear that it's it's releasing now. 527 00:32:53,840 --> 00:32:56,080 Yep, absolutely. It's, it's, it's now in public 528 00:32:56,080 --> 00:32:58,320 preview and it's available for all types of coding agents. 529 00:32:58,320 --> 00:33:01,880 So we we talk a lot about GitHub compiler, but put your your 530 00:33:01,880 --> 00:33:05,680 favorite coding agent in there and you can actually talk to it. 531 00:33:07,560 --> 00:33:10,360 Switching gear a bit you're you're working a lot with 532 00:33:10,400 --> 00:33:13,520 customer and partner. So now, what's a common belief 533 00:33:13,880 --> 00:33:18,040 or piece of advice about building copilot agents that you 534 00:33:18,440 --> 00:33:21,280 think is just wrong I. Hold. 535 00:33:21,680 --> 00:33:23,960 Yeah. That is a good question. 536 00:33:23,960 --> 00:33:35,560 So I think the first kind of wrong assumption is that you 537 00:33:35,560 --> 00:33:39,120 will build a better orchestration layer than 538 00:33:39,120 --> 00:33:43,840 Microsoft 365 Compilot Kit. So it's a little bit the way I 539 00:33:43,840 --> 00:33:46,840 see it might be a little bit like we're better and I don't 540 00:33:46,840 --> 00:33:51,920 think that's how I mean it is. There's such a huge effort to 541 00:33:51,920 --> 00:33:55,200 make the orchestration in Microsoft 365 Compilot as good 542 00:33:55,200 --> 00:33:58,920 as possible for both us. You need to realize that all the 543 00:33:58,920 --> 00:34:02,840 agents we ship are agents that are built with our own tooling. 544 00:34:03,720 --> 00:34:06,760 We're building them all with declarative agents. 545 00:34:07,880 --> 00:34:14,080 So if you think our agents are performing and doing some really 546 00:34:14,239 --> 00:34:20,719 crazy things, thing, researcher, analysts, all the agent modes in 547 00:34:21,000 --> 00:34:26,120 Word, Excel, PowerPoint, well, your agent will live on the 548 00:34:26,120 --> 00:34:29,520 exact same platform. So I think that's the biggest 549 00:34:30,000 --> 00:34:34,880 things like building that is very critical. 550 00:34:34,880 --> 00:34:40,679 Now I want to be very respectful of especially people that are 551 00:34:40,880 --> 00:34:44,199 doing more than M365 copilot. Like let's say you're building 552 00:34:44,199 --> 00:34:47,920 an agent that needs to be available with an M365, but also 553 00:34:48,600 --> 00:34:53,320 outside of M365 as a web on the website on I don't know by 554 00:34:53,320 --> 00:34:59,760 e-mail, via SMS or whatsoever. Then they are downsides to using 555 00:34:59,920 --> 00:35:04,600 our stack which is our stack will not allow you to embed your 556 00:35:04,600 --> 00:35:11,280 agent elsewhere than M365 and all the collaterals like Excel, 557 00:35:11,280 --> 00:35:13,920 PowerPoint, Outlook, Word, teams. 558 00:35:13,920 --> 00:35:18,520 If you are building for your employees or for employees 559 00:35:18,960 --> 00:35:23,960 that's your main target not the public or like I really think 560 00:35:23,960 --> 00:35:26,680 that M365 copilot should be your your go to. 561 00:35:28,160 --> 00:35:34,480 If you go outside of that, then there I'm, I'm even if it's not 562 00:35:34,880 --> 00:35:39,920 as powerful over an orchestrator than M365 is, you need to, to go 563 00:35:39,920 --> 00:35:41,120 there. And that's where I think 564 00:35:41,120 --> 00:35:44,600 compiled studio plays a, a phenomenal role because then 565 00:35:44,680 --> 00:35:47,160 they are available outside of that realm. 566 00:35:47,160 --> 00:35:50,120 And then you can have it also embedded inside M365 compiled. 567 00:35:50,120 --> 00:35:55,400 But if you're really building for employees, no question 568 00:35:55,400 --> 00:35:59,080 asked, go all in with declarative agents and pro pro 569 00:35:59,080 --> 00:36:01,880 code tooling. So that's a really interesting 570 00:36:01,880 --> 00:36:06,320 point coop out studio, the low code version will let you deploy 571 00:36:06,320 --> 00:36:11,120 on customer service if you want on your website, inside of power 572 00:36:11,120 --> 00:36:13,280 apps, etcetera. From my understanding, and it 573 00:36:13,280 --> 00:36:18,200 makes sense because you mention it, the pro code version is 574 00:36:18,200 --> 00:36:20,960 running on the same architecture as M365 copilot. 575 00:36:20,960 --> 00:36:25,560 So the pro code will not you deploy elsewhere, but the low 576 00:36:25,560 --> 00:36:30,640 code version will let you do it. But with some limitations that 577 00:36:31,360 --> 00:36:33,360 the pro code doesn't have. OK, but. 578 00:36:33,360 --> 00:36:36,680 You still have you still have the ability to do pro code and 579 00:36:36,960 --> 00:36:40,680 go elsewhere. That is but like, and as part of 580 00:36:40,680 --> 00:36:42,960 the agents toolkit, you have declarative agents you can 581 00:36:42,960 --> 00:36:44,800 create. You also have custom engine 582 00:36:44,800 --> 00:36:46,560 agents. Then that's really where you, 583 00:36:47,160 --> 00:36:50,280 you, I don't know if you remember maybe 10 years ago 584 00:36:50,640 --> 00:36:55,880 where we had that pizza as a service infographics where if 585 00:36:55,880 --> 00:37:00,120 you're ordering pizza, well, you don't own anything. 586 00:37:00,120 --> 00:37:01,640 The only thing you own is the pizza. 587 00:37:02,600 --> 00:37:06,280 But if you are baking the pizza at home and you need to do your 588 00:37:06,280 --> 00:37:10,320 own crust and you need to do everything, you own all of the 589 00:37:10,320 --> 00:37:12,680 layers of the stack. But we're, we're backing a 590 00:37:12,680 --> 00:37:14,760 little bit into this world. If you're in the declarative 591 00:37:14,760 --> 00:37:21,360 agent, you basically own a small slice of it, but the quality is 592 00:37:21,360 --> 00:37:25,400 still very great. You don't have to own all the 593 00:37:25,400 --> 00:37:27,840 pieces, but if you want to go complexity and because you need 594 00:37:27,840 --> 00:37:32,680 it, then you have the option to own more slices of that bar 595 00:37:32,680 --> 00:37:35,320 chart. And I think it's, it's important 596 00:37:35,320 --> 00:37:38,000 to realize that there's a balance. 597 00:37:38,280 --> 00:37:44,440 You will want to be somewhere in that continuum from I'm building 598 00:37:44,440 --> 00:37:48,280 an agent builder as an information worker to I build an 599 00:37:48,280 --> 00:37:51,240 Azure Foundry and I own, I own everything. 600 00:37:51,240 --> 00:37:54,800 And then I need to own deployments, runtime and so on. 601 00:37:55,960 --> 00:37:58,040 I love the analogy, it makes total sense. 602 00:37:59,560 --> 00:38:04,040 If you could give leader one a piece of advice to avoid a 603 00:38:04,040 --> 00:38:07,600 painful RE architecture lead, what would it be? 604 00:38:09,320 --> 00:38:13,160 Start with the audience. Start like what are you trying 605 00:38:13,160 --> 00:38:15,880 to do? What is the the target audience 606 00:38:15,880 --> 00:38:20,600 of this agent? Is it 5 people, 100 people, 607 00:38:21,120 --> 00:38:25,480 10,000 people to 100,000 people? Based on that, I think you're 608 00:38:25,480 --> 00:38:33,160 going to want to make different choices to deliver your, your 609 00:38:33,160 --> 00:38:36,120 agent. You will want to like, you know 610 00:38:36,120 --> 00:38:41,120 how especially in this is in this wave of AI where we're 611 00:38:41,120 --> 00:38:43,080 talking about like disposable software. 612 00:38:43,480 --> 00:38:46,800 It's like I'm building an agent. It's good for that event. 613 00:38:47,520 --> 00:38:49,800 Then afterwards that event is gone. 614 00:38:50,600 --> 00:38:52,880 It was super useful while it happened and afterwards it's 615 00:38:53,040 --> 00:38:55,560 like it's just straight to the can. 616 00:38:56,280 --> 00:39:00,600 Well, that you need to think about like how durable you think 617 00:39:00,600 --> 00:39:03,480 the agent will be and for how many people will this agent be. 618 00:39:03,480 --> 00:39:07,280 So I think that's what will trigger some of the decisions in 619 00:39:07,280 --> 00:39:11,320 there. I want to make my HR, like my HR 620 00:39:11,320 --> 00:39:14,760 support department will start with an agent and then at some 621 00:39:14,760 --> 00:39:16,760 point you're going to hit the limit. 622 00:39:16,760 --> 00:39:20,160 You're going to hit like, we want somebody to talk to you. 623 00:39:20,800 --> 00:39:24,240 Then the agent will just respond with, OK, so here's the ticket 624 00:39:24,280 --> 00:39:27,560 that you need to go away. I created a ticket for you in 625 00:39:27,560 --> 00:39:30,800 our HR system. But until that point, you might 626 00:39:30,800 --> 00:39:35,280 have this interact with with an AI agent up to a certain point, 627 00:39:35,280 --> 00:39:38,240 But this is a big piece of software still. 628 00:39:38,680 --> 00:39:40,920 And you want to be able to kind of handle that through 629 00:39:40,920 --> 00:39:42,760 versioning. You want to be able to go back. 630 00:39:42,760 --> 00:39:45,000 You want to be able to roll back your previous version because 631 00:39:45,000 --> 00:39:48,080 then you realize, Oh no, we broke everybody. 632 00:39:49,000 --> 00:39:50,920 You want to be able to go to go back to these things. 633 00:39:50,920 --> 00:39:54,800 So that's I'm rethinking about these scenarios first. 634 00:39:55,600 --> 00:40:00,000 Also, I think there is in in the world of AI we need to think 635 00:40:00,000 --> 00:40:05,280 about, OK, is it workflow based? Is it so prescribed that he 636 00:40:05,280 --> 00:40:11,080 needs to follow this very specific set of steps or this is 637 00:40:11,080 --> 00:40:16,000 very open-ended? Let AI figure out what needs to 638 00:40:16,000 --> 00:40:20,240 be done. And that is where I think we see 639 00:40:20,240 --> 00:40:25,360 more and more declarative agents are 100% AI driven. 640 00:40:26,800 --> 00:40:31,520 You can have some instructions that will guide the AI agent. 641 00:40:33,040 --> 00:40:37,400 Instructions are like instructions to children, right? 642 00:40:38,040 --> 00:40:42,560 They follow them sometimes they follow them a majority of the 643 00:40:42,560 --> 00:40:43,840 time. Sometimes they're going to go 644 00:40:43,840 --> 00:40:47,600 off script. And that is expected at this 645 00:40:47,600 --> 00:40:52,080 point in the world of AI. So if you want to really go 646 00:40:52,080 --> 00:40:56,080 through a workflow, then I maybe that's where you want to have a 647 00:40:56,080 --> 00:41:02,360 lot more control, either using Logo with boxes in Copilot 648 00:41:02,360 --> 00:41:05,360 Studio or you want to build your own orchestration. 649 00:41:05,680 --> 00:41:09,480 So that way you can own every single step of the user journey. 650 00:41:10,120 --> 00:41:12,880 Well, that's the two things that I would love to kind of have 651 00:41:12,880 --> 00:41:20,680 people think before they go all out into a in an M365 copilot 652 00:41:20,960 --> 00:41:24,760 agent project. Yeah, thank you. 653 00:41:24,760 --> 00:41:26,360 And I like your analogy about it. 654 00:41:27,120 --> 00:41:29,320 Bring it to a child. I think I will reuse this one. 655 00:41:31,200 --> 00:41:33,920 Make total sense. Like listens, but sometimes it 656 00:41:33,920 --> 00:41:36,800 doesn't. Just want to do its own thing. 657 00:41:37,200 --> 00:41:39,560 Yeah, summarize my experience with my kids. 658 00:41:39,560 --> 00:41:41,480 I can tell you. And that's why you have the 659 00:41:41,480 --> 00:41:44,680 thumbs down. It's by the way, use them. 660 00:41:44,760 --> 00:41:48,880 We like, we react to them. We like, we know when things go 661 00:41:48,880 --> 00:41:51,360 bad and we, we improve based on that. 662 00:41:51,360 --> 00:41:55,280 So use them, tell us what happened and we're, we're going 663 00:41:55,280 --> 00:42:00,480 to react to improve whatever component was broken out there. 664 00:42:00,480 --> 00:42:05,560 So that's a signal back to Dai cutting VI. 665 00:42:06,400 --> 00:42:08,840 That was bad. Yeah, I'm telling all my 666 00:42:08,840 --> 00:42:12,160 customer like use it, use it. We're like you said, we're very 667 00:42:12,160 --> 00:42:13,520 reactive. I think in that I had a 668 00:42:13,520 --> 00:42:16,600 conversation with Aaron Chapel, which is our CVP of commercial 669 00:42:16,600 --> 00:42:20,440 area and was explaining how we have this, you know, feedback 670 00:42:20,440 --> 00:42:23,400 clue very tight between a customer and the engineering 671 00:42:23,400 --> 00:42:25,520 group. And in all my years doing 672 00:42:25,520 --> 00:42:29,760 Microsoft product, I've never seen a time where we're 673 00:42:29,760 --> 00:42:35,120 reacting, you know, that fast to customer feedback and that comes 674 00:42:35,360 --> 00:42:37,960 that goes from those thumbs up, thumbs down inside of Coppad 675 00:42:37,960 --> 00:42:39,600 Studio when we're talking about agents. 676 00:42:40,120 --> 00:42:43,960 And please do thumbs up like when when things go right, it's 677 00:42:43,960 --> 00:42:46,080 always like, you know how it is, like you read these feedback, 678 00:42:46,080 --> 00:42:48,920 it's always very negative. And that's OK because that's 679 00:42:49,320 --> 00:42:53,080 when like, like it's good. No news is good news. 680 00:42:54,040 --> 00:42:56,400 But sometimes like, oh, this was awesome. 681 00:42:56,560 --> 00:42:58,440 It's fun to read some like some thoughts. 682 00:43:01,360 --> 00:43:03,320 Did I seem well almost at the end of our time. 683 00:43:03,320 --> 00:43:06,320 So I have two questions left. Those are my signature 684 00:43:06,320 --> 00:43:09,240 questions. First one is, you know a more 685 00:43:09,240 --> 00:43:14,320 personal notes, what's the single AI abit or your workflow 686 00:43:14,320 --> 00:43:18,480 that's at at the biggest impact on your own productivity. 687 00:43:21,160 --> 00:43:27,680 So I would say that it's coming from a dev like from from that 688 00:43:27,680 --> 00:43:33,880 side of the house. If you have access to a coding 689 00:43:33,880 --> 00:43:37,320 agent, you'd have compiled cloud code whatsoever. 690 00:43:38,160 --> 00:43:45,040 I invite everybody that would potentially fear to go into that 691 00:43:45,040 --> 00:43:50,360 world and start using these tools, especially now with the 692 00:43:50,360 --> 00:43:52,720 availability of Work IQ. It's connected to your 693 00:43:52,720 --> 00:43:59,320 productivity data now and see how it might redefine your way 694 00:43:59,320 --> 00:44:02,120 of using agents. And I think that's going to 695 00:44:02,120 --> 00:44:07,760 boost your love for M365 copilot because oh, wait a second, I've 696 00:44:07,760 --> 00:44:09,920 done that. I've seen that now I can go and 697 00:44:11,040 --> 00:44:14,040 create my own agent in agent builder that will replicate 698 00:44:14,040 --> 00:44:16,920 these scenarios. It's just it's it's really, 699 00:44:16,920 --> 00:44:20,040 really interesting. So like one thing it's it, it 700 00:44:20,040 --> 00:44:24,080 makes me very productive to live in the terminal, But for me, 701 00:44:24,080 --> 00:44:27,160 it's not like something that is stranger to me. 702 00:44:27,240 --> 00:44:31,400 I'm I'm used to that scenario. But for people that are not used 703 00:44:31,400 --> 00:44:33,560 to that, it's definitely feels like, Oh my God, I'm in the 704 00:44:33,560 --> 00:44:37,240 matrix. But get familiar with it. 705 00:44:37,240 --> 00:44:42,320 There's a lot of really cool innovation also happening in in 706 00:44:42,320 --> 00:44:43,960 that space for personal productivity. 707 00:44:45,120 --> 00:44:46,720 You know what, I'm not a deaf at all. 708 00:44:46,760 --> 00:44:49,920 That was one of my biggest rereads in my career and you 709 00:44:49,920 --> 00:44:52,920 just convinced me to start coding or or write coding. 710 00:44:53,040 --> 00:44:54,640 I don't think I'll become a coder but. 711 00:44:55,320 --> 00:44:57,680 I think everybody's vibe coding these days and and yeah, that's 712 00:44:57,680 --> 00:45:03,040 one of the thing like you have an idea, just express it, say 713 00:45:03,040 --> 00:45:06,280 it. And the number of magical things 714 00:45:06,280 --> 00:45:09,320 that will happen is quite phenomenal. 715 00:45:09,320 --> 00:45:12,960 It's a it's just a software for yourself and that's absolutely 716 00:45:12,960 --> 00:45:14,400 fine. And sometimes it's just hey, I 717 00:45:14,400 --> 00:45:17,880 want to one of the things like I have all these pictures in that 718 00:45:17,880 --> 00:45:20,840 folder and I want to rename them using that format. 719 00:45:21,040 --> 00:45:24,760 How do I do that? Open a coding agent, just hit 720 00:45:24,760 --> 00:45:26,360 enter. It's going to do a bunch of 721 00:45:26,360 --> 00:45:28,800 magical scripting for you. You don't need to think about it 722 00:45:29,000 --> 00:45:31,720 and it's going to do it. So it's always a a little bit of 723 00:45:31,720 --> 00:45:35,680 a, a cool scenario can be a you can be afraid a little bit at 724 00:45:35,680 --> 00:45:39,000 the at the beginning can be a little scare, but honestly, 725 00:45:39,240 --> 00:45:43,320 magical. I will totally try it and that 726 00:45:43,320 --> 00:45:46,720 bring me to my last question. Actually looking 10 years out 727 00:45:47,000 --> 00:45:52,920 now, we've seen how powerful those models have become in only 728 00:45:52,920 --> 00:45:55,880 three years. What do you think? 729 00:45:55,880 --> 00:45:59,760 People still under underestimate about how AI will change how we 730 00:45:59,760 --> 00:46:02,600 work or live. But in that case, for 731 00:46:02,600 --> 00:46:04,680 developers, what does it means? I know a lot of people are 732 00:46:04,680 --> 00:46:08,200 scared of losing their job. Some people are just, you know, 733 00:46:08,200 --> 00:46:11,840 worried about the outputs of architecture in terms of 734 00:46:11,840 --> 00:46:15,120 security, etcetera. But like going 10 years in the 735 00:46:15,120 --> 00:46:19,840 future, I will just imagine that all of this will improve. 736 00:46:19,840 --> 00:46:22,800 So what? What do you see the world look 737 00:46:22,800 --> 00:46:27,320 like in 10 years? 10 years is really, really far. 738 00:46:27,760 --> 00:46:32,000 Yeah, I know. But I think as individuals, as 739 00:46:32,000 --> 00:46:39,800 information worker, as desk workers, I think our daily work 740 00:46:39,800 --> 00:46:46,160 will be to drive and steer A-Team of agents that helps you 741 00:46:46,160 --> 00:46:49,000 be more productive. That might become autonomous, 742 00:46:49,160 --> 00:46:52,720 might become your second brain, but you're still going to be, I 743 00:46:52,720 --> 00:46:55,160 want to think that you're still going to, you're still going to 744 00:46:55,160 --> 00:47:00,440 be in control of everything that is happening from the from the 745 00:47:00,480 --> 00:47:03,120 AI space. So I think that's really what 746 00:47:03,120 --> 00:47:11,560 I'm I'm, I'm seeing you're going to be the manager and mentor of 747 00:47:11,560 --> 00:47:16,680 all of these AI agents that will do things on your behalf. 748 00:47:16,680 --> 00:47:19,480 I want to believe that the human touch will stay. 749 00:47:19,480 --> 00:47:22,360 I want to believe that we're still going to meet as humans 750 00:47:22,360 --> 00:47:26,720 and it's not going to be like just seven AI agents meeting and 751 00:47:26,720 --> 00:47:31,720 making decisions. I want to believe that the human 752 00:47:31,720 --> 00:47:38,560 has a strong play in this in this world. 753 00:47:39,440 --> 00:47:42,760 I think that's why I love so much the brand that Microsoft 754 00:47:42,760 --> 00:47:47,400 represents and how we pitch AI with copilot, where you're the 755 00:47:47,400 --> 00:47:49,200 pilot, then there's a copilot with you. 756 00:47:49,440 --> 00:47:54,480 Sometimes the the copilot can take over for some time, but at 757 00:47:54,480 --> 00:47:56,960 some point it's going to report back to you and say, hey, here's 758 00:47:56,960 --> 00:48:01,720 what I did you think. And I want to believe that in 10 759 00:48:01,720 --> 00:48:04,040 years, it's going to be a lot of the work we're going to do will 760 00:48:04,040 --> 00:48:09,120 be to manage these these folks and be more strategic in how we 761 00:48:09,120 --> 00:48:11,960 think and how we behave with the rest of our colleagues. 762 00:48:13,080 --> 00:48:14,840 And that that's how I see things as well. 763 00:48:16,480 --> 00:48:19,840 Thank you so much for your time today, Sebastian. 764 00:48:19,840 --> 00:48:23,360 I think leaders listening to this conversation will have a 765 00:48:23,400 --> 00:48:28,360 clear framework on making the decision of of going low code or 766 00:48:28,360 --> 00:48:32,160 pro code or be both because. Yeah, I do both. 767 00:48:32,320 --> 00:48:36,080 I we want people to use both. It's just that there is an 768 00:48:36,080 --> 00:48:41,240 option. There are you have more Golf 769 00:48:41,240 --> 00:48:44,720 Club in your golf bag. I'm a golfer, so I went with 770 00:48:44,720 --> 00:48:46,960 that analogy. But yeah, you have more than 771 00:48:46,960 --> 00:48:50,520 just like a putter and Steven Iron. 772 00:48:50,800 --> 00:48:59,640 You have more than that and be curious, explore, try it, fail, 773 00:49:00,280 --> 00:49:03,600 change, go in Azure, come back to M365. 774 00:49:03,800 --> 00:49:08,080 But especially with AI, it's so quick to try things these days 775 00:49:08,320 --> 00:49:12,040 that I think you can make your mind very rapidly on what really 776 00:49:12,040 --> 00:49:14,640 suits you. And from there, you're going to 777 00:49:14,640 --> 00:49:18,120 be able to progress and deliver productivity enhancements, 778 00:49:18,440 --> 00:49:20,560 deliver value directly to your own business. 779 00:49:22,440 --> 00:49:26,400 Thank you so much Sebastian for your time. 780 00:49:27,360 --> 00:49:28,920 More than welcome, thanks for having me. 781 00:49:29,640 --> 00:49:30,880 Have a great day. Cheers. 782 00:49:32,120 --> 00:49:35,560 Before we wrap up, a quick recap of the patterns that matter from 783 00:49:35,560 --> 00:49:40,880 my conversation with Sebastian. First, choosing the right agent 784 00:49:40,880 --> 00:49:43,760 architecture starts with audience and durability. 785 00:49:44,320 --> 00:49:48,000 Sebastian underscored that lightweight user owned agents 786 00:49:48,000 --> 00:49:52,680 work for narrow scenarios, but organization wide agents need 787 00:49:52,760 --> 00:49:56,240 shared ownership, source control, and stable 788 00:49:56,240 --> 00:49:59,680 environments. That shift determines whatever 789 00:49:59,680 --> 00:50:03,760 you're staying low code or move into pro code discipline. 790 00:50:04,720 --> 00:50:08,640 Second, declarative agents inside Microsoft 365 Copilot 791 00:50:08,640 --> 00:50:12,120 reduce complexity by removing the need to build your own 792 00:50:12,120 --> 00:50:15,840 orchestrator, authentication layer, or ragged pipeline. 793 00:50:16,600 --> 00:50:19,960 Sebastian emphasized that the same platform powering features 794 00:50:19,960 --> 00:50:24,440 like researchers or the built in agents mode is the one 795 00:50:24,440 --> 00:50:29,120 developers use, which shortens time to value when productivity 796 00:50:29,120 --> 00:50:34,960 data is central to the workflow. Third, the coding process itself 797 00:50:34,960 --> 00:50:37,560 is changing. As agents become more 798 00:50:37,880 --> 00:50:42,720 collaborators with Gitab, Kopalat and Work IQ pulling in 799 00:50:42,720 --> 00:50:48,080 meeting transcripts, specs, and organizational context, the work 800 00:50:48,080 --> 00:50:52,920 becomes steering and reviewing instead of writing every piece 801 00:50:52,920 --> 00:50:55,920 of code. Sebastian described it as 802 00:50:56,000 --> 00:50:59,040 pairing with an autonomous teammate that can update 803 00:50:59,040 --> 00:51:02,080 instructions and integrate systems and adjust 804 00:51:02,080 --> 00:51:06,080 configurations based on real decisions made across the 805 00:51:06,080 --> 00:51:09,840 organization. And finally, distribution shapes 806 00:51:09,840 --> 00:51:13,960 the architectural choice. If an agent lives purely inside 807 00:51:13,960 --> 00:51:18,120 the employee experience, keeping it native to Microsoft 365 is 808 00:51:18,120 --> 00:51:21,040 simpler. If it must run on websites or 809 00:51:21,280 --> 00:51:25,560 customer facing channels, Copilot Studio or custom hosting 810 00:51:25,680 --> 00:51:29,320 becomes the right path. If you found if you found this 811 00:51:29,320 --> 00:51:32,480 episode useful, subscribe to Mastering AI with the Expert 812 00:51:32,480 --> 00:51:34,120 wherever you're watching the show. 813 00:51:34,520 --> 00:51:39,040 And please stop by at Symbol boulanger.com for more practical 814 00:51:39,040 --> 00:51:42,160 ways to put AI to work. See ya.