WEBVTT 1 00:00:07.759 --> 00:00:08.039 Warren. 2 00:00:08.039 --> 00:00:10.720 I'm pretty excited for today's episode. Know why is that 3 00:00:11.759 --> 00:00:13.880 because I have a lot of questions about this topic, 4 00:00:14.279 --> 00:00:18.440 Like I've heard that I've heard the phrase MCP so much, 5 00:00:19.480 --> 00:00:21.879 and I just have many questions. 6 00:00:22.280 --> 00:00:23.679 I mean, at some point you're going to have heard 7 00:00:23.679 --> 00:00:26.719 it too much. And with that, maybe I'll drop a 8 00:00:26.760 --> 00:00:30.199 little fact for the episode. There was a little research 9 00:00:30.280 --> 00:00:33.719 done not too long ago about the adverse impacts of 10 00:00:34.679 --> 00:00:38.159 mentioning artificial intelligence in product and service names, and they 11 00:00:38.159 --> 00:00:42.479 found that it actually significantly decreases consumer trust. Really, I 12 00:00:42.520 --> 00:00:43.320 think that tracks. 13 00:00:43.880 --> 00:00:45.880 Yeah, it seems reasonable. 14 00:00:46.640 --> 00:00:48.359 Like you go to Starbucks and you get some coffee 15 00:00:48.359 --> 00:00:50.399 and they're like, now with AI included, are you like, 16 00:00:50.759 --> 00:00:51.719 are you gonna be happy for that? 17 00:00:53.920 --> 00:00:59.039 Absolutely? Can I get my double vine latte with AI 18 00:00:59.240 --> 00:01:00.000 on a blockchain? 19 00:01:00.560 --> 00:01:02.240 We'll come with MCP on the side. 20 00:01:02.560 --> 00:01:03.200 Okay. 21 00:01:03.240 --> 00:01:03.560 Cool. 22 00:01:04.159 --> 00:01:06.640 So speaking of which, Gil, you're here to talk to 23 00:01:06.760 --> 00:01:07.560 us about MCP. 24 00:01:08.200 --> 00:01:09.959 Yeah, I'm very excited too, and I am one of 25 00:01:09.959 --> 00:01:12.640 those people who's actually heard MCP way too many times, 26 00:01:12.719 --> 00:01:14.280 so happy to do that. 27 00:01:14.760 --> 00:01:19.519 All right, Before we jump into that, give me a 28 00:01:19.560 --> 00:01:20.719 little bit about your background. 29 00:01:21.159 --> 00:01:24.239 Absolutely, So, as you mentioned, I'm Gil, I am the 30 00:01:24.280 --> 00:01:27.920 co founder and CTO of Merge and Merge is a 31 00:01:27.959 --> 00:01:32.319 platform that offers unified epis to help companies offer integrations 32 00:01:32.519 --> 00:01:35.799 with a ton of different products in any specific vertical 33 00:01:35.840 --> 00:01:39.599 from ticketing, CRM, file storage, and so on, and a 34 00:01:39.599 --> 00:01:41.959 lot more coming in the AI space MCP all of 35 00:01:41.959 --> 00:01:43.400 that as we'll talk about today. 36 00:01:44.040 --> 00:01:45.120 But before that, I. 37 00:01:45.079 --> 00:01:47.359 Went to college in New York and then ended up 38 00:01:47.400 --> 00:01:49.920 going straight into tech. So I worked at LinkedIn out 39 00:01:49.959 --> 00:01:54.280 in San Francisco, and then worked at a couple smaller startups, 40 00:01:54.480 --> 00:01:57.519 which ultimately led me to this problem of integration, seeing 41 00:01:57.519 --> 00:02:00.239 how it's just bogging down space, so I decided to. 42 00:02:00.040 --> 00:02:02.200 Star Merge to tackle that problem. And it's been great. 43 00:02:02.239 --> 00:02:04.879 It's been about five years now, four years since coming 44 00:02:04.879 --> 00:02:06.640 out of Stealth. We've gone from zero to one hundred and 45 00:02:06.640 --> 00:02:09.599 ten employees. We have almost fifteen thousand free and paying 46 00:02:09.599 --> 00:02:12.520 customers all around the world. We have our three offices 47 00:02:12.560 --> 00:02:14.680 in San Francisco, New York, and Berlin. 48 00:02:15.319 --> 00:02:21.360 Right on right on. So with four years like that's 49 00:02:21.400 --> 00:02:24.199 a pretty solid record for a startup, you're feeling pretty 50 00:02:24.199 --> 00:02:25.280 confident about this. 51 00:02:26.080 --> 00:02:26.439 Yeah. 52 00:02:26.479 --> 00:02:28.840 Absolutely, I think that we're seeing that this is a 53 00:02:28.879 --> 00:02:32.400 bigger problem than we have ever even envisioned. Everyone needs integrations. 54 00:02:32.439 --> 00:02:35.199 The problems only getting worse now with AI. You have 55 00:02:35.280 --> 00:02:38.800 these models that have essentially ingested the full public corpus 56 00:02:38.800 --> 00:02:41.639 of Internet data, and all that's left is private data. 57 00:02:41.680 --> 00:02:42.639 And that's what Merge does. 58 00:02:42.680 --> 00:02:46.080 We specifically help companies get access to their customers data. 59 00:02:46.400 --> 00:02:48.520 So we're excited about the problem now, but also where 60 00:02:48.520 --> 00:02:53.159 it's going, both in the traditional API and integration building 61 00:02:53.159 --> 00:02:55.400 space as well as all of the upcoming AI and 62 00:02:55.520 --> 00:02:57.879 MCP driven integrations right on. 63 00:02:58.080 --> 00:03:02.879 So like is the concept there? Instead of having to 64 00:03:02.919 --> 00:03:05.919 go and figure out the API docs for the fifteen 65 00:03:05.960 --> 00:03:09.360 different services I need to integrate with, I just connect 66 00:03:09.400 --> 00:03:13.919 with Merge and you're like the relay to those services 67 00:03:13.919 --> 00:03:16.479 for me, and I just have to talk to one service. 68 00:03:16.520 --> 00:03:17.400 That's exactly right. 69 00:03:17.479 --> 00:03:21.520 So an example here brax Ramp, which are corporate cards. 70 00:03:21.680 --> 00:03:23.719 Their customers have merged and they use us to power 71 00:03:23.759 --> 00:03:27.400 a few use cases. But notably they want to automatically 72 00:03:27.400 --> 00:03:31.560 onboard employees of companies that use their credit cards, automatically 73 00:03:31.599 --> 00:03:33.599 mail them a credit cards to their home address based 74 00:03:33.639 --> 00:03:35.719 on their title. Maybe give them twenty dollars a day 75 00:03:35.719 --> 00:03:39.000 for lunch if they're an engineer, for example, and then 76 00:03:39.159 --> 00:03:41.599 they want to terminate those cards when person leaves the company, 77 00:03:41.599 --> 00:03:44.639 and manually managing all of that is impossible, and so 78 00:03:45.120 --> 00:03:47.199 instead they want to integrate with the HR systems to 79 00:03:47.199 --> 00:03:49.039 be able to pull in all this data. But some 80 00:03:49.080 --> 00:03:52.599 of their customers use Bamboo HR. Gusto namely worked a 81 00:03:52.639 --> 00:03:55.560 SAP and they have to build all of those integrations. 82 00:03:55.599 --> 00:03:58.719 So instead they integrate once with us, and we integrate 83 00:03:58.719 --> 00:04:01.400 and sort of normalize all that data to one format that. 84 00:04:01.360 --> 00:04:02.159 They integrate with. 85 00:04:02.680 --> 00:04:04.120 And then again we do that for HR, but we 86 00:04:04.159 --> 00:04:06.759 also do that for a lot of other platforms like ticketings, RM, 87 00:04:06.840 --> 00:04:08.000 file storage and so on. 88 00:04:08.360 --> 00:04:12.840 Right on, that's cool, that's cool. So let's jump into MCP. 89 00:04:13.360 --> 00:04:17.959 Give me the give me like the layman's version of 90 00:04:18.000 --> 00:04:19.040 what MCP is. 91 00:04:19.879 --> 00:04:22.879 Yeah, so there's a lot of ways to think about MCP, 92 00:04:23.079 --> 00:04:25.480 but I think the important node here is it's actually 93 00:04:25.480 --> 00:04:28.639 a really simple concept. It was a standard that was 94 00:04:28.720 --> 00:04:32.199 reached similar to other protocols and standards of the Internet 95 00:04:32.240 --> 00:04:35.360 that ultimately aren't so complicated, but solved the need, which 96 00:04:35.399 --> 00:04:38.120 was that there was a major lack of standardization. And 97 00:04:38.160 --> 00:04:41.439 so when we think about the history of APIs building 98 00:04:41.439 --> 00:04:45.160 an integration required you to go to API documentation and 99 00:04:45.240 --> 00:04:47.560 explicitly say I'm going to take data from this point 100 00:04:47.800 --> 00:04:49.399 and I'm going to move it to this point. A 101 00:04:49.399 --> 00:04:52.519 lot of complications there, but overall you're doing that. Now 102 00:04:52.600 --> 00:04:56.000 in the agentic era, you want to expose those API 103 00:04:56.120 --> 00:04:58.879 calls essentially to an agent so that you can actually 104 00:04:58.879 --> 00:05:01.120 give it arms as a to them being these sort 105 00:05:01.160 --> 00:05:03.240 of you know, things to say, Hey, here's how you 106 00:05:03.279 --> 00:05:06.360 solve this. Go log into your Salesforce account, click this button, 107 00:05:06.439 --> 00:05:09.439 click that instead letting the agents actually take those actions 108 00:05:09.480 --> 00:05:12.600 for you. And so MCP is essentially a way to 109 00:05:12.680 --> 00:05:15.839 make it so that those API calls or actions or 110 00:05:15.879 --> 00:05:19.920 tool calls is as you would say, are available and 111 00:05:20.000 --> 00:05:21.600 exposed to the agents in a way that it can 112 00:05:21.639 --> 00:05:24.720 easily understand, and so then it can formulate a workflow 113 00:05:24.720 --> 00:05:28.519 knowing what tools via MCP are available to it and 114 00:05:28.560 --> 00:05:29.639 then make those calls. 115 00:05:29.959 --> 00:05:33.040 What would stop an agent from being able to integrate 116 00:05:33.120 --> 00:05:36.680 with the existing API docs or one of the standards 117 00:05:36.720 --> 00:05:39.959 like open API Specification or whatever Ataboose is using with 118 00:05:40.040 --> 00:05:43.480 Smithy just consume that and be able to generate the 119 00:05:43.519 --> 00:05:45.079 appropriate calls into the APIs. 120 00:05:45.639 --> 00:05:46.639 So there's a. 121 00:05:46.639 --> 00:05:48.959 Few things that are that are differentiators, but overall, I 122 00:05:49.000 --> 00:05:50.759 will say, at a high level, I'm with you. I 123 00:05:50.759 --> 00:05:52.759 think it's it's not even a hot take anymore to 124 00:05:52.759 --> 00:05:55.639 say that MCP doesn't actually do all that much. You 125 00:05:55.720 --> 00:05:59.279 have API documentation that's built for humans to build integrations. 126 00:05:59.439 --> 00:06:01.120 You have open API specs, which. 127 00:06:00.920 --> 00:06:03.439 Were kind of the next evolution of hey, some static 128 00:06:03.480 --> 00:06:06.439 script can understand my API and build docs or SDKs 129 00:06:06.480 --> 00:06:08.560 out of it. And then finally you have MCP, which 130 00:06:08.600 --> 00:06:12.360 is just another type of rapper that allows agents to interact. 131 00:06:12.560 --> 00:06:15.959 I think the difference is MCP is not stateless like 132 00:06:16.000 --> 00:06:19.319 documentation or like you know, like an open API spect 133 00:06:19.480 --> 00:06:22.240 but instead it's an actual running server that stores credentials, 134 00:06:22.399 --> 00:06:26.519 that manages sessions and can be stateful. So it's it's 135 00:06:26.560 --> 00:06:29.399 not ultimately adding a ton more, but it does unlock 136 00:06:29.439 --> 00:06:33.160 a few additional abilities that are necessary for an agent 137 00:06:33.199 --> 00:06:33.959 to take actions. 138 00:06:34.519 --> 00:06:36.600 That's I feel like a little scary, like I've I've 139 00:06:36.680 --> 00:06:39.639 designed the perfect service that is stateless in every way 140 00:06:39.680 --> 00:06:41.879 with you know, I've thought a lot about what the 141 00:06:41.959 --> 00:06:44.120 endpoint should be, and in order for an agent to 142 00:06:44.160 --> 00:06:47.160 work with it, we're saying you need to actually forget 143 00:06:47.199 --> 00:06:50.560 everything that you've built so far, make a different doc 144 00:06:50.639 --> 00:06:53.199 that is readable, and also start storing state and do 145 00:06:53.519 --> 00:06:55.680 a lot of extra stuff that you specifically didn't want 146 00:06:55.720 --> 00:06:56.839 in your service to begin with. 147 00:06:58.000 --> 00:07:00.360 Yeah, it's true, I think that, you know, when you 148 00:07:00.399 --> 00:07:02.480 think about like a statically coded app, though you would 149 00:07:02.480 --> 00:07:04.079 pull from you know, let's say that you want to 150 00:07:04.160 --> 00:07:07.639 update all records, all ticketing system tickets that have Gill 151 00:07:07.680 --> 00:07:09.800 in the title. You have to pull in all those tickets, 152 00:07:09.879 --> 00:07:11.720 then you have to iterate through them and modify them, 153 00:07:11.759 --> 00:07:13.360 and then you need to write them back. So there's 154 00:07:13.360 --> 00:07:17.680 this notion of like state needs to be maintained between calls. 155 00:07:17.839 --> 00:07:19.800 And an agent. An agent can do that. 156 00:07:19.759 --> 00:07:22.519 But you can also write tools on the server that actually, 157 00:07:22.519 --> 00:07:24.879 you know, manage a lot of that. So what you're 158 00:07:24.879 --> 00:07:27.639 doing is you're deciding how much complexity you want to 159 00:07:27.680 --> 00:07:29.600 expose to the agent and how much you want to 160 00:07:29.639 --> 00:07:32.720 wrap behind a statically coded tool that the agent can 161 00:07:32.759 --> 00:07:33.240 then call. 162 00:07:33.600 --> 00:07:37.360 So normally you would push the if your API was 163 00:07:37.759 --> 00:07:42.079 fetch ticket and then update ticket as two independent API calls, 164 00:07:42.120 --> 00:07:44.199 and that's all you offered. You were like, we don't 165 00:07:44.240 --> 00:07:48.199 offer bulk support, and the client would be responsible for 166 00:07:48.240 --> 00:07:51.079 actually pulling each one of those and then updating them individually. 167 00:07:51.079 --> 00:07:52.959 And if we're saying that you're building a proxy in 168 00:07:53.000 --> 00:07:55.959 between there, it's because the API that you're offering to 169 00:07:56.040 --> 00:07:59.240 end users wasn't valuable enough that it wasn't actually solving 170 00:07:59.279 --> 00:08:01.680 the needs that they would frequently have. I don't think 171 00:08:01.800 --> 00:08:04.399 bulk actions are very common, so I think in that 172 00:08:04.439 --> 00:08:06.439 way this is a little interesting to saying, like, well, 173 00:08:06.680 --> 00:08:08.279 now you need to actually start thinking about what the 174 00:08:08.360 --> 00:08:11.279 value of your service is actually offering, because if people 175 00:08:11.319 --> 00:08:13.920 want to do these bulk actions, you may think about 176 00:08:14.000 --> 00:08:16.600 wanting to provide that, and this is where you would 177 00:08:16.639 --> 00:08:20.439 prevent actually provide this logic in a agent based system. 178 00:08:21.120 --> 00:08:24.040 I mean absolutely, and I even say this about MCP service. 179 00:08:24.079 --> 00:08:27.240 They're not unlocking anything that new. They are fully limited 180 00:08:27.319 --> 00:08:30.600 by the capabilities of the underlying API. Notably this is 181 00:08:30.759 --> 00:08:34.360 MCP for API interactions. You can use MCP to wrap 182 00:08:34.440 --> 00:08:36.600 SEQL calls to a local database, that sort of thing. 183 00:08:37.320 --> 00:08:39.480 But yeah, and one of the big problems you see 184 00:08:39.519 --> 00:08:42.279 in the integration space and have seen for years, is 185 00:08:42.320 --> 00:08:44.960 that you often have to pull full data sets. And 186 00:08:45.000 --> 00:08:47.480 the reason you have to pull full data sets is 187 00:08:47.639 --> 00:08:49.919 that you don't have good search endpoints in these APIs, 188 00:08:50.039 --> 00:08:52.759 So I would say, I would say most inefficient integrations 189 00:08:52.799 --> 00:08:56.679 are built by bad APIs, not necessarily bad consumption patterns 190 00:08:56.879 --> 00:08:57.559 or anything else. 191 00:08:57.759 --> 00:09:00.200 Oh they're going to say bad engineers, and that I would. 192 00:09:01.200 --> 00:09:06.080 I mean, yeah, hopefully that one's not going to be 193 00:09:06.279 --> 00:09:09.200 The engineering skill level won't matter as much anymore. We're 194 00:09:09.200 --> 00:09:10.600 working on a new product right now, and I can 195 00:09:10.639 --> 00:09:12.919 tell you from using AI to build it, the skill 196 00:09:12.919 --> 00:09:14.600 gap is closing really really fast. 197 00:09:14.639 --> 00:09:16.720 It is so good now interesting. 198 00:09:16.799 --> 00:09:21.879 So so you think the vibe coding is leveling up? 199 00:09:22.120 --> 00:09:23.919 Oh yeah, Oh yeah, it's scary to say. 200 00:09:24.039 --> 00:09:26.399 Look, I'm an engineer who's I've been coding for for 201 00:09:26.440 --> 00:09:28.159 almost actually over twenty years now. 202 00:09:28.440 --> 00:09:28.960 I love it. 203 00:09:29.000 --> 00:09:31.720 I'm so passionate about building, and it's scary to see 204 00:09:31.720 --> 00:09:34.159 it just writing code that you would have written. But 205 00:09:34.200 --> 00:09:37.000 I also think it's really leveling me up too. We 206 00:09:37.200 --> 00:09:39.879 vibe coded, so me, a product manager, and a software 207 00:09:39.919 --> 00:09:42.840 engineer are building this new product. Our product manager, who 208 00:09:42.879 --> 00:09:45.000 has never written a line of code in his life, 209 00:09:45.240 --> 00:09:47.960 built our entire front end for it. We imported that 210 00:09:48.000 --> 00:09:50.559 into GitHub. Then we took over with Windsurf and started 211 00:09:50.960 --> 00:09:53.919 started you know, actually doing some a little bit more 212 00:09:53.919 --> 00:09:55.679 guided AI coding on top of it. 213 00:09:56.120 --> 00:09:58.159 But it worked and it was great, and then we 214 00:09:58.159 --> 00:10:00.440 just connected that straight to the back end right on. 215 00:10:00.600 --> 00:10:03.360 Yeah, I think that's been my experience. Like it's really 216 00:10:03.600 --> 00:10:08.440 easy to get started with AI, but then I think 217 00:10:08.679 --> 00:10:12.799 after like those initial few steps, I think it's still 218 00:10:12.799 --> 00:10:16.720 important to have the technical skills. And I treat AI, 219 00:10:17.440 --> 00:10:19.360 and the way I try to get people to treat 220 00:10:19.360 --> 00:10:22.759 AI is think of it as if you had your 221 00:10:22.799 --> 00:10:26.480 own intern or your own junior engineer, you know, like, 222 00:10:26.600 --> 00:10:30.919 don't treat the AI as a principal architect. Treat it 223 00:10:30.960 --> 00:10:35.960 as an intern, and give it very small scoped tasks 224 00:10:36.039 --> 00:10:38.879 that you can check up on, because it's it does 225 00:10:38.919 --> 00:10:42.440 sometimes get things wrong and just like your intern would, 226 00:10:42.480 --> 00:10:44.960 so you've got to give it a task that you 227 00:10:45.000 --> 00:10:48.120 can follow up and make sure that it's continuing to 228 00:10:48.200 --> 00:10:51.200 build towards the same end goal that you are. And 229 00:10:51.240 --> 00:10:53.159 I think that's where a lot of projects get off 230 00:10:53.200 --> 00:10:56.000 the rails, as they just give AI like this vague 231 00:10:56.120 --> 00:11:01.279 task without the guardrails to keep it from wandering off 232 00:11:01.279 --> 00:11:02.080 and hallucinating. 233 00:11:02.679 --> 00:11:05.480 I want to probe you on that well, definitely, because 234 00:11:05.759 --> 00:11:09.200 I see a lot of companies are not hiring interns 235 00:11:09.679 --> 00:11:11.919 and have no idea what to do with some interns, 236 00:11:11.919 --> 00:11:16.480 And yet they're coming up and hiring llms out there 237 00:11:16.519 --> 00:11:19.639 in the world to interact with, and I don't have 238 00:11:19.720 --> 00:11:22.759 the faith that they are capable of understanding how to 239 00:11:22.799 --> 00:11:26.320 provide that additional contact. So are you optimistic about where 240 00:11:26.639 --> 00:11:28.519 the quality of software in the world is going? 241 00:11:29.279 --> 00:11:33.159 I think so, yeah, because like on a large enough 242 00:11:33.200 --> 00:11:35.879 time frame, like this is going to work. 243 00:11:35.759 --> 00:11:37.080 Itself out, you know. 244 00:11:37.159 --> 00:11:39.320 Yeah, the death of the universe right around the corner. 245 00:11:39.679 --> 00:11:42.399 Yeah, absolutely, Like in the big scale of things, it 246 00:11:42.399 --> 00:11:46.200 doesn't really matter. But no, like specifically to your question, 247 00:11:46.279 --> 00:11:51.440 in people like engineers not having the skills to guide 248 00:11:51.480 --> 00:11:54.559 an intern, I would agree with that. I would include 249 00:11:54.559 --> 00:11:58.559 myself in that bucket as well as something I've had 250 00:11:58.600 --> 00:12:04.200 to learn and improve on. And I think that's where 251 00:12:05.200 --> 00:12:08.679 I think that's the skill gap that AI doesn't cover. 252 00:12:08.879 --> 00:12:13.360 So AI can write the technical code, but as an engineer, 253 00:12:15.399 --> 00:12:20.639 your value add into that equation is maintaining the big 254 00:12:20.679 --> 00:12:25.039 picture and breaking that down into concrete, isolated tasks. It 255 00:12:25.080 --> 00:12:31.320 can be distributed to junior engineers, interns, AIS, senior engineers, 256 00:12:31.559 --> 00:12:33.559 whatever is on your team. 257 00:12:34.039 --> 00:12:34.879 I totally agree with that. 258 00:12:34.919 --> 00:12:36.480 And I think we see this a lot. We see 259 00:12:36.480 --> 00:12:38.799 a fear of adopting AI from a lot of engineers 260 00:12:38.919 --> 00:12:41.200 or and it's not necessarily a fear of you know, oh, 261 00:12:41.200 --> 00:12:42.799 it's going to take my job, But I think it's 262 00:12:42.840 --> 00:12:45.960 a fear of it building bad software, or it just 263 00:12:46.120 --> 00:12:48.759 changing the way that someone is used to building in general. 264 00:12:49.519 --> 00:12:51.080 And I think the way that I explain this to 265 00:12:51.120 --> 00:12:53.600 my team is, you know, yeah, you're frustrated because it 266 00:12:53.679 --> 00:12:56.919 generated bad code number one, that's time for introspection. 267 00:12:57.039 --> 00:13:00.360 What did I do wrong? How do I prompt better? Yeah? 268 00:13:00.480 --> 00:13:03.120 But also, you know, I think that a lot of 269 00:13:03.120 --> 00:13:05.000 people see this as like, Okay, it's spent out really 270 00:13:05.120 --> 00:13:07.639 terrible code, I'm going to waste so much time cleaning 271 00:13:07.720 --> 00:13:09.759 it up. But what I'm not thinking about it's like, Okay, 272 00:13:09.840 --> 00:13:12.279 what if that eight hours buildings by hand? Or I 273 00:13:12.679 --> 00:13:15.320 can spend thirty minutes prompting it to create something, and 274 00:13:15.320 --> 00:13:17.519 then thirty minutes to an hour cleaning up the code, 275 00:13:17.559 --> 00:13:19.279 and then I've spent you know, a fourth of the 276 00:13:19.360 --> 00:13:22.080 time just maybe not how I'm used to building. 277 00:13:23.440 --> 00:13:23.799 Yeah. 278 00:13:23.840 --> 00:13:26.399 One of the big takeaways I've had from working with 279 00:13:26.519 --> 00:13:31.080 AI has been applying that to the rest of my life. 280 00:13:31.519 --> 00:13:34.519 Like I give AI a bad prompt, it writes bad code, 281 00:13:34.559 --> 00:13:37.039 and I spend a bunch of time cleaning it up 282 00:13:37.120 --> 00:13:40.200 and then thinking about what should I have said differently? 283 00:13:40.799 --> 00:13:44.600 And then I started looking at like other conversations in 284 00:13:44.639 --> 00:13:47.159 my life, and when I'm talking to humans now, I 285 00:13:47.279 --> 00:13:50.120 use that same process, and I realize, like, a lot 286 00:13:50.159 --> 00:13:52.399 of the pain and suffering I've had in my life 287 00:13:52.440 --> 00:13:55.000 is because I gave another human a bad prompt. 288 00:13:55.799 --> 00:14:02.399 Yeah, wow, you have relationships built on lies. Right, It's 289 00:14:02.440 --> 00:14:04.120 like I need to go back to this first conversation 290 00:14:04.159 --> 00:14:07.120 I ever had with this person and maybe change what 291 00:14:07.200 --> 00:14:09.600 I said to them, because that that's that set me 292 00:14:09.679 --> 00:14:11.440 up for a success or failure. 293 00:14:12.240 --> 00:14:12.440 Right. 294 00:14:13.240 --> 00:14:17.200 Yeah, It's like this whole chain of events that happened 295 00:14:17.240 --> 00:14:20.919 in my life could have been corrected had I given 296 00:14:20.919 --> 00:14:23.240 that person the right prompt to begin with. 297 00:14:24.440 --> 00:14:26.799 You very well may have just changed my dating life forever. 298 00:14:26.840 --> 00:14:30.639 It wasn't expected outcome of this call, but it really hasn't. 299 00:14:30.639 --> 00:14:35.559 It just become like you just sick one agentic agent 300 00:14:35.879 --> 00:14:38.399 at some dating service, and it will talk to another 301 00:14:38.759 --> 00:14:41.440 LM out there and then they'll decide collectively whether or 302 00:14:41.519 --> 00:14:42.639 not to start your relationship. 303 00:14:43.440 --> 00:14:45.639 Yeah, I mean it's once once Tender release is their 304 00:14:45.639 --> 00:14:46.720 official MCP server. 305 00:14:46.759 --> 00:14:47.399 We're going. I'm in. 306 00:14:48.639 --> 00:14:51.120 I mean, I assume people are doing this. They're just 307 00:14:51.120 --> 00:14:54.759 scraping the app and you know, uploading data anyway. I mean, 308 00:14:55.000 --> 00:14:56.679 or it's a human in the loop still, like it's 309 00:14:56.679 --> 00:14:59.559 still telling you what to type. Uh, you're just you're 310 00:14:59.639 --> 00:15:02.159 just doing Yeah, Like I don't. I don't think automating 311 00:15:02.200 --> 00:15:03.639 it is going to mean maybe that's what we're missing. 312 00:15:03.679 --> 00:15:05.519 We're missing the automation so people can get on more 313 00:15:05.600 --> 00:15:06.279 dates faster. 314 00:15:07.279 --> 00:15:09.600 I've been I've been seeing the tiktoks because I guess 315 00:15:09.600 --> 00:15:12.440 the dating apps are going after people who are automating. 316 00:15:12.200 --> 00:15:14.120 Of some people who set up an actual phone in 317 00:15:14.120 --> 00:15:15.600 a room with a little rubber hand on. 318 00:15:15.559 --> 00:15:18.159 A screwdriver or on a on a drill that's just swiping. 319 00:15:20.519 --> 00:15:22.600 I mean, it's a it's a miss market for them. 320 00:15:22.600 --> 00:15:24.519 Like if all of your users are doing something we 321 00:15:24.600 --> 00:15:26.320 know from there's like some great books out there like 322 00:15:26.360 --> 00:15:29.399 Platform Revolution that you should actually, other than trying to 323 00:15:29.559 --> 00:15:32.120 prevent that behavior and punish people for it, realize that 324 00:15:32.240 --> 00:15:35.720 is where the value is being added. And like I 325 00:15:35.720 --> 00:15:38.679 think the swiping left or right is the action which 326 00:15:38.720 --> 00:15:41.759 makes people feel invested in the action they're taking so 327 00:15:41.799 --> 00:15:44.039 that they're more likely to continue. So, you know, give 328 00:15:44.039 --> 00:15:47.039 them that capability, but give them what they want, which 329 00:15:47.080 --> 00:15:50.080 is like maybe a multi select option or or you know, 330 00:15:50.120 --> 00:15:51.840 something like that, and to take it to the next level. 331 00:15:51.919 --> 00:15:54.480 Rather than banning those people, let them have that functionality 332 00:15:54.559 --> 00:15:57.200 like once a day and then you know, up charge 333 00:15:57.279 --> 00:16:00.000 for it for the you know, next one hundred matches. 334 00:16:01.200 --> 00:16:03.559 Yeah, the AI matchmaker, Yeah right. 335 00:16:03.440 --> 00:16:07.279 Exactly, well, yeah right, just be open with it, like say, okay, 336 00:16:07.399 --> 00:16:09.919 I see, I see what you're up for here, and 337 00:16:09.960 --> 00:16:12.039 I'm gonna make it easy for you. You just gotta 338 00:16:12.519 --> 00:16:14.960 upgrade your service level here, sign up for the next 339 00:16:14.960 --> 00:16:16.279 plan and I got you dude. 340 00:16:17.039 --> 00:16:18.240 Mm hm cool. 341 00:16:18.320 --> 00:16:22.960 So so back to MCP, it's it sounds like it's 342 00:16:23.200 --> 00:16:26.879 very agent focused. Is that an accurate statement? 343 00:16:27.639 --> 00:16:27.919 Yeah? 344 00:16:27.960 --> 00:16:30.360 So, so people still ask us this a lot, you know, 345 00:16:30.399 --> 00:16:32.200 when when do we want to use a traditional API 346 00:16:32.279 --> 00:16:34.519 versus MCP? And I think that the way to think 347 00:16:34.519 --> 00:16:37.080 about it is a lot of what we've built is 348 00:16:37.120 --> 00:16:38.720 still the best possible. 349 00:16:38.399 --> 00:16:39.120 UI or something. 350 00:16:39.200 --> 00:16:41.240 So if I want to cancel an order is it 351 00:16:41.279 --> 00:16:43.200 easier for me to go to Amazon, go to my 352 00:16:43.320 --> 00:16:45.360 orders and click cancel, or to open up a chat 353 00:16:45.399 --> 00:16:47.759 butt and say hi, the order for this, and it's like, okay, 354 00:16:47.759 --> 00:16:49.679 are you Are you referring to disorder and I'm like yes, 355 00:16:49.720 --> 00:16:52.240 And it's it's just not the best the interface to cancel, 356 00:16:52.279 --> 00:16:55.039 and so for that there's a button, and to have 357 00:16:55.120 --> 00:16:58.080 a button then prompt an agent and say hey, the 358 00:16:58.080 --> 00:17:00.360 customer would like to cancel disorder, and it then has 359 00:17:00.399 --> 00:17:02.200 to decide on a tool, and then it has to 360 00:17:02.200 --> 00:17:03.879 make sure it's calling the right tool and make that call. 361 00:17:04.200 --> 00:17:07.480 One it's just inefficient and really slow. But two, you're 362 00:17:07.519 --> 00:17:09.960 going from a world of a very deterministic action I 363 00:17:10.000 --> 00:17:13.160 click a button that runs this code that cancels this 364 00:17:13.359 --> 00:17:16.319 order to we think that the agent should be able 365 00:17:16.319 --> 00:17:18.720 to figure out the right tool to call, but it 366 00:17:18.799 --> 00:17:20.160 might not get it right every time. 367 00:17:20.599 --> 00:17:23.400 And again we're spending a lot of money on having 368 00:17:23.440 --> 00:17:24.920 the agent decide what to actually do. 369 00:17:25.000 --> 00:17:28.039 And so I think in a lot of cases, classic 370 00:17:28.079 --> 00:17:30.799 APIs are still going to be the really valuable one. 371 00:17:31.559 --> 00:17:36.079 MCP is for agentic interfaces, it's for bots, it's for communications. 372 00:17:36.599 --> 00:17:38.920 It could be a customer facing bot it could be 373 00:17:39.000 --> 00:17:41.640 an internal sorry by bottom referring to it could be 374 00:17:41.680 --> 00:17:44.480 a customer facing agent, it could be an internal agent, 375 00:17:45.559 --> 00:17:48.519 or it could be some form of non exposed agent 376 00:17:48.599 --> 00:17:50.160 that's actually just taking actions. 377 00:17:50.640 --> 00:17:53.240 I really like this take. I think it's really interesting 378 00:17:53.279 --> 00:17:56.359 and I want to repeat this. Basically, if the thing 379 00:17:56.400 --> 00:17:58.400 that could bring you value in your business or your 380 00:17:58.480 --> 00:18:03.640 API is allowing increased volume or speed for execution, this 381 00:18:03.680 --> 00:18:05.599 could be the right thing to do. But if you 382 00:18:05.599 --> 00:18:07.640 don't want people to cancel your orders, don't make it 383 00:18:07.680 --> 00:18:08.799 easier for them to do that. 384 00:18:09.160 --> 00:18:13.119 So that's not where it was going. 385 00:18:13.160 --> 00:18:16.000 But that's why that's what I no, but. 386 00:18:15.960 --> 00:18:16.559 I think I will. 387 00:18:17.240 --> 00:18:18.920 It's interesting that you chose Amazon though, Like you know, 388 00:18:18.920 --> 00:18:20.400 if you pick something and it has a really great 389 00:18:20.480 --> 00:18:24.119 user experience, then it doesn't necessarily make sense to automate that. 390 00:18:24.400 --> 00:18:26.960 But as we know out there, there's like every company 391 00:18:27.000 --> 00:18:30.359 sucks at UX. So you know, you think about that, 392 00:18:30.400 --> 00:18:32.000 do you want to invest in you you actually do 393 00:18:32.079 --> 00:18:35.160 want to just throw money at the problem and run 394 00:18:35.200 --> 00:18:39.319 a really expensive service somewhere to manage state for letting 395 00:18:39.359 --> 00:18:41.680 agents manage the user experience. 396 00:18:42.440 --> 00:18:43.160 Yeah, yeah, I. 397 00:18:43.200 --> 00:18:45.759 Mean ultimately they're going to that specific example they're going 398 00:18:45.839 --> 00:18:48.640 to do both anyway, But in general, the idea is 399 00:18:48.680 --> 00:18:52.200 like a static action that's button driven, classic API integration 400 00:18:52.480 --> 00:18:53.920 AGENTIC MCP. 401 00:18:54.440 --> 00:18:58.519