1 00:00:00,080 --> 00:00:03,960 Open AI might have just made the biggest strategic bet yet, and 2 00:00:03,960 --> 00:00:07,200 it's not making AI smarter. They're shifting from raw 3 00:00:07,200 --> 00:00:11,320 intelligence with new language model updates to moving up the 4 00:00:11,320 --> 00:00:14,880 application layer to the top of the stock to try and make AI 5 00:00:14,880 --> 00:00:17,720 more useful. Just this week, Open AI Dev Day 6 00:00:17,720 --> 00:00:20,720 happened, and it's one of the biggest events of the year for 7 00:00:20,720 --> 00:00:23,240 companies like Open AI, where they tried to sell their vision 8 00:00:23,240 --> 00:00:25,920 and their new product features to developers all over the 9 00:00:25,920 --> 00:00:27,560 world. And for those that have been 10 00:00:27,560 --> 00:00:30,360 listening in every single week, you'll really understand and how 11 00:00:30,360 --> 00:00:33,840 important it is for companies like Open AI to retain the 12 00:00:33,840 --> 00:00:36,640 developer ecosystem. It always gives us a really 13 00:00:36,640 --> 00:00:40,000 interesting snapshot into the future of Open AI strategy. 14 00:00:40,160 --> 00:00:42,800 In today's episode, we're going to go through three of the 15 00:00:42,800 --> 00:00:45,200 biggest feature releases announced during Dev day. 16 00:00:45,560 --> 00:00:48,160 Whether you should even care about it, If so, what you should 17 00:00:48,160 --> 00:00:51,240 know about it. And finally, what this is going 18 00:00:51,240 --> 00:00:54,680 to tell us about the future of AI in the next few years. 19 00:00:54,800 --> 00:00:56,240 This is in the Loop with Jack Horton. 20 00:00:56,240 --> 00:01:11,680 I hope you enjoy the show. So let's start today's episode 21 00:01:11,680 --> 00:01:15,280 with the App Store announcement from ChatGPT. 22 00:01:16,160 --> 00:01:18,920 So what they actually launched was essentially this new feature 23 00:01:18,920 --> 00:01:22,640 of apps and ChatGPT that enables you to bring third party 24 00:01:22,640 --> 00:01:26,360 applications from major companies like Spotify, Zillow, 25 00:01:26,360 --> 00:01:30,920 Canva, booking.com, Expedia, Figma directly inside the 26 00:01:30,920 --> 00:01:33,480 ChatGPT interface. Now I think what's funny here is 27 00:01:33,480 --> 00:01:37,360 that obviously they tried with GPT plug insurance that failed. 28 00:01:37,800 --> 00:01:41,160 They definitely tried with the GPT store and that failed as 29 00:01:41,160 --> 00:01:43,720 well. So this is their third attempt 30 00:01:43,800 --> 00:01:46,800 at a platform strategy and ecosystem strategy around App 31 00:01:46,800 --> 00:01:49,400 Store ideas. And it basically lets you use 32 00:01:49,400 --> 00:01:53,600 the applications from ChatGPT and get the benefit of chat GBTS 33 00:01:53,600 --> 00:01:56,320 intelligence and it's context it has on you and other 34 00:01:56,320 --> 00:02:00,200 applications plus the third party applications context and 35 00:02:00,440 --> 00:02:03,000 intelligence. And so the key innovation here 36 00:02:03,000 --> 00:02:06,560 has been the deep context integration that they're 37 00:02:06,560 --> 00:02:08,840 enabling or they're saying they're going to enable. 38 00:02:09,400 --> 00:02:14,960 So these aren't just, you know, let's say Expedia or Zillow 39 00:02:15,240 --> 00:02:19,240 embedded into ChatGPT. They actually talk to ChatGPT 40 00:02:19,240 --> 00:02:21,120 and share information back and forth. 41 00:02:21,320 --> 00:02:24,000 So what they launched was an app's SDK that enables 42 00:02:24,000 --> 00:02:26,680 developers to build this into ChatGPT. 43 00:02:27,000 --> 00:02:29,560 And it's all built on top of model context protocols. 44 00:02:29,560 --> 00:02:33,480 So MCPS, which Open AI has aligned with, is an open 45 00:02:33,480 --> 00:02:35,520 standard just like pretty much everyone in the industry right 46 00:02:35,520 --> 00:02:36,960 now. They've launched an app 47 00:02:36,960 --> 00:02:39,880 directory and apparently going to be seen bringing out 48 00:02:39,880 --> 00:02:43,440 monetization features, including the Agentic Commerce protocol 49 00:02:43,720 --> 00:02:46,920 for direct in chat purchases. I might cover this in a future 50 00:02:46,920 --> 00:02:50,200 episode, but it's another one of those model context protocols or 51 00:02:50,200 --> 00:02:53,880 agent agent protocols that are emerging, but this time to 52 00:02:53,880 --> 00:02:58,520 standardise how AI systems can purchase goods via the likes of 53 00:02:58,520 --> 00:03:00,320 a ChatGPT. And this all makes sense. 54 00:03:00,320 --> 00:03:05,240 They've got huge distribution for 3rd party businesses and app 55 00:03:05,240 --> 00:03:08,920 builders and things like that. ChatGPT has 800 million weekly 56 00:03:08,920 --> 00:03:11,880 active users. The fact that that's weekly is 57 00:03:12,120 --> 00:03:14,480 ridiculous. And they've got 4 million 58 00:03:14,480 --> 00:03:17,440 developers building on Open AI's platform of some sorts. 59 00:03:18,120 --> 00:03:20,560 And just a caveat to this, although I say building on top 60 00:03:20,560 --> 00:03:24,800 of using their API, so they'll be using that API, many other 61 00:03:24,800 --> 00:03:29,120 providers, but just through the API, the processing about 6 62 00:03:29,120 --> 00:03:32,160 billion tokens through their APIs per minute. 63 00:03:32,160 --> 00:03:35,160 To put that in perspective, that's more weekly users and 64 00:03:35,160 --> 00:03:37,760 usage than TikTok ever had. And what I think is really 65 00:03:37,760 --> 00:03:40,880 important to understand about this is it really showcases the 66 00:03:40,880 --> 00:03:43,960 future of all software interactions because every 67 00:03:43,960 --> 00:03:48,120 application in the world, whether you're AB to C consumer 68 00:03:48,120 --> 00:03:52,640 product or your B to B SAS product or your fintech product, 69 00:03:52,640 --> 00:03:55,920 biotech product, whatever you do, it is going to be 70 00:03:55,920 --> 00:03:59,080 conversational. This is exactly what we do with 71 00:03:59,080 --> 00:04:01,160 our customers. We work with B to B SAS 72 00:04:01,160 --> 00:04:04,840 providers and product management and developers and designers to 73 00:04:05,240 --> 00:04:09,400 enable them to quickly turn all of their technology into a 74 00:04:09,400 --> 00:04:13,320 conversational interface that controls all the features, gets 75 00:04:13,320 --> 00:04:16,240 context about users, takes actions in other applications. 76 00:04:16,839 --> 00:04:19,839 And we see this happening in real time. 77 00:04:19,839 --> 00:04:22,960 And I think this is just another major proof point along that 78 00:04:22,960 --> 00:04:24,520 journey. And think about it, in many 79 00:04:24,520 --> 00:04:28,240 instances or use cases, you don't want to have to visit 10 80 00:04:28,240 --> 00:04:30,680 different application interfaces just to get something done. 81 00:04:30,680 --> 00:04:32,360 And the demos were really cool too. 82 00:04:32,520 --> 00:04:35,560 I have to say, in one, someone logged into Cosia through 83 00:04:35,560 --> 00:04:38,840 ChatGPT and was able to watch educational content and courses. 84 00:04:38,920 --> 00:04:40,880 And they did the same with Zillow, which for those in the 85 00:04:40,880 --> 00:04:45,480 UK is the platform for people to buy and rent and sell houses. 86 00:04:46,000 --> 00:04:47,240 And they also did it with Spotify. 87 00:04:47,440 --> 00:04:50,200 And this follows really a pattern we've seen many times 88 00:04:50,200 --> 00:04:53,440 before with big, big platform providers, you know, Facebook, 89 00:04:53,440 --> 00:04:57,120 Google, Apple, iOS. And really what we can see is 90 00:04:57,120 --> 00:04:59,640 that we're in step 0 right now, which is crazy competition 91 00:04:59,640 --> 00:05:03,400 between, you know, 5 to 7 to 8 providers in the world with no 92 00:05:03,400 --> 00:05:05,520 real winning yet. Opening eyes obviously 93 00:05:05,520 --> 00:05:07,640 identified them context to memory. 94 00:05:07,640 --> 00:05:11,240 And they're trying to obviously press that advantage by opening 95 00:05:11,240 --> 00:05:15,280 up a third party platform through the app SDK. 96 00:05:15,280 --> 00:05:17,400 But execute solution is going to be really important here. 97 00:05:18,160 --> 00:05:21,400 You know, back in the day of Apple building, you know, iTunes 98 00:05:21,400 --> 00:05:24,160 and platforms like that, there was many, many companies trying 99 00:05:24,160 --> 00:05:27,120 to do the same, Even those that should have done a really, 100 00:05:27,120 --> 00:05:31,240 really good job but failed because it felt just wrong. 101 00:05:31,240 --> 00:05:33,560 It didn't feel natural. They never nailed it. 102 00:05:33,680 --> 00:05:36,560 For example, Sony were trying to build the exact same all the 103 00:05:36,560 --> 00:05:39,240 music producer companies in the world were trying to produce 104 00:05:39,240 --> 00:05:42,560 their own version of iTunes, but they all did it just wrong. 105 00:05:43,560 --> 00:05:46,680 And I felt that sometimes with chat GB TS demos, it felt a bit 106 00:05:46,680 --> 00:05:50,120 gimmicky. In the demo, they asked you to 107 00:05:50,120 --> 00:05:53,040 design a company logo through a chat interface, speaking to 108 00:05:53,040 --> 00:05:55,960 Camber and then make a presentation deck. 109 00:05:56,280 --> 00:05:59,720 And I can see it with the deck maybe, but it just felt like, I 110 00:05:59,720 --> 00:06:04,520 guess weird, you know, I wouldn't go to ChatGPT to design 111 00:06:04,720 --> 00:06:08,480 A logo through Camber. I would just go to Camber for 112 00:06:08,480 --> 00:06:11,360 some of those tasks. I think that really speaks to 113 00:06:11,560 --> 00:06:14,440 how people in the world should see this, which is that for many 114 00:06:14,440 --> 00:06:16,720 use cases, it's perfect for buying and finding and 115 00:06:16,720 --> 00:06:19,360 researching houses. But for many use cases, you just 116 00:06:19,360 --> 00:06:23,120 want to be in deep work mode and really engage with the platform. 117 00:06:23,840 --> 00:06:28,120 Now that might change over time. Chat GB TS technology might 118 00:06:28,120 --> 00:06:31,640 evolve as to a point where you don't need to visit third party 119 00:06:31,640 --> 00:06:33,640 providers. But what I think this speaks to 120 00:06:33,920 --> 00:06:37,920 really is open a eyes bid to capture just consuming fact all 121 00:06:37,920 --> 00:06:41,120 user context possible. So once users, you know, 122 00:06:41,480 --> 00:06:45,280 experience Koscia with ChatGPT as their personal tutor, you 123 00:06:45,280 --> 00:06:47,520 don't want to go necessary back to this standard and experience. 124 00:06:47,520 --> 00:06:50,440 And ChatGPT has all the context of everything. 125 00:06:50,440 --> 00:06:53,400 So the conversation running up to that, the interactions that 126 00:06:53,400 --> 00:06:56,480 happened during that and after that, it's how Open AI could 127 00:06:56,480 --> 00:07:01,040 become the everyday assistant by owning all interaction context. 128 00:07:12,080 --> 00:07:14,840 I think this is a worthy thing to dig into and deep dive into a 129 00:07:14,840 --> 00:07:18,560 little bit more with this idea of context and opening eye 130 00:07:18,560 --> 00:07:20,120 trying to capture all that context. 131 00:07:20,280 --> 00:07:23,040 Because apps running in ChatGPT will access all the 132 00:07:23,040 --> 00:07:25,280 conversational context, they'll share information back and 133 00:07:25,280 --> 00:07:27,080 forth. They'll integrate with Chat GBTS 134 00:07:27,080 --> 00:07:30,240 memory feature for personalized experience across different 135 00:07:30,240 --> 00:07:32,640 applications. Open IO calls is talking to apps 136 00:07:32,640 --> 00:07:35,360 functionality from ChatGPT. The first time you'll see one of 137 00:07:35,360 --> 00:07:38,560 these applications, ChatGPT will show a privacy policy and what 138 00:07:38,560 --> 00:07:40,960 data is likely to be shared. Just like Google does. 139 00:07:41,040 --> 00:07:43,480 Apps will pull contacts from your chat history and provide 140 00:07:43,480 --> 00:07:46,240 personalized responses based on previous conversations. 141 00:07:46,600 --> 00:07:49,760 And what we're really seeing here is them trying to be that 142 00:07:49,760 --> 00:07:52,120 everyday system. But it raises really big 143 00:07:52,120 --> 00:07:54,960 questions because, you know, some of this makes perfect sense 144 00:07:54,960 --> 00:07:59,160 using Zillow to look at houses, then using GPT to, I don't know, 145 00:07:59,160 --> 00:08:03,360 look for data such as how close is this to a good park or 146 00:08:03,360 --> 00:08:05,680 school? Has there been any good 147 00:08:05,680 --> 00:08:09,560 restaurants opening soon? The IRS, all that context from 148 00:08:09,560 --> 00:08:12,840 your house browsing and can provide very relevant answers. 149 00:08:13,040 --> 00:08:15,360 And I think once users experience this level of 150 00:08:15,360 --> 00:08:18,080 contextual help, going back is going to be quite tough. 151 00:08:18,200 --> 00:08:21,080 This is how open AI probably will start to really stress out 152 00:08:21,080 --> 00:08:23,800 the likes of Google because although Google is way out in 153 00:08:23,800 --> 00:08:26,920 front, things could change quite quickly over time. 154 00:08:27,120 --> 00:08:29,520 And as I said, at Mindset, we see this all the time. 155 00:08:29,800 --> 00:08:31,600 We've just launched a memory feature in beta. 156 00:08:32,159 --> 00:08:34,919 Our customers are turning their applications into essentially 157 00:08:35,159 --> 00:08:38,440 lots of different agents that feel like a single conversation. 158 00:08:38,440 --> 00:08:42,000 And as a user, you can navigate their applications through 159 00:08:42,000 --> 00:08:44,720 simple conversations and then hop into taking actions across 160 00:08:44,720 --> 00:08:46,960 other applications. And so it's really interesting 161 00:08:46,960 --> 00:08:51,440 how the type of technology and work we're doing alongside some 162 00:08:51,480 --> 00:08:55,880 incredible companies will actually start to be further 163 00:08:55,880 --> 00:08:58,480 developed and improved by the likes of Chat Chi BT bringing 164 00:08:58,480 --> 00:09:00,920 out this contextual awareness feature. 165 00:09:01,400 --> 00:09:03,920 And I, I assume most providers will start to do the same. 166 00:09:04,400 --> 00:09:07,520 So when we combine these two worlds of our customers, 167 00:09:07,520 --> 00:09:11,040 applications becoming conversational with these new 168 00:09:11,040 --> 00:09:14,080 features by Chat Chi BT and soon all of the providers, no doubt 169 00:09:14,920 --> 00:09:17,480 technology will increasingly just feel quite human. 170 00:09:17,640 --> 00:09:20,080 Now, personally, I don't like the idea of opening and having 171 00:09:20,160 --> 00:09:23,080 all this context about me. I'm not a big fan about this 172 00:09:23,080 --> 00:09:25,600 idea. I'm not a big fan of my 173 00:09:25,600 --> 00:09:30,320 conversation history suddenly being used as part of web 174 00:09:30,320 --> 00:09:33,200 searching or the equivalent of modern day advertising as it 175 00:09:33,200 --> 00:09:35,720 probably will emerge in the next couple of years. 176 00:09:35,880 --> 00:09:38,440 Because the more details they have on me, the more they'll try 177 00:09:38,440 --> 00:09:40,600 to sell to me. For some, they live there. 178 00:09:40,600 --> 00:09:43,760 I, I always have my tin foil hat on with these things because 179 00:09:43,760 --> 00:09:46,800 really we never got social media right many years ago. 180 00:09:46,800 --> 00:09:50,440 And I think many of us regret the decisions or the lack of 181 00:09:50,440 --> 00:09:53,320 decisions, shall I say, that were made during that time on 182 00:09:53,320 --> 00:09:55,880 regulation, data protection and privacy. 183 00:10:07,340 --> 00:10:10,660 Anyway, going to the third part of what I think is really 184 00:10:10,660 --> 00:10:13,540 interesting here with the Dev Day release, which is agent 185 00:10:13,540 --> 00:10:16,440 care. So Agent Kit is essentially an 186 00:10:16,440 --> 00:10:18,360 agent and workflow builder for AI agents. 187 00:10:18,360 --> 00:10:22,360 It included a visual agent workflow builder with drag and 188 00:10:22,360 --> 00:10:24,840 drop workflows. I think like NAN Orchestration, 189 00:10:24,960 --> 00:10:28,320 there was a chat box interface that you could put your 190 00:10:28,320 --> 00:10:31,920 workflows into to test and play and access those workflows. 191 00:10:32,440 --> 00:10:35,480 It came with an evaluation platform with trace grading, 192 00:10:35,840 --> 00:10:39,800 which can essentially score multi step agent decisions and 193 00:10:39,840 --> 00:10:42,680 ensure that it's producing relevant and high quality 194 00:10:42,680 --> 00:10:44,600 outputs. They've got a connector registry 195 00:10:44,600 --> 00:10:48,280 which is their objective of trying to handle enterprise 196 00:10:48,280 --> 00:10:51,520 integrations, and they're also coming out the box with support 197 00:10:51,520 --> 00:10:55,600 for MCP, which is fantastic. And they're teasing the release 198 00:10:55,600 --> 00:10:59,240 of a reinforcement fine tuning feature for specialised agents, 199 00:10:59,520 --> 00:11:02,600 so really being able to make them hyper specialised at some 200 00:11:02,600 --> 00:11:04,160 use cases. I really like some of their 201 00:11:04,160 --> 00:11:06,760 approaches here. It's very similar to NAN, to 202 00:11:06,760 --> 00:11:09,960 Zapier, to Lindy AI. I'm yet to use the work for a 203 00:11:09,960 --> 00:11:13,120 builder from Open AI, but I've used all the other main 204 00:11:13,120 --> 00:11:16,240 providers that are basically the same as that Open AI set of 205 00:11:16,240 --> 00:11:18,840 features and it'll be really interesting. 206 00:11:19,000 --> 00:11:21,040 It reinforces something we've been saying for a long time at 207 00:11:21,040 --> 00:11:23,680 Mindset, which is that anything to do, building a genetic 208 00:11:23,680 --> 00:11:28,120 experiences must happen in configuration and not code. 209 00:11:28,160 --> 00:11:30,640 There are many cases that you want to build in code, that's 210 00:11:30,640 --> 00:11:33,080 fine. But really all an agent and a 211 00:11:33,080 --> 00:11:36,000 workflow is, is a series of different prompts using 212 00:11:36,000 --> 00:11:40,440 different tools and context of information at the right moment 213 00:11:40,640 --> 00:11:44,400 and prompts being sent to the model to give the model context 214 00:11:44,400 --> 00:11:47,440 or hyper specific context in that moment over a series of 215 00:11:47,440 --> 00:11:49,040 steps. That's all that's happening. 216 00:11:49,200 --> 00:11:52,520 All the magic happens in the language model layer, and we've 217 00:11:52,520 --> 00:11:54,440 got a workflow builder and an agent builder. 218 00:11:54,680 --> 00:12:00,680 So it's really good to see Open AI further prove our belief that 219 00:12:01,000 --> 00:12:04,400 building agents should happen in configuration and not code. 220 00:12:05,480 --> 00:12:07,840 But one thing that I would say here with this launch is that 221 00:12:07,840 --> 00:12:10,440 it's extremely myopic and it surprises me a little bit. 222 00:12:10,440 --> 00:12:12,800 To be honest, I thought that it'd been a little bit more 223 00:12:12,800 --> 00:12:17,600 expansive because everything is just Open AI ecosystem only. 224 00:12:17,800 --> 00:12:21,240 You can only use their models, their evaluation system, their 225 00:12:21,240 --> 00:12:23,520 deployment methods. It's all locked into their 226 00:12:23,560 --> 00:12:25,520 ecosystem. And although they're using some 227 00:12:25,520 --> 00:12:30,880 open standards such as MCP, their proposition really is once 228 00:12:30,880 --> 00:12:33,880 you're here in the Open AI ecosystem, you can't really 229 00:12:33,880 --> 00:12:35,800 leave. And that makes sense given their 230 00:12:35,800 --> 00:12:39,040 clear strategy of making ChatGPT into that next major 231 00:12:39,040 --> 00:12:41,240 distribution platform, just like Apple did. 232 00:12:41,680 --> 00:12:45,600 And it makes sense because Open AI is soon to launch their own 233 00:12:45,600 --> 00:12:48,520 product features with Johnny I've so the legendary designer 234 00:12:48,520 --> 00:12:51,840 behind most of the best Apple products ever released. 235 00:12:51,960 --> 00:12:54,800 All this further does is try and lock people in when people don't 236 00:12:54,800 --> 00:12:58,240 want that from a developer perspective, from a technology 237 00:12:58,240 --> 00:13:00,560 and vendor perspective. So I think it's going to be 238 00:13:00,560 --> 00:13:04,200 fantastic for internal use cases as a result, but not so much for 239 00:13:04,200 --> 00:13:07,360 those that want to be able to use many best of breed systems. 240 00:13:08,520 --> 00:13:11,480 It also creates this huge platform dependency problem 241 00:13:12,240 --> 00:13:15,320 because what happens then is these platforms open everybody 242 00:13:15,320 --> 00:13:20,200 up to cool exciting things to attract customers and developers 243 00:13:20,200 --> 00:13:22,600 with great distribution and new features and tools. 244 00:13:22,920 --> 00:13:24,840 But they just gradually close things down. 245 00:13:24,840 --> 00:13:28,200 Whether that's reducing organic reach on social media, whether 246 00:13:28,200 --> 00:13:30,680 it's increasing higher and higher revenue cuts like the App 247 00:13:30,680 --> 00:13:34,080 Store or just building competing first party products like 248 00:13:34,080 --> 00:13:37,240 Facebook did with the games. And it minds that almost all of 249 00:13:37,240 --> 00:13:39,920 our customers want to be able to use best of breed systems. 250 00:13:40,080 --> 00:13:43,320 They don't want to be locked in. Nobody knows what the next 10 251 00:13:43,320 --> 00:13:46,320 years will look like. So you want to be able to build 252 00:13:46,560 --> 00:13:49,640 and create value over time and you not have to RIP something 253 00:13:49,640 --> 00:13:51,440 out because you're locked into an approach. 254 00:14:02,840 --> 00:14:05,320 Anyway, let's let's conclude and summarize, I guess what we've 255 00:14:05,320 --> 00:14:08,240 been talking about. I think really there's been a 256 00:14:08,240 --> 00:14:11,080 bunch of nuclear features from the App Store, agent builder, 257 00:14:11,080 --> 00:14:14,360 workflow builder, new contacts being shared to companies that 258 00:14:14,560 --> 00:14:16,840 build on ChatGPT. And what we're really seeing 259 00:14:16,840 --> 00:14:22,040 here is an operating system war. Open AI is trying to own all 260 00:14:22,040 --> 00:14:24,440 interactions between humans and systems. 261 00:14:25,040 --> 00:14:28,440 And no doubt Google or Microsoft and Apple are going to respond, 262 00:14:28,640 --> 00:14:31,640 especially given that I've heard that Apple is looking to replace 263 00:14:31,640 --> 00:14:34,480 Tim Cook as CEO. So we're looking at this big 264 00:14:34,480 --> 00:14:37,760 platform battle that will shape how we interact with technology 265 00:14:37,760 --> 00:14:41,760 over the next few years. And I, for one, I'm really 266 00:14:41,880 --> 00:14:43,880 interested in how this all plays out. 267 00:14:44,160 --> 00:14:47,800 And I think what I've really been pleased with here is that 268 00:14:48,240 --> 00:14:52,640 this dev day by Open AI will only further open the world up 269 00:14:52,640 --> 00:14:56,320 to building agentic technology and systems and testing it out 270 00:14:56,320 --> 00:14:58,640 for themselves. Anyway, that's it for this week. 271 00:14:58,640 --> 00:15:01,200 I hope you enjoyed the show and I'll see you next week.