1 00:00:00,200 --> 00:00:03,320 All right, so at the beginning of April I went to New York for 2 00:00:03,320 --> 00:00:07,400 the MCP Dev Summit, and while I was there I got the chance to 3 00:00:07,400 --> 00:00:11,080 record a few podcasts with attendees that were at the 4 00:00:11,080 --> 00:00:12,600 event. This is one of those 5 00:00:12,600 --> 00:00:14,160 conversations. Hope you enjoy. 6 00:00:14,440 --> 00:00:19,760 I think we have to look at these agents as untrusted, maybe even 7 00:00:19,760 --> 00:00:21,600 adversarial. How many people are reading this 8 00:00:21,600 --> 00:00:24,880 code or like think about, I think events of the last month 9 00:00:24,880 --> 00:00:27,840 with like Anthropic in the DoD, Think about what we're doing. 10 00:00:27,880 --> 00:00:32,040 We're asking citizen developers to deploy untrusted agents 11 00:00:32,040 --> 00:00:34,440 behind the firewall. And we're frantically wiring 12 00:00:34,440 --> 00:00:36,800 those things up to every single system we can find. 13 00:00:37,480 --> 00:00:41,320 And, and every CEO is basically standing up and saying we must 14 00:00:41,320 --> 00:00:42,440 do this. We must do it now. 15 00:00:42,440 --> 00:00:44,800 We have to take more risk. We have to act like a startup. 16 00:00:44,800 --> 00:00:48,200 I can't think of any part of our business that isn't rapidly 17 00:00:48,200 --> 00:00:53,480 changing where we look at it first and foremost as what do 18 00:00:53,480 --> 00:00:55,440 these tools let us do? But they're only useful, They're 19 00:00:55,440 --> 00:00:59,440 only useful if they have the context of all of the, the 20 00:00:59,440 --> 00:01:01,160 systems and the services and the APIs. 21 00:01:01,160 --> 00:01:03,040 And that's, that's what APIs are at the end of the day. 22 00:01:03,040 --> 00:01:04,959 They're the, they're the front door to all of that stuff. 23 00:01:16,440 --> 00:01:19,560 Hi, I'm Alex Alkever with the Linux Foundation and the Agentic 24 00:01:19,600 --> 00:01:22,080 AI Foundation. We're here at MCP Dev Summit in 25 00:01:22,080 --> 00:01:26,360 New York City and I'm chatting today with Matt Deburgolis. 26 00:01:26,360 --> 00:01:29,040 If I said it the right way, who is the CEO of Apollo Graph QL? 27 00:01:29,840 --> 00:01:32,200 And forgive me if I get anything wrong, correct me, I have no 28 00:01:32,200 --> 00:01:34,880 ego. Thank you for joining us. 29 00:01:34,960 --> 00:01:37,880 Before we get started on this conversation, which is going to 30 00:01:37,880 --> 00:01:41,760 be about API lessons for MCPI, believe is what we were 31 00:01:41,760 --> 00:01:44,960 discussing and tell us a little bit about what a polygraph QL 32 00:01:44,960 --> 00:01:47,400 does. And then how we can chat a 33 00:01:47,400 --> 00:01:52,720 little more about how you're leaning into MCP, particularly 34 00:01:52,720 --> 00:01:56,000 since you're essentially a graph QL as a service provider and you 35 00:01:56,000 --> 00:01:58,800 know, we're generating everything around the API space. 36 00:01:58,800 --> 00:02:02,880 So MCP is a very interesting translational layer for you. 37 00:02:03,120 --> 00:02:05,520 Yeah, sounds good. Thanks for thanks for doing 38 00:02:05,520 --> 00:02:06,640 this. It's a great conference. 39 00:02:07,600 --> 00:02:09,520 Thanks for coming. Energy is awesome. 40 00:02:09,520 --> 00:02:12,480 It it you can't make this stuff up, right? 41 00:02:12,480 --> 00:02:14,520 It it shows there. It's crazy, literally like 42 00:02:14,520 --> 00:02:17,400 things are so I mean like I'm following all. 43 00:02:17,400 --> 00:02:20,320 So each week I do a scrape of all the protocol repos and 44 00:02:20,320 --> 00:02:23,400 you're just seeing things move so quickly, like, oh wait, 45 00:02:23,400 --> 00:02:26,280 somebody signed ACLA or oh wait, you know, they just shipped the 46 00:02:26,280 --> 00:02:28,440 new securities back. It's it's, it's crazy. 47 00:02:28,520 --> 00:02:33,880 I think what's cool about MCP is it, it reminds me of the the 48 00:02:33,880 --> 00:02:36,720 mash up era, right? It's it's, it's for builders, 49 00:02:36,720 --> 00:02:39,640 it's for making stuff. And you know, we could talk 50 00:02:39,640 --> 00:02:44,000 about agents are useful because they can connect to stuff. 51 00:02:44,160 --> 00:02:47,800 So we need an answer to that. And MCP is a big part of it. 52 00:02:47,800 --> 00:02:51,920 And what we've been doing it Apollo for the AP is of large 53 00:02:51,920 --> 00:02:54,600 companies, especially is, is we're, we're basically solving 54 00:02:54,600 --> 00:02:56,680 this problem of how do you, how do you take a whole bunch of 55 00:02:56,680 --> 00:02:59,120 different systems? And AP is that you've got, you 56 00:02:59,120 --> 00:03:01,800 might have thousands of them in a typical enterprise. 57 00:03:01,800 --> 00:03:04,280 And how do you glue those together to make a great user 58 00:03:04,280 --> 00:03:06,600 experience? And this got its start back when 59 00:03:06,600 --> 00:03:10,440 we're talking about building a mobile app or a really great 60 00:03:10,440 --> 00:03:13,200 JavaScript app in the browser. But it's always been about 61 00:03:13,200 --> 00:03:16,240 builders. It's always been about, I want 62 00:03:16,240 --> 00:03:18,200 to make a cool thing. I want to make something my 63 00:03:18,200 --> 00:03:22,880 customers are going to love. And I want to do that on top of 64 00:03:22,880 --> 00:03:26,760 real stuff. And and so now we can have that 65 00:03:26,760 --> 00:03:30,680 whole conversation again, but at an even bigger scale because 66 00:03:30,680 --> 00:03:35,200 we're talking about with AI being able to build so much more 67 00:03:35,640 --> 00:03:40,560 and not just the sort of user facing, you know, first class 68 00:03:40,560 --> 00:03:43,840 mobile app, but think about all the things that are happening 69 00:03:43,840 --> 00:03:46,440 inside the business now that we've got agents on people's 70 00:03:46,440 --> 00:03:50,880 desks, now that we can build software large and small really 71 00:03:50,880 --> 00:03:53,040 quickly. And the part that I'm just so 72 00:03:53,040 --> 00:04:00,000 excited about is we know how much amazing stuff is out there 73 00:04:00,080 --> 00:04:05,240 inside micro services or inside little pieces of SAS or inside 74 00:04:05,240 --> 00:04:07,560 these systems that people have built over the years. 75 00:04:08,000 --> 00:04:10,280 And the more of those we can connect together, the more of 76 00:04:10,280 --> 00:04:11,840 those we can connect to an agent. 77 00:04:11,960 --> 00:04:13,200 I think it's going to change so much. 78 00:04:13,200 --> 00:04:14,760 It's going to make businesses faster. 79 00:04:14,760 --> 00:04:17,200 It's going to be better customer experiences. 80 00:04:18,120 --> 00:04:21,399 It's real exciting. And so from your viewpoint too, 81 00:04:21,399 --> 00:04:24,240 because you've spent a lot of time in the API world and now 82 00:04:24,240 --> 00:04:29,440 you're essentially living in 2 worlds, what is the transition 83 00:04:29,440 --> 00:04:30,680 or? I wouldn't say transition 84 00:04:30,680 --> 00:04:32,040 because that implies one goes away. 85 00:04:32,040 --> 00:04:35,320 Ape guys don't go at all. No, they certainly don't. 86 00:04:35,320 --> 00:04:36,840 No. And like even with like graph 87 00:04:36,840 --> 00:04:39,480 QL, it was still like kind of ascending, you know. 88 00:04:39,880 --> 00:04:42,600 And rest is still, there's a lot more people use rest than graph, 89 00:04:42,600 --> 00:04:46,080 even though graph is arguably much better for AI. 90 00:04:46,800 --> 00:04:50,680 So if you could maybe take a stab at like, what is this 91 00:04:50,800 --> 00:04:55,040 translational layer look like if it works really well for what 92 00:04:55,040 --> 00:04:57,720 we're trying to to to build that world you're describing? 93 00:04:57,800 --> 00:05:00,240 Yeah, I think all these things are complimentary. 94 00:05:01,200 --> 00:05:05,200 With Graphical, for example, it's not about getting rid of 95 00:05:05,200 --> 00:05:08,320 REST AP is or whatever you had before and replacing it with 96 00:05:08,320 --> 00:05:09,880 Graph QL. That's not what Graph QL is. 97 00:05:09,880 --> 00:05:13,000 Graphical is a query language, so it's, it's something that you 98 00:05:13,000 --> 00:05:16,280 can layer on top of your REST AP is and it makes them better. 99 00:05:16,560 --> 00:05:21,240 Like graph QL introduced the idea of strong typing in AP is 100 00:05:21,560 --> 00:05:23,200 it's got this idea of introspection. 101 00:05:23,200 --> 00:05:28,440 So you can ask the graph, what, what do you have for me? 102 00:05:28,440 --> 00:05:30,040 What are the types? What are the fields? 103 00:05:30,840 --> 00:05:32,840 How do they relate? How are they used? 104 00:05:32,840 --> 00:05:34,920 There's all this metadata that comes along with that. 105 00:05:35,720 --> 00:05:38,880 And I think MCP very similarly, right? 106 00:05:38,880 --> 00:05:42,800 Nobody's saying you're going to throw away your underlying AP is 107 00:05:43,440 --> 00:05:45,040 we're talking about a new way to use them. 108 00:05:45,040 --> 00:05:49,320 We're talking about adapting them for the needs of an agent. 109 00:05:49,800 --> 00:05:54,320 But I think really the, the magic thing about MCP is the 110 00:05:54,320 --> 00:05:57,080 flexibility in the free form part of it. 111 00:05:58,640 --> 00:06:01,480 You know, we're, we used to think about AP is as these rigid 112 00:06:02,120 --> 00:06:05,200 things that got designed very carefully a long time ago and 113 00:06:05,200 --> 00:06:08,320 they never change, right? Like, but that's not the world 114 00:06:08,320 --> 00:06:10,520 we're in anymore. Systems can change every day, 115 00:06:10,520 --> 00:06:15,120 and MCP gives us a way to to talk about that and to expose 116 00:06:15,360 --> 00:06:18,560 those ideas to a system. Dive a little deeper on that, if 117 00:06:18,560 --> 00:06:21,080 you can tell, I mean, if you will, on what is 118 00:06:21,480 --> 00:06:23,880 architecturally, what does that start to look like if you're 119 00:06:23,880 --> 00:06:27,120 trying to build out something that takes advantage of the 120 00:06:27,120 --> 00:06:31,040 goodness of graph QL and melds it to the goodness of MCP? 121 00:06:31,160 --> 00:06:34,040 Yeah. So a company has been using 122 00:06:34,040 --> 00:06:37,240 graph QL has, you know, we call it a super graph at Apollo. 123 00:06:37,240 --> 00:06:44,200 They've, they've got this semantic catalog of not just the 124 00:06:45,600 --> 00:06:50,560 low level details of what's in the AP is, but the higher level 125 00:06:51,000 --> 00:06:56,320 meaning of that stuff. And that's magical for AI 126 00:06:56,480 --> 00:06:59,520 because the agent needs that information to make good 127 00:06:59,520 --> 00:07:01,520 decisions. If if I sit down one day and I 128 00:07:01,520 --> 00:07:04,760 say, hey, now I've got clawed code so I can, I can build a 129 00:07:04,760 --> 00:07:08,360 business dashboard with that dashboards using the wrong 130 00:07:08,360 --> 00:07:10,840 stuff. If it misinterprets the meaning 131 00:07:10,840 --> 00:07:13,600 of what's in your AP is that's really bad. 132 00:07:13,600 --> 00:07:15,680 That's, that's worse than not building it at all, right? 133 00:07:15,680 --> 00:07:21,880 So what we're excited about is MCP gives us a way to connect 134 00:07:21,880 --> 00:07:25,160 that graph and connect all those business objects that that 135 00:07:25,160 --> 00:07:31,720 people have now defined to this new world of interactive LLM 136 00:07:31,720 --> 00:07:35,680 harnesses to a world of software that's built by LO Ms. like 137 00:07:35,680 --> 00:07:37,520 Cloud code or, or Codex or whatnot. 138 00:07:39,280 --> 00:07:42,440 And think about what this means. Like now we can sit down and we 139 00:07:42,440 --> 00:07:43,920 can ask questions about our business. 140 00:07:43,920 --> 00:07:48,920 We can say, hey, I've got, I've got a bunch of amazing data 141 00:07:48,920 --> 00:07:52,480 inside this piece of SAS, the screen and the SAS product 142 00:07:52,480 --> 00:07:54,840 doesn't do what I need, but I, I don't need that anymore. 143 00:07:54,840 --> 00:07:56,240 I can just. Right, the users. 144 00:07:56,240 --> 00:07:58,080 Don't talk and the. Users don't want that anymore 145 00:07:58,160 --> 00:07:59,160 either. They're changing, yeah. 146 00:07:59,560 --> 00:08:02,920 Or, you know, a big problem the enterprise has is, is 147 00:08:02,920 --> 00:08:04,760 essentially you've got a silo problem. 148 00:08:05,440 --> 00:08:08,800 Big companies spent the last 20 years building a whole bunch of 149 00:08:08,800 --> 00:08:11,640 silos. That's what a micro service is, 150 00:08:11,640 --> 00:08:13,040 right? It's it's, it goes back to the 151 00:08:13,040 --> 00:08:14,880 Jeff Bezos memo like you're, you're. 152 00:08:14,880 --> 00:08:17,480 Going to do an API for everything. 1 little silo and 153 00:08:17,480 --> 00:08:18,800 you can only talk to it this way. 154 00:08:19,280 --> 00:08:21,080 And that made a lot of sense architecturally. 155 00:08:21,080 --> 00:08:24,400 But now we face this problem, which is I've got all this stuff 156 00:08:25,320 --> 00:08:30,080 and it's it's separated. And what what makes it valuable 157 00:08:30,080 --> 00:08:31,840 is when I can bring it back together. 158 00:08:33,360 --> 00:08:35,640 Most of the questions I want to ask about my business aren't 159 00:08:35,640 --> 00:08:36,919 going to be answered out of one thing. 160 00:08:36,919 --> 00:08:38,559 They're going to be answered out of a whole bunch of them. 161 00:08:39,360 --> 00:08:40,520 And it's the draft, and it's the. 162 00:08:40,520 --> 00:08:43,200 Relationship between? Yeah, exactly. 163 00:08:44,039 --> 00:08:46,640 And, and so give me a couple examples of how you're seeing 164 00:08:46,640 --> 00:08:50,520 people do this in the real world in interesting ways or 165 00:08:50,520 --> 00:08:53,040 surprising ways. Or because like to your point, 166 00:08:54,360 --> 00:08:58,160 having a, a query language that can supply context when you 167 00:08:58,160 --> 00:09:00,640 start to think of things like, you know, additional tuning on 168 00:09:00,640 --> 00:09:04,400 top of open source models, it, it can probably be semi magical 169 00:09:04,400 --> 00:09:08,640 or just have astounding impact. So versus like just working off 170 00:09:08,640 --> 00:09:11,160 of straight, you know, API calls. 171 00:09:11,560 --> 00:09:17,200 Yeah, I think a big part of it is, is this idea that we can 172 00:09:17,680 --> 00:09:20,760 look forward to a world where there are lots more builders and 173 00:09:20,760 --> 00:09:23,280 they don't all have a formal background in software 174 00:09:23,280 --> 00:09:26,120 engineering. Like if you sit down with Claude 175 00:09:26,120 --> 00:09:30,600 code and you ask it to build on top of a bunch of rest AP is 176 00:09:31,160 --> 00:09:34,400 it'll, it'll get there, it'll muddle its way through. 177 00:09:34,400 --> 00:09:38,040 But I think what a lot of us have found is it needs some 178 00:09:38,040 --> 00:09:40,640 guidance. And that guidance means the 179 00:09:40,640 --> 00:09:45,200 person doing the work should have some understanding of how 180 00:09:45,200 --> 00:09:47,680 these AP is work. Like what's a rest AP? 181 00:09:47,680 --> 00:09:49,160 Oh, it's doing this thing. OK, I need to steer it in this 182 00:09:49,160 --> 00:09:52,200 other direction. And so that really blocks the 183 00:09:52,240 --> 00:09:55,920 the thing I want, which is I, I want, you know, somebody in 184 00:09:55,920 --> 00:10:01,440 sales to be able to use Claude to ask sales questions, Right, 185 00:10:01,520 --> 00:10:02,880 Right. Essentially like few shots, 186 00:10:02,880 --> 00:10:07,120 single shot, no shot. And what we get when we've got 187 00:10:07,120 --> 00:10:12,760 this semantics, when we've got a clearer explanation, when we've 188 00:10:12,760 --> 00:10:16,640 got MCP and Graph QL working together so that Claude's not 189 00:10:16,640 --> 00:10:22,120 caught up in the nuance of like, oh, I went to the SAS, I found 190 00:10:22,120 --> 00:10:24,800 it's open API spec, but guess what, that's not actually 191 00:10:24,800 --> 00:10:29,720 accurate. So I had to iterate 345 times 192 00:10:29,720 --> 00:10:32,240 and it it tries to use one API, but it turns out that thing 193 00:10:32,240 --> 00:10:35,280 returns the wrong pagination key because this is just a use of 194 00:10:35,280 --> 00:10:38,040 the API that the the product never really. 195 00:10:38,040 --> 00:10:40,240 Or it's old they didn't Rev it or something. 196 00:10:40,240 --> 00:10:42,880 Like that exactly. If we can cut. 197 00:10:42,880 --> 00:10:47,160 Through that and it's, it's just the crank turning roll up the 198 00:10:47,160 --> 00:10:49,760 sleeve stuff, right. But if we can cut through that 199 00:10:50,200 --> 00:10:57,440 and give the AIA more semantic structured entry point to all of 200 00:10:57,440 --> 00:11:00,520 those systems, then it gets magical. 201 00:11:00,720 --> 00:11:03,120 Because then, then you can actually sit down and say, hey, 202 00:11:04,640 --> 00:11:07,440 go over all the conversations that we've had, the transcripts 203 00:11:07,440 --> 00:11:10,080 of all the conversations that I've got in a, in a, in a, you 204 00:11:10,080 --> 00:11:15,400 know, SAS database and search for mentions of this particular 205 00:11:15,400 --> 00:11:18,040 thing. Then go off and read the annual 206 00:11:18,040 --> 00:11:21,520 reports from that company and go figure out how that matches up 207 00:11:21,520 --> 00:11:24,040 to like what the CE OS talking about in terms of their 208 00:11:24,040 --> 00:11:27,680 priorities. And then I want you to go hunt 209 00:11:27,680 --> 00:11:33,080 down, you know, maybe the competitors of, of that company 210 00:11:33,080 --> 00:11:34,400 and, and what they're talking about. 211 00:11:34,400 --> 00:11:40,600 And suddenly a salesperson is in 5 minutes with no special 212 00:11:40,600 --> 00:11:45,400 technical skill, making use of data that we've always had, but 213 00:11:45,400 --> 00:11:49,920 never had a way to really take advantage of the same way. 214 00:11:50,520 --> 00:11:52,320 And they can do it right before their sales call. 215 00:11:52,320 --> 00:11:56,760 So there's not some complicated, you know, process where I've, 216 00:11:56,800 --> 00:11:58,680 I've got to anticipate that I'm going to have this call, maybe 217 00:11:58,680 --> 00:12:00,440 the call just got scheduled. They almost set it up with an 218 00:12:00,440 --> 00:12:02,360 automated process. They said create a skill. 219 00:12:02,360 --> 00:12:04,360 They run it before every ticks their calendar. 220 00:12:04,400 --> 00:12:06,120 Yeah. And I can give you 20 examples 221 00:12:06,120 --> 00:12:09,880 like this, like where certainly every part of our business is 222 00:12:09,880 --> 00:12:12,200 changing. Yeah, like the way we think 223 00:12:12,200 --> 00:12:15,160 about interviewing, how do we coach people on how they're 224 00:12:15,160 --> 00:12:17,080 doing their interviews? How can we make those better? 225 00:12:17,080 --> 00:12:20,880 There's a feedback cycle that we can start doing in real time 226 00:12:20,880 --> 00:12:23,160 that might have happened once a quarter before, right? 227 00:12:23,360 --> 00:12:28,320 Or how do we do market research? Or I mean, obviously there's a 228 00:12:28,320 --> 00:12:30,840 whole world of how we build software in a different way and, 229 00:12:30,840 --> 00:12:33,520 and how we think about Rd. maps and, and all the things that 230 00:12:33,520 --> 00:12:36,280 come with that. I can't think of any part of our 231 00:12:36,280 --> 00:12:41,080 business that isn't rapidly changing where we look at it 232 00:12:41,080 --> 00:12:44,080 first and foremost as what are these tools? 233 00:12:44,080 --> 00:12:45,480 Let us do? But they're only useful. 234 00:12:45,480 --> 00:12:49,640 They're only useful if they have the context of all of the, the 235 00:12:49,640 --> 00:12:52,240 systems and the services and the AP is, and that's, that's what 236 00:12:52,240 --> 00:12:54,280 AP is are at the end of the day, they're the, they're the front 237 00:12:54,280 --> 00:12:57,800 door to all of that stuff. So in your view then, do we need 238 00:12:57,800 --> 00:13:03,320 to modify the way the AP is work at all if we want to lean into 239 00:13:03,320 --> 00:13:05,600 MCP? Yeah, and agentic better. 240 00:13:05,880 --> 00:13:09,040 The big problem is the AP is were never designed for any of 241 00:13:09,040 --> 00:13:11,880 what I just talked to no. So I I gave you the the 242 00:13:11,880 --> 00:13:13,760 exciting. Version here's the AP is the 243 00:13:13,760 --> 00:13:15,560 rest was not designed. For like here's the cold hard 244 00:13:15,560 --> 00:13:18,760 reality. I mean the the the biggest most 245 00:13:18,760 --> 00:13:22,320 immediate problem is a security problem in my view. 246 00:13:22,720 --> 00:13:26,640 Like these AP is were built for trusted users. 247 00:13:27,800 --> 00:13:30,160 And what I mean by that is like they were meant to be used by 248 00:13:30,160 --> 00:13:33,840 engineers who know how to do engineering, who have some, you 249 00:13:33,840 --> 00:13:36,280 know, they, they review each other's code. 250 00:13:36,280 --> 00:13:40,000 There's a whole bunch of rituals and processes that we've built 251 00:13:40,000 --> 00:13:43,000 over the last couple decades. The rate limited key, etcetera, 252 00:13:43,000 --> 00:13:44,560 etcetera. All that stuff and, and, and, 253 00:13:44,560 --> 00:13:47,240 and underneath all that is human judgement, good human judgement 254 00:13:47,240 --> 00:13:50,880 of people I trust, right. So if you wipe all that away, 255 00:13:50,880 --> 00:13:54,240 you've got a big problem. Because if you think about a 256 00:13:54,240 --> 00:14:00,120 typical API, for example, it's probably going to mix data that 257 00:14:00,120 --> 00:14:02,080 has very different levels of sensitivity. 258 00:14:03,080 --> 00:14:07,840 Like I've got a, I've got an HR system in our company. 259 00:14:08,680 --> 00:14:13,840 That HR system has a list of everybody and their time zone. 260 00:14:13,840 --> 00:14:15,000 That's really useful. I think. 261 00:14:15,040 --> 00:14:17,160 I think that should be at everybody's fingertips, right? 262 00:14:17,240 --> 00:14:19,440 Sure. Who the manager is, it's also 263 00:14:19,440 --> 00:14:22,320 got probably, I'm guessing, I'm not totally sure, but I think it 264 00:14:22,320 --> 00:14:24,600 has like their four O 1K contribution information. 265 00:14:24,720 --> 00:14:25,720 Or their health. Information. 266 00:14:25,720 --> 00:14:28,040 Probably not something right that people should. 267 00:14:28,400 --> 00:14:32,440 Have exactly. And you know, I think just 268 00:14:32,440 --> 00:14:35,400 across the board, whether we're talking about third party SAS 269 00:14:35,400 --> 00:14:40,000 that you bought, we're talking about first party services that 270 00:14:40,000 --> 00:14:42,560 you built. The AP is, are too coarse 271 00:14:42,560 --> 00:14:45,600 grained. They commingle stuff that has a 272 00:14:45,600 --> 00:14:50,240 very different level of, you know, sensitivity or importance. 273 00:14:51,760 --> 00:14:54,960 That was OK in a world where I had control of the other side 274 00:14:54,960 --> 00:14:56,520 and and knew what that code was doing. 275 00:14:56,520 --> 00:15:02,360 But I think we have to look at these agents as untrusted, maybe 276 00:15:02,360 --> 00:15:04,280 even adversarial. But how many people are reading 277 00:15:04,280 --> 00:15:07,680 this code or like think about, I think events of the last month 278 00:15:07,680 --> 00:15:11,920 with like Anthropic and the DoD really tell us how little we 279 00:15:11,920 --> 00:15:15,400 really know about what's happening inside some of the 280 00:15:15,400 --> 00:15:20,000 models in the harnesses. And if you think about the risk 281 00:15:20,000 --> 00:15:26,080 of data leaks and exfiltration, if you think about this as like 282 00:15:26,640 --> 00:15:32,680 a savvy adversary, a competitor, maybe an estate actor, this is 283 00:15:32,680 --> 00:15:36,520 the biggest opportunity to crack through. 284 00:15:36,520 --> 00:15:37,880 I mean, think about what we're doing. 285 00:15:37,960 --> 00:15:42,120 We're asking citizen developers to deploy untrusted agents 286 00:15:42,120 --> 00:15:44,520 behind the firewall. And we're frantically wiring 287 00:15:44,520 --> 00:15:46,880 those things up to every single system we can find. 288 00:15:47,560 --> 00:15:51,400 And, and every CEO is basically standing up and saying we must 289 00:15:51,400 --> 00:15:53,600 do this, we must do it now. We have to take more risk. 290 00:15:53,600 --> 00:15:56,440 We have to act like a startup, you know, like. 291 00:15:56,880 --> 00:15:59,680 Please compromise our AP. Is it's, it's a rock and a hard 292 00:15:59,680 --> 00:16:03,280 place because I think the businesses that do this are 293 00:16:03,280 --> 00:16:05,760 going to go so much faster and they're going to have such a 294 00:16:05,760 --> 00:16:09,520 competitive advantage. But like, at what cost if you 295 00:16:09,520 --> 00:16:11,200 get it wrong? Yeah, no, totally. 296 00:16:11,200 --> 00:16:15,320 I mean, like we were, we were. I was in Amsterdam last week and 297 00:16:15,560 --> 00:16:19,400 chatted with, you know, some of the folks on the kernel, the 298 00:16:19,400 --> 00:16:23,280 Linux kernel and they're seeing now essentially, you know, 0 299 00:16:23,280 --> 00:16:29,000 days come in that are out of not modified frontier models like 300 00:16:29,000 --> 00:16:32,360 consumer frontier models, you know, with very simple prompts 301 00:16:33,080 --> 00:16:35,560 to call them out so totally and like what you're saying. 302 00:16:35,560 --> 00:16:38,040 And if you really want to be targeted, it's even worse. 303 00:16:38,720 --> 00:16:42,840 So what how do you think about solving this? 304 00:16:42,840 --> 00:16:47,400 Or orders like graph QL start because it is so faceted and can 305 00:16:47,400 --> 00:16:49,680 supply exactly so. I'll give you an example. 306 00:16:49,680 --> 00:16:54,360 I, I think, I think there's no question that at a high level, 307 00:16:54,440 --> 00:16:55,960 AI is going to change every part of the stack. 308 00:16:55,960 --> 00:16:57,520 It's going to change how we build software. 309 00:16:57,520 --> 00:17:01,160 So one example of that is I think we have to move to call it 310 00:17:01,160 --> 00:17:05,640 a zero trust model where a a piece of software, an agent has 311 00:17:05,640 --> 00:17:09,760 access to the thinnest slice of what it what it should. 312 00:17:10,520 --> 00:17:14,720 And graph QL is a great architectural starting point for 313 00:17:14,720 --> 00:17:21,400 that because it has field by field semantics. 314 00:17:22,040 --> 00:17:25,560 So a graph QL schema says here are the objects in your 315 00:17:25,560 --> 00:17:27,119 business. Each of those objects has a set 316 00:17:27,119 --> 00:17:29,320 of properties. We call them fields, right? 317 00:17:29,720 --> 00:17:32,760 We can assign rights to those fields. 318 00:17:32,760 --> 00:17:35,960 So we can say a given agent which has, you know, 1 319 00:17:35,960 --> 00:17:39,440 credential, you get this field, this field, this field in this 320 00:17:39,440 --> 00:17:45,920 field and that agent is going to make that request explicit. 321 00:17:46,720 --> 00:17:50,680 When I call a REST endpoint, I just say here's, here's the 322 00:17:50,680 --> 00:17:53,680 arguments, give me back the result and the the API decides 323 00:17:53,680 --> 00:17:56,600 what I'm going to get back. That model doesn't work for an 324 00:17:56,600 --> 00:17:59,480 agent because the API, the micro service, doesn't know what the 325 00:17:59,480 --> 00:18:01,280 agents trying to do. It doesn't know what it is and 326 00:18:01,280 --> 00:18:03,680 it doesn't. It doesn't sound in a position 327 00:18:03,680 --> 00:18:05,640 to enforce. Policy agents composing if the 328 00:18:05,640 --> 00:18:07,920 agents composing the other four or five things or yeah. 329 00:18:08,440 --> 00:18:09,840 And, and this stuff goes pretty deep. 330 00:18:09,840 --> 00:18:12,640 You know, we work with financial companies that I think one 331 00:18:12,640 --> 00:18:19,080 example is support agents only have access to a customer's 332 00:18:19,960 --> 00:18:22,680 billing records if there's an open support ticket. 333 00:18:23,800 --> 00:18:25,680 So those are two totally separate systems that if you 334 00:18:25,680 --> 00:18:27,800 haven't joined together, but like that's a law, that's, 335 00:18:27,960 --> 00:18:30,640 that's not some desired policy. That's actually a legal 336 00:18:30,640 --> 00:18:34,640 requirements placed on a lot of financial companies as part of a 337 00:18:34,640 --> 00:18:38,560 consumer protection stance. So the client has to be able to 338 00:18:38,560 --> 00:18:44,240 specify what it's trying to do. And I think Graph QL gives us a 339 00:18:44,240 --> 00:18:45,960 template for how to go about doing that, right? 340 00:18:45,960 --> 00:18:48,640 Because in Graph QL, you write a query and the query is explicit, 341 00:18:48,880 --> 00:18:52,360 says I want, I want XYZ and W out of the graph and I want them 342 00:18:52,360 --> 00:18:54,600 in this order. And now we're in a position to 343 00:18:54,600 --> 00:18:58,080 say, Yep, that's fine. Or like, if you want that piece 344 00:18:58,080 --> 00:19:01,600 of personal information, we're going to block that request 345 00:19:02,080 --> 00:19:05,600 until we do an approval process that goes through, you know, 346 00:19:05,600 --> 00:19:08,680 somebody who owns the relevant system and can dole out that 347 00:19:08,680 --> 00:19:11,560 access. So essentially almost as a 348 00:19:11,960 --> 00:19:14,360 either as a Shim or if they're just using that as sort of a 349 00:19:14,360 --> 00:19:19,720 native in a primitive construct of their API, it allows them to 350 00:19:20,160 --> 00:19:24,360 assign much more a granularity to access attributes, anything 351 00:19:24,360 --> 00:19:26,840 that they want to share. So a question on top of that, 352 00:19:26,840 --> 00:19:29,680 because this is actually an interesting real world problem 353 00:19:29,680 --> 00:19:32,880 that I saw, I think IBM wrote about it or somebody, but there 354 00:19:32,880 --> 00:19:36,200 was an instance where you essentially had two bots last 355 00:19:36,200 --> 00:19:40,040 agents inside of a company. You know, 1 was tasked with 356 00:19:40,040 --> 00:19:43,760 customer, you know, customer maximizing customer 357 00:19:43,760 --> 00:19:45,960 satisfaction. The other was you may have heard 358 00:19:45,960 --> 00:19:49,600 the story was tasked with, you know, whether to give people 359 00:19:49,600 --> 00:19:52,120 refunds or not. And and like the reward 360 00:19:52,120 --> 00:19:54,600 structure was, you know, customer satisfaction was the 361 00:19:54,600 --> 00:19:56,840 biggest reward. So it actually started handing 362 00:19:56,840 --> 00:19:58,960 out refunds Willy nilly automatically. 363 00:19:59,560 --> 00:20:02,560 So like this is agent to agent. So when we start to have these. 364 00:20:02,760 --> 00:20:05,560 Kinds of dynamics? How do? 365 00:20:05,600 --> 00:20:10,840 How do you think about writing or at the API level starting to 366 00:20:11,200 --> 00:20:13,280 put better strictures or controls around this? 367 00:20:13,640 --> 00:20:16,400 I think the rabbit hole goes pretty deep here, right? 368 00:20:16,400 --> 00:20:19,160 Like you can start at the surface level, which is should 369 00:20:19,160 --> 00:20:21,640 an agent be able to give a refund autonomously? 370 00:20:22,160 --> 00:20:24,920 Maybe not today, right? Maybe tomorrow, maybe maybe 371 00:20:24,920 --> 00:20:27,080 below 50 bucks, right? Like, right. 372 00:20:27,280 --> 00:20:30,760 But you're getting at this much deeper, much more interesting 373 00:20:30,880 --> 00:20:32,800 question, I think, which is like, what's it really going to 374 00:20:32,800 --> 00:20:37,080 look like to have an autonomous agent that's doing the work that 375 00:20:37,840 --> 00:20:39,520 a group of people would have done in the past? 376 00:20:40,120 --> 00:20:43,680 And how do we, you know, not just think about it in terms of 377 00:20:43,680 --> 00:20:46,480 the sort of the basics of security, but like maybe a 378 00:20:46,480 --> 00:20:49,160 quality metric that comes with that? 379 00:20:50,040 --> 00:20:52,960 Are we making a good business decision and how do we measure 380 00:20:52,960 --> 00:20:55,440 that and how do we know? I think the interesting thing 381 00:20:55,440 --> 00:20:59,360 is, have we really had a good way to do that in the in the pre 382 00:20:59,360 --> 00:21:01,120 AI world? Oh, no, absolutely. 383 00:21:01,120 --> 00:21:02,240 This is a new question. A new. 384 00:21:02,360 --> 00:21:03,280 Concept. Yeah. 385 00:21:04,360 --> 00:21:07,280 So I think there's going to be a lot of innovation here. 386 00:21:07,440 --> 00:21:12,720 And I mean, it's interesting, like maybe a faster turn around 387 00:21:12,720 --> 00:21:17,200 time on that decision is worth it if you're offering maybe more 388 00:21:17,200 --> 00:21:18,680 refunds than you really needed to. 389 00:21:19,280 --> 00:21:21,400 Maybe it's not. Maybe that trade off changes 390 00:21:21,400 --> 00:21:25,760 over time. And I think we're going to find 391 00:21:25,760 --> 00:21:31,840 a lot of energy around exploring the space of business decision 392 00:21:31,840 --> 00:21:34,840 making with more and more autonomy, because I, I think at 393 00:21:34,840 --> 00:21:37,040 the end of the day, it's the clock speed of the business that 394 00:21:37,040 --> 00:21:41,680 really determines that company's prospects. 395 00:21:41,680 --> 00:21:43,800 I mean, it's a, it's a very rapidly changing world. 396 00:21:43,800 --> 00:21:48,760 And if you know, fast beats right, I think in a lot of in a 397 00:21:48,760 --> 00:21:51,640 lot of moments today and that's an example of it. 398 00:21:51,880 --> 00:21:54,000 Yeah, I completely agree. Because I think that that's like 399 00:21:54,000 --> 00:21:57,640 an instance where user expectations either in the 400 00:21:57,640 --> 00:22:00,520 enterprise Oregon consumer, they want everything faster all the 401 00:22:00,520 --> 00:22:03,400 time, you know, and they're and, and, and it really tends to 402 00:22:03,400 --> 00:22:06,200 drive process in that direction. And sort of the flip side of 403 00:22:06,200 --> 00:22:08,640 that which I'm actually interested in your thoughts on 404 00:22:08,640 --> 00:22:11,080 which I think builds on top of this on on the one hand, you 405 00:22:11,080 --> 00:22:14,400 have security, on the other hand, you have trust, you know, 406 00:22:14,400 --> 00:22:18,840 in A2 package and trust will enable speed if you do it right. 407 00:22:19,680 --> 00:22:22,960 So what are, how are you seeing people think or how do you think 408 00:22:22,960 --> 00:22:28,520 about, you know, creating trust systems within these complex 409 00:22:28,520 --> 00:22:30,000 autonomous interactions? Yeah, and. 410 00:22:30,000 --> 00:22:33,080 Can I add a third because I think about a stool, There's 411 00:22:33,080 --> 00:22:37,960 also there's experience or friction because because I I 412 00:22:37,960 --> 00:22:41,320 think a great security footprint isn't going to be worth anything 413 00:22:41,320 --> 00:22:42,400 if you have everybody. Works right? 414 00:22:42,400 --> 00:22:44,360 Everybody's mad about it. Yeah, yeah. 415 00:22:44,360 --> 00:22:49,760 And and, and they will. So I think of it as a it's, it's 416 00:22:49,760 --> 00:22:53,240 really a, an experience problem at the end of the day. 417 00:22:53,280 --> 00:22:58,920 Like if we only solve an access control problem, it's putting up 418 00:22:58,920 --> 00:23:02,520 a bunch of stop signs. But if we can look at it as, 419 00:23:03,440 --> 00:23:09,560 hey, there is a wealth of incredibly valuable data that 420 00:23:09,560 --> 00:23:12,680 you now you can have access to and you don't have to ask 421 00:23:13,080 --> 00:23:15,800 because all that is taken care of behind the scenes for you, 422 00:23:16,120 --> 00:23:18,320 right? And a lot of stuff doesn't 423 00:23:18,320 --> 00:23:22,120 require a ton of scrutiny. You probably want to know what's 424 00:23:22,120 --> 00:23:23,800 happening. You probably want an audit log 425 00:23:23,800 --> 00:23:25,880 of all your agents and what they're up to. 426 00:23:26,720 --> 00:23:30,720 But if we can empower people to take that first step, that 427 00:23:30,720 --> 00:23:33,400 builds a lot of momentum, I think for what comes second and 428 00:23:33,400 --> 00:23:35,480 third and 4th, because I think everybody had that experience 429 00:23:35,480 --> 00:23:38,040 with AI, right? Like you kind of read about it, 430 00:23:38,040 --> 00:23:41,200 but it's when you have your first touch with Oh my gosh, 431 00:23:41,200 --> 00:23:42,920 that really went as well as I thought. 432 00:23:42,920 --> 00:23:46,920 Or you see a hackathon where somebody, you know, build 433 00:23:46,920 --> 00:23:49,320 something and you're like, wait, you built that and you did it in 434 00:23:49,320 --> 00:23:51,960 a day, right? Kind of forces you to rethink 435 00:23:51,960 --> 00:23:53,120 your assumptions about. A lot. 436 00:23:53,120 --> 00:23:56,080 I built a custom MD scraper that basically pulls from YouTube 437 00:23:56,080 --> 00:23:57,760 linked. I mean like anything I wanted to 438 00:23:57,760 --> 00:24:00,240 do the half hour vibe coding on on clock code. 439 00:24:00,480 --> 00:24:02,200 They brought it locally, but still, you know, yeah. 440 00:24:03,200 --> 00:24:05,120 So I just keep coming back to I, I think there's a really 441 00:24:05,120 --> 00:24:09,400 positive opportunity here for companies to get really better 442 00:24:09,400 --> 00:24:12,040 at what they do. I mean, I've, I've loved the 443 00:24:12,040 --> 00:24:14,800 technology, I've, I've been into infrastructure for a long time. 444 00:24:14,800 --> 00:24:19,640 But I think what really drives me for a lot of this is I, I 445 00:24:19,640 --> 00:24:24,200 want to, I'm an end user. I want to see better products, I 446 00:24:24,200 --> 00:24:26,280 want to see better customer experiences. 447 00:24:26,960 --> 00:24:28,680 I want that stuff to really be great. 448 00:24:29,000 --> 00:24:34,960 And the AI thing goes way beyond helping companies write that 449 00:24:34,960 --> 00:24:37,040 software faster that I use everyday. 450 00:24:37,040 --> 00:24:40,040 It's going to make the company a more customer centric company 451 00:24:40,800 --> 00:24:43,000 and it's going to show up in all kinds of different places, I 452 00:24:43,000 --> 00:24:48,800 think. And so it's it's, it is a 453 00:24:48,800 --> 00:24:52,640 security and a trust problem. But I think it's also more and 454 00:24:52,640 --> 00:24:56,600 more, the way I think about it at least is you get there by 455 00:24:56,600 --> 00:25:01,000 meeting people where they are and showing them a, an exciting 456 00:25:01,000 --> 00:25:02,520 path. Because I think everybody wants 457 00:25:02,520 --> 00:25:06,120 to be a builder at some level, you know, like it's fun. 458 00:25:06,640 --> 00:25:10,480 So, so related to that, although much deeper down in the weeds, 459 00:25:10,960 --> 00:25:15,720 if you had a wish list of changes to the MCP spec or to 460 00:25:15,720 --> 00:25:19,280 the Graph QL spec to further this vision, what would be the 461 00:25:19,280 --> 00:25:21,760 things that you would you would sort of say you think could 462 00:25:21,760 --> 00:25:23,840 really help things along and if I'm getting you into how? 463 00:25:23,880 --> 00:25:25,880 Much time do we? Have if I'm getting, I mean, if 464 00:25:25,880 --> 00:25:27,480 I'm getting into political trouble, then. 465 00:25:27,480 --> 00:25:30,080 You know, No, no, no. I mean, I've, I've talked about 466 00:25:30,080 --> 00:25:32,480 the graph QL piece in, in a lot of places. 467 00:25:33,280 --> 00:25:36,520 Well, I think the more context you can give the model, the more 468 00:25:36,520 --> 00:25:40,720 capable they get. So a lot of this stuff in, in 469 00:25:40,720 --> 00:25:45,680 both MCP and graph QL I think goes to the idea of providing 470 00:25:45,680 --> 00:25:50,520 more structured context. There's an opportunity to, I 471 00:25:50,520 --> 00:25:56,520 think provide a richer amount of like, if you think about what 472 00:25:56,520 --> 00:25:58,920 this means, how does a human think about what something means 473 00:25:58,920 --> 00:26:01,480 while they think about how's it being used elsewhere, They think 474 00:26:01,480 --> 00:26:04,200 about access patterns. They think about, I mean, all 475 00:26:04,200 --> 00:26:07,840 that human judgement. We don't, it's not always at the 476 00:26:07,840 --> 00:26:10,760 front of our mind, but it's what's driving our our decisions 477 00:26:10,760 --> 00:26:12,280 as developers. So we've got to get all that to 478 00:26:12,280 --> 00:26:14,440 the models and there's a whole bunch of stuff you can do there. 479 00:26:14,760 --> 00:26:21,800 I think another piece is just if we think about adapting the AP 480 00:26:21,800 --> 00:26:23,800 is that we have which aren't going anywhere as you said to 481 00:26:23,800 --> 00:26:27,600 what we're trying to do in practice, an agent in many cases 482 00:26:27,600 --> 00:26:30,120 is going to want to fetch a bunch of data out of a system 483 00:26:30,120 --> 00:26:33,200 and the only way to do that is going to be to repeatedly call 484 00:26:33,200 --> 00:26:36,080 something to paginate across a whole bunch of stuff. 485 00:26:36,080 --> 00:26:39,440 Well that that's really not there's a lot of problems with 486 00:26:39,440 --> 00:26:40,840 that access pattern. Sure. 487 00:26:40,840 --> 00:26:45,440 So I think there are capabilities we can think about. 488 00:26:46,080 --> 00:26:49,560 Sequel's probably a good model for some of this stuff where the 489 00:26:50,600 --> 00:26:54,320 AP is get more sophisticated. We can move beyond sort of a 490 00:26:54,320 --> 00:26:59,320 basic request response model and start talking about things like 491 00:26:59,360 --> 00:27:02,760 aggregation and sorting and filtering and the kinds of 492 00:27:02,760 --> 00:27:05,880 things that an agent can quickly express. 493 00:27:06,480 --> 00:27:08,840 They can be pushed further down the stack because the closer we 494 00:27:08,840 --> 00:27:12,720 can get those actions down to the core systems, the more 495 00:27:12,720 --> 00:27:13,880 efficient all that's going to get. 496 00:27:14,040 --> 00:27:17,120 So like essentially the task or the whatever the you know the 497 00:27:17,120 --> 00:27:20,920 goal, the work becomes expressed in a more complex query 498 00:27:20,920 --> 00:27:24,400 structure that further down it can avoid some of the some of 499 00:27:24,400 --> 00:27:26,640 the pitfalls of traditional API behaviors because. 500 00:27:26,920 --> 00:27:28,960 I mean, as an example of like, the higher we can go up in the 501 00:27:28,960 --> 00:27:31,440 abstraction chain, the more capable it's going to become. 502 00:27:31,560 --> 00:27:36,240 And and you know, I, I think that applies to APIs just as 503 00:27:36,240 --> 00:27:39,080 well. And at a high level, I think the 504 00:27:39,080 --> 00:27:42,640 question is just like, and do agents want a different kind of 505 00:27:42,640 --> 00:27:45,840 set of access patterns like most AP is were built either for 506 00:27:45,840 --> 00:27:50,080 screens or they were built for least common denominator, you 507 00:27:50,080 --> 00:27:53,200 know, service to service communication like a very domain 508 00:27:53,200 --> 00:27:57,160 oriented, you need object 25 here you go, right. 509 00:27:57,760 --> 00:28:01,560 So I just think what we're going to find is that agents want 510 00:28:02,360 --> 00:28:05,120 other things than that. They're going to want to crawl 511 00:28:05,120 --> 00:28:06,440 over that data in a different way. 512 00:28:06,440 --> 00:28:08,880 We can anticipate some of that. I think some of it will still 513 00:28:08,880 --> 00:28:12,000 learn as we go, but I think there's going to be a lot of 514 00:28:12,000 --> 00:28:16,960 value in adding that kind of higher level semantics to these 515 00:28:16,960 --> 00:28:19,440 things. There's a lot more though. 516 00:28:19,440 --> 00:28:22,160 I mean, we could talk about real time and streaming and push. 517 00:28:22,160 --> 00:28:24,080 That was a topic from this morning, right? 518 00:28:24,080 --> 00:28:27,040 Like AII think fundamentally is kind of a real time thing. 519 00:28:27,560 --> 00:28:29,560 It's certainly accelerating the way we interact with our 520 00:28:29,560 --> 00:28:30,640 technology. Yeah. 521 00:28:30,640 --> 00:28:33,440 And and I think both MCP and Graph QL have some, you know, 522 00:28:34,520 --> 00:28:37,520 room for improvement on that front because again, it wasn't 523 00:28:37,640 --> 00:28:41,120 maybe at the heart of what the original workloads most needed. 524 00:28:41,640 --> 00:28:43,760 There's hints of it, but but there's a lot more that we can 525 00:28:43,760 --> 00:28:47,600 build there. I'm very interested sort of as 526 00:28:47,680 --> 00:28:49,920 because you live in at the intersection of two very 527 00:28:49,920 --> 00:28:52,920 interesting worlds. I mean, because in graph QL, I 528 00:28:52,920 --> 00:28:55,360 mean, you don't see a lot of people here that are very active 529 00:28:55,360 --> 00:28:57,120 in the API spec community as well. 530 00:28:57,520 --> 00:29:00,560 I mean some, but it's, you know, not, not you, you sit squarely 531 00:29:00,560 --> 00:29:04,200 in the middle and it's a so you mean you have a very different 532 00:29:04,200 --> 00:29:07,280 and sort of not different, but a nuanced view of this or, you 533 00:29:07,280 --> 00:29:08,600 know, an understanding of both sides. 534 00:29:08,600 --> 00:29:11,200 It's really different. Yeah, it's it's, I feel like 535 00:29:11,200 --> 00:29:16,200 we've made contact with business world now and there's no 536 00:29:16,200 --> 00:29:18,120 surprises there at the technical level. 537 00:29:18,440 --> 00:29:20,920 But there's a lot here, here. I'll leave you with this. 538 00:29:21,080 --> 00:29:26,440 I've learned over the work we've done that AP is, if you think 539 00:29:26,440 --> 00:29:29,520 about it, are a handshake, their contract. 540 00:29:29,640 --> 00:29:33,560 And in a company, especially a mature company, especially one 541 00:29:33,560 --> 00:29:37,040 where you've got thousands of these, this is really a people 542 00:29:37,040 --> 00:29:39,440 problem. At its heart, this is all about 543 00:29:40,200 --> 00:29:43,720 a multi stakeholder alignment and negotiation. 544 00:29:44,240 --> 00:29:50,000 I think AI can offer a lot of benefit to navigating those 545 00:29:50,000 --> 00:29:53,120 conversations and navigating the kinds of changes that you're 546 00:29:53,120 --> 00:29:55,160 going to want to make. But I think it's also important 547 00:29:55,200 --> 00:29:57,120 to remember that like none of that's going to go away. 548 00:29:57,120 --> 00:30:01,000 I think we're, we're always going to have people accountable 549 00:30:01,000 --> 00:30:05,360 for the quality of these systems and the correctness and the 550 00:30:05,360 --> 00:30:08,160 reliability and so on. And I think we're going to have 551 00:30:08,160 --> 00:30:10,080 people accountable for the quality and correctness and 552 00:30:10,080 --> 00:30:13,600 reliability of the agentic workloads that they build. 553 00:30:14,480 --> 00:30:17,960 And if we recognize that upfront and if we approach this not just 554 00:30:17,960 --> 00:30:20,720 from the point of view of like what's the transport look like 555 00:30:20,720 --> 00:30:23,960 and how can we like add these semantics here to this protocol, 556 00:30:23,960 --> 00:30:26,920 but we also look at it from the point of view of how is the 557 00:30:26,920 --> 00:30:28,880 organization at scale going to change? 558 00:30:28,880 --> 00:30:31,160 I know that sounds really abstract, but I've I've just 559 00:30:31,160 --> 00:30:35,480 found over and over and over again, people are are in a 560 00:30:35,480 --> 00:30:38,200 position to help drive these things. 561 00:30:38,200 --> 00:30:46,120 But it's really about because the API is is has two sides or 562 00:30:46,120 --> 00:30:48,480 because you might have 1000 consumers of your API. 563 00:30:48,760 --> 00:30:51,720 It really is about an organizational transformation, 564 00:30:52,280 --> 00:30:54,040 which is exciting. That's a really good point to 565 00:30:54,040 --> 00:30:57,080 end on. I mean essentially change NCP 566 00:30:57,280 --> 00:30:59,880 and plus API equals change management because you've 567 00:30:59,880 --> 00:31:01,440 democratized access to all of these. 568 00:31:01,440 --> 00:31:03,480 Things at the speed of. AI, for better or for worse. 569 00:31:03,560 --> 00:31:06,640 Yeah, exactly. Yeah, it's for better but but 570 00:31:06,640 --> 00:31:09,400 it, but it's not going to come without challenges. 571 00:31:09,400 --> 00:31:12,440 And I'm just, I'm really eager to see that through because I, I 572 00:31:12,440 --> 00:31:15,280 think there's a, an exciting world on the other side of I. 573 00:31:15,600 --> 00:31:18,000 Think somebody said recently Yolo responsibly or something 574 00:31:18,000 --> 00:31:21,000 like that which but. I like that, yeah. 575 00:31:21,120 --> 00:31:23,480 Yeah, well, you're enabling parts of that. 576 00:31:23,560 --> 00:31:25,440 Yeah, yeah. Thank you very much, Matt for 577 00:31:25,440 --> 00:31:26,840 taking the time. I appreciate it. 578 00:31:27,000 --> 00:31:29,760 Likewise, hopefully get to talk to you in another conference. 579 00:31:29,880 --> 00:31:30,320 Sounds good.