AI Means “Knowledge Work Is Cooked”, How Should Leaders Respond | Mark Rotheram, Rowan Gill

1 00:00:00,800 - > 00:00:04,000 Mark Rotheram: So it went from zero to a hundred thousand 2 00:00:04,000 - > 00:00:07,200 GitHub stars, which basically means it's the fastest growing 3 00:00:07,200 - > 00:00:08,720 open source project in history. 4 00:00:08,720 - > 00:00:11,119 So huge, huge adoption. 5 00:00:11,119 - > 00:00:15,519 And yeah, it can pretty much do anything you let it, which is 6 00:00:15,519 - > 00:00:16,399 really interesting. 7 00:00:16,879 - > 00:00:19,920 Sinéad Hammond: Interesting, exciting, terrifying, I think is 8 00:00:19,920 - > 00:00:21,519 probably the words I'd describe that. 9 00:00:21,760 - > 00:00:24,000 Rowan Gill: Things are changing so rapidly, as Mark has 10 00:00:24,000 - > 00:00:24,719 highlighted. 11 00:00:24,719 - > 00:00:28,559 It is really difficult to keep up, and it's having an open 12 00:00:28,559 - > 00:00:31,519 mind, being aware of what's happening, but then really 13 00:00:31,519 - > 00:00:34,719 bringing it back down to evaluate what will make the big 14 00:00:34,719 - > 00:00:36,560 difference and is it safe. 15 00:00:39,600 - > 00:00:41,920 Sinéad Hammond: Welcome to the AI for business podcast with 16 00:00:41,920 - > 00:00:44,960 BCN, where we put AI and automation at the centre of 17 00:00:44,960 - > 00:00:45,920 every episode. 18 00:00:45,920 - > 00:00:49,200 This is a regular debrief for business leaders to digest the 19 00:00:49,200 - > 00:00:52,159 business news and the biggest stories and development in AI as 20 00:00:52,159 - > 00:00:53,200 they hit the headlines. 21 00:00:53,200 - > 00:00:55,840 And when it feels like there's a new story and technology 22 00:00:55,840 - > 00:00:58,799 pretty much daily, we're cutting through all of the noise and 23 00:00:58,799 - > 00:01:01,280 we're giving you a rundown of what's important to businesses, 24 00:01:01,280 - > 00:01:04,159 why it matters, and what you can do to keep up with the pace. 25 00:01:04,159 - > 00:01:07,920 So I'm your host, Sinéad Hammond, and today I'm joined by 26 00:01:07,920 - > 00:01:12,640 CTO of BCN, Mark Rotherham, and Head of Marketing at BCN, Rowan 27 00:01:12,640 - > 00:01:13,040 Gill. 28 00:01:13,040 - > 00:01:14,560 Thank you both for joining us today. 29 00:01:14,879 - > 00:01:16,560 Rowan Gill: Looking forward to it. 30 00:01:17,040 - > 00:01:17,040 Mark Rotheram: Thank you 31 00:01:17,920 - > 00:01:20,079 Sinéad Hammond: The first thing we want to cover is the 32 00:01:20,079 - > 00:01:23,120 headlines that are new in the news for AI. 33 00:01:23,120 - > 00:01:24,319 So let's get into those. 34 00:01:24,319 - > 00:01:27,680 Mark, if you had to pick three developments this month that UK 35 00:01:27,680 - > 00:01:30,480 business leaders should actually have on our radar, what do you 36 00:01:30,480 - > 00:01:31,519 think they should be? 37 00:01:31,920 - > 00:01:34,159 Mark Rotheram: Yeah, first thing we should definitely cover off 38 00:01:34,159 - > 00:01:37,120 is all the recent stuff coming out around knowledge work. 39 00:01:37,120 - > 00:01:40,319 You know, the phrase knowledge work is cooked is definitely 40 00:01:40,319 - > 00:01:42,159 something we should be digging into. 41 00:01:42,159 - > 00:01:45,840 We couldn't go ahead today without talking about um 42 00:01:45,840 - > 00:01:49,920 OpenClaw and permissionless AI and what's going on there and 43 00:01:49,920 - > 00:01:53,840 the whole ecosystem that sprung up in the last few weeks. 44 00:01:53,840 - > 00:01:57,120 And then it's probably worth touching on things like the 45 00:01:57,120 - > 00:02:01,120 latest AI models and what it means and how the different 46 00:02:01,120 - > 00:02:04,719 kinds of AI providers are approaching it in different ways 47 00:02:04,719 - > 00:02:06,159 and what it means for us. 48 00:02:06,560 - > 00:02:06,959 Sinéad Hammond: Nice. 49 00:02:06,959 - > 00:02:08,639 Lots to dive into there. 50 00:02:08,639 - > 00:02:11,199 And then Rowan, from a marketing leader's point of 51 00:02:11,199 - > 00:02:14,159 view, what sort of things are on your radar this month around 52 00:02:14,159 - > 00:02:14,719 AI? 53 00:02:15,120 - > 00:02:18,719 Rowan Gill: Um, so this month, Google has launched a new AI 54 00:02:18,719 - > 00:02:22,000 mode uh that replaces traditional search results with 55 00:02:22,000 - > 00:02:25,840 a conversational, fully reasoned AI answer that's all powered by 56 00:02:25,840 - > 00:02:26,400 Gemini. 57 00:02:26,400 - > 00:02:29,199 So we're obviously looking to that and that what it means for 58 00:02:29,199 - > 00:02:30,800 us and our and our marketing. 59 00:02:30,800 - > 00:02:34,400 It's yet another change to the search landscape. 60 00:02:34,400 - > 00:02:39,919 And as zero-click behaviour becomes a norm, brands have to 61 00:02:39,919 - > 00:02:44,080 optimise to be cited with inside the AI answer, not just ranked 62 00:02:44,080 - > 00:02:45,280 in the traditional way. 63 00:02:45,599 - > 00:02:46,719 Sinéad Hammond: Yeah, absolutely. 64 00:02:46,719 - > 00:02:49,599 I think there's so many individual little changes going 65 00:02:49,599 - > 00:02:51,919 on with every change that happens with AI at the moment. 66 00:02:51,919 - > 00:02:55,280 So just these small things and how they're changing big time in 67 00:02:55,280 - > 00:02:57,919 terms of the strategy and the way that marketing departments 68 00:02:57,919 - > 00:02:59,919 are looking at their AI is really interesting. 69 00:02:59,919 - > 00:03:01,759 And it's a good job you're here, because that's what we'll 70 00:03:01,759 - > 00:03:04,240 be talking about a little bit later in the episode. 71 00:03:04,240 - > 00:03:07,599 Um, so Mark, you mentioned knowledge work is cooked. 72 00:03:07,599 - > 00:03:08,639 What does that mean? 73 00:03:08,639 - > 00:03:11,360 In plain English, what's changed, what's still very 74 00:03:11,360 - > 00:03:15,360 human, and what change has come on that has made this such an 75 00:03:15,360 - > 00:03:16,080 important topic? 76 00:03:16,479 - > 00:03:19,439 Mark Rotheram: Yeah, so I think if we go back a week or so, 77 00:03:19,439 - > 00:03:23,120 there was a really interesting article that kind of summarised 78 00:03:23,120 - > 00:03:24,800 the key message. 79 00:03:24,800 - > 00:03:29,360 Uh and it it was Microsoft's AI CEO that in the Financial Times 80 00:03:29,360 - > 00:03:33,759 was talking about how we are reaching the point where AI has 81 00:03:33,759 - > 00:03:37,919 already got human-level performance for a lot of 82 00:03:37,919 - > 00:03:38,800 professional tasks. 83 00:03:38,800 - > 00:03:41,680 So anything, you know, where you're sitting down at a 84 00:03:41,680 - > 00:03:46,800 computer doing something, AI is already very capable in that. 85 00:03:46,800 - > 00:03:50,800 And and the prediction that he made was in the next 12 to 18 86 00:03:50,800 - > 00:03:52,479 months, AI will be doing it all. 87 00:03:52,479 - > 00:03:54,319 So really interesting article. 88 00:03:54,319 - > 00:03:58,080 But I think it's come off the back of capabilities that have 89 00:03:58,080 - > 00:04:01,520 emerged, you know, in the last not just six months, but the 90 00:04:01,520 - > 00:04:02,639 last few weeks as well. 91 00:04:02,639 - > 00:04:06,479 And what I mean by that is let's look at benchmarks, the 92 00:04:06,479 - > 00:04:11,840 ability that the agents are demonstrating is pretty much 93 00:04:11,840 - > 00:04:13,120 doubling every four months. 94 00:04:13,120 - > 00:04:15,840 So if you rock back four months, we had some awesome 95 00:04:15,840 - > 00:04:21,279 models, Opus 4.5 came out, and it could do five hours of fully 96 00:04:21,279 - > 00:04:24,480 autonomous work, and the quality was very, very impressive. 97 00:04:24,480 - > 00:04:28,800 A few weeks ago, 4.6 came out, and it's quicker, it's faster, 98 00:04:28,800 - > 00:04:32,160 it's better, and it you know, the ability and capability has 99 00:04:32,160 - > 00:04:32,639 doubled. 100 00:04:32,639 - > 00:04:36,399 You know, we're ending up with this really, really smart brain 101 00:04:36,399 - > 00:04:40,319 that's able to do things as good, if not better, than people 102 00:04:40,319 - > 00:04:42,079 that are doing those jobs today. 103 00:04:42,079 - > 00:04:46,399 So, yeah, it's the ability that is being demonstrated, 104 00:04:46,399 - > 00:04:50,079 basically being able to do really complex cognitive work. 105 00:04:50,079 - > 00:04:52,480 So that that's the disruption that we're seeing. 106 00:04:52,480 - > 00:04:55,439 And I think it it's really interesting to see where that's 107 00:04:55,439 - > 00:05:00,079 gonna take us as we look at the adoption of that and what it 108 00:05:00,079 - > 00:05:02,959 means for the roles and the tasks that people are doing 109 00:05:02,959 - > 00:05:03,439 today. 110 00:05:04,399 - > 00:05:05,120 Sinéad Hammond: Absolutely. 111 00:05:05,120 - > 00:05:08,319 And I mean, from your perspective, I think we talked 112 00:05:08,319 - > 00:05:11,360 about this in the last episode, you know, jobs and tasks are 113 00:05:11,360 - > 00:05:13,360 gonna change massively with this. 114 00:05:13,360 - > 00:05:15,839 What's gonna be the differentiator now? 115 00:05:15,839 - > 00:05:19,040 What is that thing that people are going to need to kind of 116 00:05:19,040 - > 00:05:22,879 have that's going to require that human intervention? 117 00:05:23,360 - > 00:05:26,000 Mark Rotheram: Yeah, so we we always talk about human in the 118 00:05:26,000 - > 00:05:30,560 loop, and and I think that is gonna be the main theme here 119 00:05:30,560 - > 00:05:34,480 around where the the human actually adds value, because we 120 00:05:34,480 - > 00:05:37,040 we've got an awful lot of value to bring to this conversation. 121 00:05:37,040 - > 00:05:39,519 It's figuring out where that is and and why it is. 122 00:05:39,519 - > 00:05:42,560 Um the word that I'm gonna use is judgment. 123 00:05:42,560 - > 00:05:47,920 So we can use our AI agents, our little AI friends, to do an 124 00:05:47,920 - > 00:05:52,079 awful lot of wonderful work for us, but it comes down to 125 00:05:52,079 - > 00:05:56,000 judgment at the beginning of that process to articulate what 126 00:05:56,000 - > 00:05:58,399 you want it to do and why you want it to do it and how you 127 00:05:58,399 - > 00:05:59,839 want it to feed back. 128 00:05:59,839 - > 00:06:03,519 And then it's at the end of the process, again, exercising that 129 00:06:03,519 - > 00:06:06,240 judgment to say, has it actually done something useful? 130 00:06:06,240 - > 00:06:10,160 Um the AIs are super intelligent, but if you ask it 131 00:06:10,160 - > 00:06:13,360 to do something and it makes the wrong assumption, you will end 132 00:06:13,360 - > 00:06:16,879 up with an awesome amount of stuff that's not very useful. 133 00:06:16,879 - > 00:06:19,360 And and I think it's that judgment that that we're gonna 134 00:06:19,360 - > 00:06:23,279 bring, and and that's gonna come from people that that can think 135 00:06:23,279 - > 00:06:25,759 about you know, critical thinking of what it is we're 136 00:06:25,759 - > 00:06:28,399 trying to get this thing to do that's really gonna make the 137 00:06:28,399 - > 00:06:28,800 difference. 138 00:06:28,800 - > 00:06:31,920 You know, it's the difference between getting an AI to create 139 00:06:31,920 - > 00:06:37,199 a a wonderful 40-page PowerPoint presentation that is drivel, 140 00:06:37,199 - > 00:06:39,600 uh, versus three slides that absolutely nail it. 141 00:06:39,600 - > 00:06:42,480 You know, the AI can do that 40-page deck and it'll love it 142 00:06:42,480 - > 00:06:44,000 and it'll be look awesome. 143 00:06:44,000 - > 00:06:46,560 But that that you know, that judgment bit that's gonna come 144 00:06:46,560 - > 00:06:50,240 in is really where we're gonna see the human value come through 145 00:06:50,240 - > 00:06:52,879 and control these agents that that are emerging. 146 00:06:53,199 - > 00:06:54,000 Sinéad Hammond: Amazing. 147 00:06:54,000 - > 00:06:57,439 And um, Rowan, I'm gonna come to you now because I think this 148 00:06:57,439 - > 00:07:00,720 is something that you're doing a lot in your team and making 149 00:07:00,720 - > 00:07:05,360 sure that we aren't just coming out with drivel as Mark puts it 150 00:07:05,360 - > 00:07:05,600 there. 151 00:07:05,600 - > 00:07:09,360 So talk us through on a practical level for business 152 00:07:09,360 - > 00:07:14,319 leaders what you're doing as a department leads to make sure 153 00:07:14,319 - > 00:07:18,319 that you're using AI and you're using it for that, you know, 154 00:07:18,319 - > 00:07:20,800 vast amount of work that's coming out, but keeping that 155 00:07:20,800 - > 00:07:24,160 quality, keeping that relevancy, tell talk us through a little 156 00:07:24,160 - > 00:07:24,639 bit of that. 157 00:07:24,959 - > 00:07:25,600 Rowan Gill: Yeah, sure. 158 00:07:25,600 - > 00:07:28,720 And I'm really happy to hear obviously Mark talk about 159 00:07:28,720 - > 00:07:32,399 judgment and where bringing value through having the human 160 00:07:32,399 - > 00:07:35,759 in the loop and people with experience and expertise. 161 00:07:35,759 - > 00:07:42,000 AI is enabling us as a team to do more faster, but where we're 162 00:07:42,000 - > 00:07:46,560 really focused is ensuring that we are using AI in the right 163 00:07:46,560 - > 00:07:51,360 place and also that the output is actually a great quality. 164 00:07:51,360 - > 00:07:56,879 There's lots and lots of um fatigue around AI and having AI 165 00:07:56,879 - > 00:08:00,800 slop and people being able to produce lots and lots of 166 00:08:00,800 - > 00:08:06,240 content, but for us as a team, it's really important for us to 167 00:08:06,240 - > 00:08:11,040 always remain on brand, keep the thing that makes BCN different 168 00:08:11,040 - > 00:08:14,800 and special, make sure that that is retained in all of our 169 00:08:14,800 - > 00:08:18,079 communications across all our channels and any marketing that 170 00:08:18,079 - > 00:08:18,639 we do. 171 00:08:18,639 - > 00:08:22,959 So, from a practical perspective, that's around using 172 00:08:22,959 - > 00:08:26,079 our best judgment to decide what we're adopting and what's 173 00:08:26,079 - > 00:08:27,040 going to make the difference. 174 00:08:27,040 - > 00:08:31,199 But it's also then when we are adopting a new tool and 175 00:08:31,199 - > 00:08:34,960 utilising AI to help us with things, it's also making sure 176 00:08:34,960 - > 00:08:39,679 that we have the right skill set to put the right prompts in, to 177 00:08:39,679 - > 00:08:42,480 put the right brand guidelines in. 178 00:08:42,480 - > 00:08:46,399 So, for example, everything in terms of our brand guidelines, 179 00:08:46,399 - > 00:08:51,039 our tone of voice, we will align it to our strategic priorities, 180 00:08:51,039 - > 00:08:54,879 key messages we we are wanting as a business to convey within 181 00:08:54,879 - > 00:08:55,759 our marketing. 182 00:08:55,759 - > 00:09:00,559 And then we will also layer on things that we know people 183 00:09:00,559 - > 00:09:02,399 instantly spot as AI. 184 00:09:02,399 - > 00:09:06,080 So, what we don't want it to tell us, we don't want it to 185 00:09:06,080 - > 00:09:10,399 come back with our terms that we we want to put in room 101. 186 00:09:10,399 - > 00:09:14,799 So, you know, those long M dashes, those Americanisms, you 187 00:09:14,799 - > 00:09:15,600 know, all of that. 188 00:09:15,600 - > 00:09:20,159 We we have a list of phrases that we constantly update, 189 00:09:20,159 - > 00:09:23,120 making sure that we've got all of the right prompts in place 190 00:09:23,120 - > 00:09:26,399 and giving the AI the right knowledge it needs to produce 191 00:09:26,399 - > 00:09:29,039 great quality content that aligns with our brand. 192 00:09:29,039 - > 00:09:33,519 And that's really important because more than ever, AI is 193 00:09:33,519 - > 00:09:36,320 doing a lot more of the groundwork for marketing. 194 00:09:36,320 - > 00:09:40,480 So bringing the human back into the loop to look at that 195 00:09:40,480 - > 00:09:45,840 40-page PowerPoint or look at that blog post or look at the 196 00:09:45,840 - > 00:09:48,799 social posts or an entire campaign and know what will and 197 00:09:48,799 - > 00:09:52,559 what won't work is really where people can use their um use 198 00:09:52,559 - > 00:09:54,879 their experience to add value around AI. 199 00:09:55,200 - > 00:09:56,799 Sinéad Hammond: I like what you said there about using your 200 00:09:56,799 - > 00:09:57,200 judgment. 201 00:09:57,200 - > 00:09:58,480 I think that's really important. 202 00:09:58,480 - > 00:10:02,080 And I think going on to the next topic, Mark, that you 203 00:10:02,080 - > 00:10:06,320 mentioned there was around the open claw news that happened 204 00:10:06,320 - > 00:10:06,799 last week. 205 00:10:06,799 - > 00:10:09,360 You're gonna have to start from the beginning and tell me 206 00:10:09,360 - > 00:10:11,039 first, what is open claw? 207 00:10:11,039 - > 00:10:14,159 Describe to me what happened last week, why is this so 208 00:10:14,159 - > 00:10:15,840 important, and what does this mean for businesses? 209 00:10:15,840 - > 00:10:17,039 Why should they be listening? 210 00:10:17,360 - > 00:10:21,279 Mark Rotheram: Yeah, so open claw is a really interesting 211 00:10:21,279 - > 00:10:21,759 development. 212 00:10:21,759 - > 00:10:24,320 So it's probably been about a couple of weeks now since this 213 00:10:24,320 - > 00:10:24,559 hit. 214 00:10:24,559 - > 00:10:26,080 So effectively, what happened? 215 00:10:26,080 - > 00:10:28,799 There was a single guy, a little developer, who'd been 216 00:10:28,799 - > 00:10:31,919 push publishing software for a while, and he published this 217 00:10:31,919 - > 00:10:34,480 thing called it wasn't called OpenClaw, there's been a few 218 00:10:34,480 - > 00:10:36,240 names, but we'll leave it as OpenClaw now. 219 00:10:36,240 - > 00:10:41,039 But effectively, it was a scaffold that allows you to 220 00:10:41,039 - > 00:10:43,519 install a piece of software on a computer. 221 00:10:43,519 - > 00:10:46,559 This is where we had lots of fun around, uh a rush for 222 00:10:46,559 - > 00:10:51,440 MacBooks and Mac Minis and most recently Raspberry Pis, all 223 00:10:51,440 - > 00:10:55,120 around people wanting to host this um this bot. 224 00:10:55,120 - > 00:10:57,759 And the best way to describe the bot is if you think about 225 00:10:57,759 - > 00:11:00,080 Iron Man and Jarvis, where you've got your personal 226 00:11:00,080 - > 00:11:00,480 assistant. 227 00:11:00,480 - > 00:11:04,159 You know, we're not talking uh Siri or Alexa or um that those 228 00:11:04,159 - > 00:11:04,559 kind of things. 229 00:11:04,559 - > 00:11:09,360 We're talking actually a real AI that can do things for you. 230 00:11:09,360 - > 00:11:11,440 That's kind of what this was all about. 231 00:11:11,440 - > 00:11:15,600 So it went from zero to a hundred thousand GitHub stars, 232 00:11:15,600 - > 00:11:18,480 which basically means it's the fastest growing open source 233 00:11:18,480 - > 00:11:19,360 project in history. 234 00:11:19,360 - > 00:11:21,840 So huge, huge adoption. 235 00:11:21,840 - > 00:11:25,440 And yeah, it it launched, it was on your device, you 236 00:11:25,440 - > 00:11:29,600 communicate to it using WhatsApp or Telegram or iMessage, and 237 00:11:29,600 - > 00:11:33,279 yeah, it can pretty much do anything you let it, uh, which 238 00:11:33,279 - > 00:11:34,639 is really interesting. 239 00:11:35,039 - > 00:11:37,279 Sinéad Hammond: It sounds interesting, exciting. 240 00:11:37,279 - > 00:11:40,799 All of a sudden, I'm thinking, I assume it has access to 241 00:11:40,799 - > 00:11:41,360 everything. 242 00:11:41,360 - > 00:11:45,200 Is there like a level of permissions that you might 243 00:11:45,200 - > 00:11:47,679 already my head's thinking, what can't it access? 244 00:11:47,679 - > 00:11:51,120 Does it have access to bank details and all of that stuff? 245 00:11:51,120 - > 00:11:52,320 Is that all part of it? 246 00:11:52,799 - > 00:11:55,279 Mark Rotheram: Yeah, again, this is this is where we we end up 247 00:11:55,279 - > 00:11:59,759 with a a very interesting take from in the corporate world 248 00:11:59,759 - > 00:12:02,879 where we operate and we advise all of our customers. 249 00:12:02,879 - > 00:12:05,759 Um the first thing I said, but let's not do this quite yet. 250 00:12:05,759 - > 00:12:08,159 You know, we're we're on a journey to agentic, this is 251 00:12:08,159 - > 00:12:11,679 looking awesome, but let let's not quite do that because at its 252 00:12:11,679 - > 00:12:15,200 heart it is a permissionless deployment of AI. 253 00:12:15,200 - > 00:12:19,679 And what I mean by that is, yeah, you deploy it on a device 254 00:12:19,679 - > 00:12:22,879 and you give it access to a lot of things, yeah. 255 00:12:22,879 - > 00:12:26,159 And you can control what you give it access to, but the 256 00:12:26,159 - > 00:12:28,879 out-of-the-box deployments that people were doing were, you 257 00:12:28,879 - > 00:12:31,360 know, you can access my bank, you can access Amazon, you can 258 00:12:31,360 - > 00:12:34,159 access um my iMessage, you can access all these things. 259 00:12:34,159 - > 00:12:38,000 Now, the the first few days were a nightmare, if I'm honest, 260 00:12:38,000 - > 00:12:41,840 for um the community because the original name had to change, 261 00:12:41,840 - > 00:12:44,240 they've they changed the name a few times, and in between those 262 00:12:44,240 - > 00:12:46,960 name changes, mainly because Anthropic weren't happy with it 263 00:12:46,960 - > 00:12:48,879 being called ClaudeBot to start off with. 264 00:12:48,879 - > 00:12:52,559 We had hackers get involved, they managed to get domain names 265 00:12:52,559 - > 00:12:54,639 and inject malicious code. 266 00:12:54,639 - > 00:12:58,320 Even just deploying it um for the first few days was a bit 267 00:12:58,320 - > 00:13:00,000 fraught with people getting hacked. 268 00:13:00,000 - > 00:13:03,279 And then we've got this really interesting kind of disclaimer 269 00:13:03,279 - > 00:13:07,679 that says, you shouldn't do this if you're not technical, which 270 00:13:07,679 - > 00:13:10,320 is like saying, well, it's it's one of those ridiculous 271 00:13:10,320 - > 00:13:12,879 disclaimers that says, you know, nuts might contain traces of 272 00:13:12,879 - > 00:13:13,279 nuts. 273 00:13:13,279 - > 00:13:14,639 People are going to do it anyway. 274 00:13:14,639 - > 00:13:17,120 So you've got all these people that are installing it that 275 00:13:17,120 - > 00:13:20,240 don't understand how to secure it, which was leaving lots and 276 00:13:20,240 - > 00:13:24,080 lots of holes for people to get in and hack and steal keys, 277 00:13:24,080 - > 00:13:25,840 accounts, all that kind of stuff. 278 00:13:25,840 - > 00:13:29,279 So yeah, we we went from a an awesome thing to actually this 279 00:13:29,279 - > 00:13:33,679 is quite insecure and scary, to kind of actually a huge 280 00:13:33,679 - > 00:13:35,600 evolution of kind of what's happening now. 281 00:13:36,000 - > 00:13:39,120 Sinéad Hammond: Interesting, exciting, terrifying, I think is 282 00:13:39,120 - > 00:13:40,480 probably the words I'd describe that. 283 00:13:40,480 - > 00:13:48,000 And I think for us at BCM, we are quite aware of how much AI 284 00:13:48,000 - > 00:13:49,679 is getting access to. 285 00:13:49,679 - > 00:13:51,919 And I know you mentioned that when you were talking about what 286 00:13:51,919 - > 00:13:54,879 businesses that we work with, how we kind of help them with 287 00:13:54,879 - > 00:13:58,639 adoption there and enable it to be something that is innovative 288 00:13:58,639 - > 00:14:03,120 and supportive, but also doesn't have those gaps that enable 289 00:14:03,120 - > 00:14:05,919 hackers and threats to kind of infiltrate. 290 00:14:05,919 - > 00:14:09,200 Sounds like Open Claw last week was an example of that 291 00:14:09,200 - > 00:14:12,240 happening on scale in a very bad way. 292 00:14:12,240 - > 00:14:17,440 Um it's gonna be exciting to see what this next stage of AI 293 00:14:17,440 - > 00:14:18,960 looks like off the back of this. 294 00:14:18,960 - > 00:14:21,840 What do businesses need to think about or business leaders 295 00:14:21,840 - > 00:14:24,639 need to think about with the announcement of something like 296 00:14:24,639 - > 00:14:24,799 this? 297 00:14:24,799 - > 00:14:27,360 What does that bring up again to the surface? 298 00:14:27,679 - > 00:14:31,120 Mark Rotheram: Yeah, so I guess this was done by an individual. 299 00:14:31,120 - > 00:14:34,080 There were no new models, there was nothing groundbreaking, it 300 00:14:34,080 - > 00:14:37,440 was literally a framework of how to use current gen AI. 301 00:14:37,440 - > 00:14:40,399 And it the the concept has exploded. 302 00:14:40,399 - > 00:14:46,159 So if if you take just that concept of current gen AI with a 303 00:14:46,159 - > 00:14:50,000 whole ecosystem that was thought out by one guy in his 304 00:14:50,000 - > 00:14:53,039 bedroom just knocking things together and making such an 305 00:14:53,039 - > 00:14:56,799 impact, bringing that concept to the business is really 306 00:14:56,799 - > 00:14:57,200 interesting. 307 00:14:57,200 - > 00:15:00,799 We've got this really wonderful AI capability. 308 00:15:00,799 - > 00:15:04,720 What are the use cases that we could deploy in a safe way? 309 00:15:04,720 - > 00:15:07,840 What are the outcomes we could be targeting that you haven't 310 00:15:07,840 - > 00:15:08,559 thought of yet? 311 00:15:08,559 - > 00:15:11,840 A lot of the the most interesting things that we end 312 00:15:11,840 - > 00:15:13,519 up doing are new things. 313 00:15:13,519 - > 00:15:16,639 Yeah, we spend an awful lot of time talking about how can we 314 00:15:16,639 - > 00:15:20,879 use AI to solve an existing outcome or an existing process. 315 00:15:20,879 - > 00:15:23,919 But the fun and the real exciting thing is when we we 316 00:15:23,919 - > 00:15:27,039 actually enable something we've not been able to do before in a 317 00:15:27,039 - > 00:15:28,720 way that AI is allowing us to. 318 00:15:28,720 - > 00:15:31,759 If you look at all the new things that have come off the 319 00:15:31,759 - > 00:15:36,240 back of this, you know, we ended up with MaltBook, which was a 320 00:15:36,240 - > 00:15:39,200 place for all these agents that they're called Maltis to go and 321 00:15:39,200 - > 00:15:41,440 chat with each other on a social platform. 322 00:15:41,440 - > 00:15:45,759 We ended up with a whole array of other social things being 323 00:15:45,759 - > 00:15:48,000 created by the agents for the agents. 324 00:15:48,000 - > 00:15:51,440 We've got Coinbase who enabled them to be banked now. 325 00:15:51,440 - > 00:15:55,039 So there is an AI wallet for the agents, so they've got money 326 00:15:55,039 - > 00:15:55,919 that can do things. 327 00:15:55,919 - > 00:16:01,440 We have examples where people have used AI to negotiate a car 328 00:16:01,440 - > 00:16:04,159 price and make, you know, they've they've got dealers 329 00:16:04,159 - > 00:16:07,120 pacing off with the AI in the middle to get a better deal. 330 00:16:07,120 - > 00:16:11,360 Uh we've got them going rogue, where these bots are taking over 331 00:16:11,360 - > 00:16:12,000 iMessage. 332 00:16:12,000 - > 00:16:16,480 We had one guy that he he left it open and his his little bot 333 00:16:16,480 - > 00:16:19,840 sent 1,500 messages to his wife on iMessage, you know, before he 334 00:16:19,840 - > 00:16:20,240 could stop it. 335 00:16:20,240 - > 00:16:22,879 So you've got some some really interesting stuff going on. 336 00:16:22,879 - > 00:16:27,279 But I guess to take it back to the business context, it's a 337 00:16:27,279 - > 00:16:29,600 demonstration of what we can do today. 338 00:16:29,600 - > 00:16:30,000 Yeah. 339 00:16:30,000 - > 00:16:33,759 All those use cases went through didn't exist a couple of 340 00:16:33,759 - > 00:16:34,399 weeks ago. 341 00:16:34,399 - > 00:16:35,519 They exist now. 342 00:16:35,519 - > 00:16:37,600 They're ideas of things we can do. 343 00:16:37,600 - > 00:16:43,039 We can build safer versions of OpenClaw or Moltbot. 344 00:16:43,039 - > 00:16:46,399 We can build these little agents, and it's really now the 345 00:16:46,399 - > 00:16:47,120 art of the possible. 346 00:16:47,120 - > 00:16:49,759 What what is the the thing that's really going to make that 347 00:16:49,759 - > 00:16:51,120 outcome and bring it to life? 348 00:16:51,440 - > 00:16:54,000 Sinéad Hammond: And it's a what would then happen in three weeks 349 00:16:54,000 - > 00:16:55,679 from now, six weeks, twelve weeks? 350 00:16:55,679 - > 00:16:58,799 Like there is just so much, so fast happening. 351 00:16:58,799 - > 00:16:59,840 I think it's incredible. 352 00:16:59,840 - > 00:17:01,360 It's so hard to keep up with. 353 00:17:01,360 - > 00:17:05,279 Um, Rowan, from your side, I can imagine your kind of alarm 354 00:17:05,279 - > 00:17:07,039 bells going off a little bit with this. 355 00:17:07,039 - > 00:17:10,480 Because I know that when you're deploying or implementing 356 00:17:10,480 - > 00:17:14,640 tools, there's always that consideration of the permission 357 00:17:14,640 - > 00:17:15,200 level. 358 00:17:15,200 - > 00:17:19,839 What do you think on a practical level we can do or the 359 00:17:19,839 - > 00:17:23,200 marketing teams can do that enable us to have this 360 00:17:23,200 - > 00:17:24,799 capability that Mark's talking about? 361 00:17:24,799 - > 00:17:28,240 It's just really exciting, cool, innovative things that we 362 00:17:28,240 - > 00:17:30,720 were never able to do before, but in a way that's not going to 363 00:17:30,720 - > 00:17:32,000 expose us to threats. 364 00:17:32,640 - > 00:17:37,599 Rowan Gill: Yeah, security is absolutely high on the agenda 365 00:17:37,599 - > 00:17:41,200 for us as a marketing function, but also as a business. 366 00:17:41,200 - > 00:17:45,440 We've got really, really strong security foundation and an 367 00:17:45,440 - > 00:17:47,920 amazing internal cybersecurity team. 368 00:17:47,920 - > 00:17:51,680 So, as well as all of the like the lead-in-age tools that are 369 00:17:51,680 - > 00:17:54,400 protecting us in the background and running every day that we 370 00:17:54,400 - > 00:17:59,279 don't realise, we've also got a really robust cybersecurity 371 00:17:59,279 - > 00:18:03,440 awareness training program that happens throughout our business, 372 00:18:03,440 - > 00:18:09,519 which is making us aware of the risks that AI can bring if we 373 00:18:09,519 - > 00:18:12,480 are not using it correctly, safely, and securely. 374 00:18:12,480 - > 00:18:17,359 So for us, those security considerations are baked into 375 00:18:17,359 - > 00:18:21,599 the thought process when we're evaluating any new tool that we 376 00:18:21,599 - > 00:18:24,319 might want to bring in to help us solve some issues or some 377 00:18:24,319 - > 00:18:26,640 problems or make us do more faster. 378 00:18:26,640 - > 00:18:31,119 So we will always check in with our cybersecurity team to 379 00:18:31,119 - > 00:18:33,680 check: does that new tool have the permissions? 380 00:18:33,680 - > 00:18:35,359 Does it have MFA? 381 00:18:35,359 - > 00:18:37,440 Where is our data going? 382 00:18:37,440 - > 00:18:41,359 Are we potentially putting the business at risk, allowing 383 00:18:41,359 - > 00:18:43,279 somebody to come in via these tools? 384 00:18:43,279 - > 00:18:47,440 And if if there's any risk at all, we will not use that tool, 385 00:18:47,440 - > 00:18:50,319 even though sometimes it's really disappointing because 386 00:18:50,319 - > 00:18:52,880 they could be really shiny, wonderful new tools that could 387 00:18:52,880 - > 00:18:53,119 help. 388 00:18:53,119 - > 00:18:55,519 But for us, security is far more important. 389 00:18:55,519 - > 00:18:59,680 In terms of what we do know, what is secure, which does help 390 00:18:59,680 - > 00:19:02,480 us massively from a really practical level. 391 00:19:02,480 - > 00:19:06,799 We utilize Microsoft Copilot all the time, obviously, because 392 00:19:06,799 - > 00:19:10,799 it sits within our business, it's safe and it's secure, and 393 00:19:10,799 - > 00:19:16,480 we know that is a great place for our team to adopt and move 394 00:19:16,480 - > 00:19:17,599 forwards with AI. 395 00:19:17,599 - > 00:19:20,480 And then because we're not stopping the adoption, but what 396 00:19:20,480 - > 00:19:22,799 we're doing here is doing it safely and securely. 397 00:19:23,039 - > 00:19:25,119 Sinéad Hammond: Yeah, I think it sounds like there's a huge 398 00:19:25,119 - > 00:19:26,319 amount that we can do. 399 00:19:26,319 - > 00:19:29,759 It's just making sure you put those steps in place in advance 400 00:19:29,759 - > 00:19:32,720 because I think a lot of people will be wanting to use and test 401 00:19:32,720 - > 00:19:33,839 out some of these cool things. 402 00:19:33,839 - > 00:19:36,559 I mean, I'm not I'm not sure how much I want a bot to send 403 00:19:36,559 - > 00:19:38,240 1500 messages to a contact. 404 00:19:38,240 - > 00:19:41,200 I'm not feeling that's quite what I'm gonna ask AI to do or 405 00:19:41,200 - > 00:19:42,240 I'm gonna want it to do. 406 00:19:42,240 - > 00:19:45,200 But there's all this capability which is really exciting and 407 00:19:45,200 - > 00:19:47,759 things that I, you know, we really want to explore, and I as 408 00:19:47,759 - > 00:19:49,440 an employee want to explore. 409 00:19:49,440 - > 00:19:52,799 But I think it's quite good to have almost having that second 410 00:19:52,799 - > 00:19:55,039 guess stage when you're thinking, okay, I want to be 411 00:19:55,039 - > 00:19:56,799 able to do this, but what are the things that I need to 412 00:19:56,799 - > 00:19:58,160 consider before I'm pulling it in? 413 00:19:58,160 - > 00:20:00,640 And that's a business-wide thing, that's a culture thing, 414 00:20:00,640 - > 00:20:01,839 which I think is really important. 415 00:20:02,240 - > 00:20:05,519 Rowan Gill: Businesses want their employees to utilize AI 416 00:20:05,519 - > 00:20:06,559 and to experiment. 417 00:20:06,559 - > 00:20:09,839 And any business that thinks their employees aren't doing 418 00:20:09,839 - > 00:20:10,960 that are probably wrong. 419 00:20:10,960 - > 00:20:14,400 They probably are utilising AI tools, but they're doing it 420 00:20:14,400 - > 00:20:15,680 without anybody knowing about it. 421 00:20:15,680 - > 00:20:19,599 So this we've seen this massive rise in in shadow AI. 422 00:20:19,599 - > 00:20:22,960 So utilising tools that aren't approved, that aren't safe, that 423 00:20:22,960 - > 00:20:23,759 aren't secure. 424 00:20:23,759 - > 00:20:27,839 Business leaders need to empower their teams to utilize 425 00:20:27,839 - > 00:20:32,240 AI in a safe way rather than stop them or have them do it 426 00:20:32,240 - > 00:20:35,119 behind the scenes, which could be a potential security risk. 427 00:20:35,519 - > 00:20:36,319 Sinéad Hammond: Is that what you mean? 428 00:20:36,319 - > 00:20:37,839 Is that what shadow AI is? 429 00:20:37,839 - > 00:20:40,799 Do you want to just explain that term a little bit more what 430 00:20:40,799 - > 00:20:41,599 that actually means? 431 00:20:41,920 - > 00:20:45,839 Mark Rotheram: It it's basically where we we try and encourage 432 00:20:45,839 - > 00:20:49,119 people to use the right AI, but adoption's really hard. 433 00:20:49,119 - > 00:20:49,599 Yeah. 434 00:20:49,599 - > 00:20:53,119 So everyone that's in the Microsoft ecosystem has free 435 00:20:53,119 - > 00:20:54,079 co-pilot chat. 436 00:20:54,079 - > 00:20:56,880 But if they don't know that it's there or know what it can 437 00:20:56,880 - > 00:20:59,200 do, then they go to what they know. 438 00:20:59,200 - > 00:20:59,519 Yeah. 439 00:20:59,519 - > 00:21:02,880 So that's where you start seeing people go and use Gemini, 440 00:21:02,880 - > 00:21:06,720 Chat GPT, Claude, all these different ones, but they use the 441 00:21:06,720 - > 00:21:07,519 public versions. 442 00:21:07,519 - > 00:21:11,839 Now, the public versions are fine from a public perspective, 443 00:21:11,839 - > 00:21:14,640 but there's a reason why we have enterprise grade security and 444 00:21:14,640 - > 00:21:16,000 enterprise AI tools. 445 00:21:16,000 - > 00:21:19,519 So the shadow IT is where typically we've not done the 446 00:21:19,519 - > 00:21:24,480 best job in articulating what the AI tools are that we want to 447 00:21:24,480 - > 00:21:26,960 be using and why and the capabilities, or they've not 448 00:21:26,960 - > 00:21:28,240 been made available at all. 449 00:21:28,240 - > 00:21:30,880 And people are just left to go and find their own way. 450 00:21:30,880 - > 00:21:33,839 And when they do that, and you know, we've got lots of curious 451 00:21:33,839 - > 00:21:36,400 people, they will go and find tools that do the job for them 452 00:21:36,400 - > 00:21:37,200 because they're there. 453 00:21:37,200 - > 00:21:41,759 So um this ties really well into how you manage adoption and 454 00:21:41,759 - > 00:21:45,039 make sure you have got an AI policy because when you thought 455 00:21:45,039 - > 00:21:47,279 about it and you write your policy and you think, actually, 456 00:21:47,279 - > 00:21:49,279 right, how do I encourage people to do this? 457 00:21:49,279 - > 00:21:50,799 You know, it's carrot and stick. 458 00:21:50,799 - > 00:21:52,799 You know, you want to stop people doing the bad stuff where 459 00:21:52,799 - > 00:21:56,640 data might go out of control, but you really want to encourage 460 00:21:56,640 - > 00:21:59,599 people to use the ones that that are safe and explain. 461 00:21:59,599 - > 00:22:01,920 Them, what you can do with them. 462 00:22:01,920 - > 00:22:05,759 To go to the point, shadow AI is where people are basically 463 00:22:05,759 - > 00:22:09,359 left for whatever reason to go and use AI that you don't really 464 00:22:09,359 - > 00:22:12,079 want them to use instead of the tools that you're making 465 00:22:12,079 - > 00:22:12,880 available for them. 466 00:22:13,279 - > 00:22:14,160 Sinéad Hammond: That makes sense. 467 00:22:14,160 - > 00:22:17,519 And it leads us on to the third point that you talked about, 468 00:22:17,519 - > 00:22:19,759 the third headline that's come out, which is around some of the 469 00:22:19,759 - > 00:22:24,000 new updates and moderates, loads going on with open AI, 470 00:22:24,000 - > 00:22:27,759 more agentic AI, I think is what's kind of happening. 471 00:22:27,759 - > 00:22:31,519 Again, tell us a little bit about kind of what's new, what's 472 00:22:31,519 - > 00:22:35,920 happened, what's going on, and which updates businesses should 473 00:22:35,920 - > 00:22:37,440 be paying attention to. 474 00:22:37,839 - > 00:22:41,599 Mark Rotheram: Yeah, so there's a lot of tech in there, and I'm 475 00:22:41,599 - > 00:22:44,319 not going to go through all of the finer details. 476 00:22:44,319 - > 00:22:47,279 I mean, just suffice it to say, there's been some massive 477 00:22:47,279 - > 00:22:51,519 changes in the ability in the models, you know, from a numbers 478 00:22:51,519 - > 00:22:56,000 perspective, you know, we've seen benchmarks continuously 479 00:22:56,000 - > 00:22:57,039 improved upon. 480 00:22:57,039 - > 00:23:01,039 Um, the context windows are increasing, and that means 481 00:23:01,039 - > 00:23:04,960 basically that AI can think for for longer and uh cover more 482 00:23:04,960 - > 00:23:05,359 topics. 483 00:23:05,359 - > 00:23:07,599 Imagine you can throw more books worth of data into it 484 00:23:07,599 - > 00:23:09,279 before it runs out of of memory. 485 00:23:09,519 - > 00:23:09,920 Sinéad Hammond: Nice. 486 00:23:10,079 - > 00:23:12,799 Mark Rotheram: Um, but what we're also seeing is new skills 487 00:23:12,799 - > 00:23:14,559 arrive within these models. 488 00:23:14,559 - > 00:23:17,119 And the skills are things that we might have been doing 489 00:23:17,119 - > 00:23:19,680 manually with the AI models previously. 490 00:23:19,680 - > 00:23:23,759 So some really good examples are where we're seeing agent 491 00:23:23,759 - > 00:23:25,200 teams come into play. 492 00:23:25,200 - > 00:23:29,039 And this is where instead of saying AI, go and do this job, 493 00:23:29,039 - > 00:23:32,319 and it goes and does that job, it will then spin up a team of 494 00:23:32,319 - > 00:23:35,119 agents that will coordinate the work, they'll split it up, 495 00:23:35,119 - > 00:23:37,519 they'll work together, and they'll all run in parallel and 496 00:23:37,519 - > 00:23:38,799 they'll come back at the end. 497 00:23:38,799 - > 00:23:43,440 And this means that you you can offload with the right judgment 498 00:23:43,440 - > 00:23:45,519 and planning even more work. 499 00:23:45,519 - > 00:23:48,799 And that ties into another kind of really interesting feature, 500 00:23:48,799 - > 00:23:51,440 which is the length of time these models can run. 501 00:23:51,440 - > 00:23:55,039 Previously they'd run out of grunt after seconds, then 502 00:23:55,039 - > 00:23:56,720 minutes, then hours. 503 00:23:56,720 - > 00:24:00,319 We're now getting into the point where they can run for a 504 00:24:00,319 - > 00:24:03,759 long time, and and that is increasing all the time as well. 505 00:24:03,759 - > 00:24:09,440 So again, feeding in the right input becomes paramount to make 506 00:24:09,440 - > 00:24:13,279 sure that the output after that length of time is what you 507 00:24:13,279 - > 00:24:14,000 expected. 508 00:24:14,000 - > 00:24:17,920 So, yeah, we we're seeing these kind of skills and abilities 509 00:24:17,920 - > 00:24:21,359 creep up and you know the intelligence levels just tick up 510 00:24:21,359 - > 00:24:21,759 as well. 511 00:24:21,759 - > 00:24:25,920 So uh I guess they're not just smarter chatbots, they're models 512 00:24:25,920 - > 00:24:28,880 that are designed to run for hours, they're designed to 513 00:24:28,880 - > 00:24:32,160 coordinate work, split it up, run in parallel, bring it all 514 00:24:32,160 - > 00:24:35,200 back together again, and they've got the ability to have much 515 00:24:35,200 - > 00:24:37,680 broader context than they've ever had before. 516 00:24:37,680 - > 00:24:40,319 And we're seeing you know these things happen across the 517 00:24:40,319 - > 00:24:41,839 different vendors, you know. 518 00:24:41,839 - > 00:24:46,240 Anthropic are very much focused on bringing more and more 519 00:24:46,240 - > 00:24:49,359 intelligence and capability at roughly the same price point 520 00:24:49,359 - > 00:24:50,559 that the previous model was. 521 00:24:50,559 - > 00:24:55,440 OpenAI have have done some amazing stuff with GPT 5.2 and 522 00:24:55,440 - > 00:24:59,920 most recently 5.3, where their approach is, you know, we're 523 00:24:59,920 - > 00:25:02,720 going to make everything cheaper as well as nudging up the 524 00:25:02,720 - > 00:25:03,839 intelligence, I'd say. 525 00:25:03,839 - > 00:25:06,480 And then we start to get into what's happening in the open 526 00:25:06,480 - > 00:25:09,279 source world, which again is is really interesting. 527 00:25:09,279 - > 00:25:15,039 You know, Microsoft initially partner with OpenAI for co-pilot 528 00:25:15,039 - > 00:25:16,880 that was powered by their models. 529 00:25:16,880 - > 00:25:19,759 They then brought in some anthropic, um, so they're 530 00:25:19,759 - > 00:25:22,240 they're looking at the enterprise grade, and now we're 531 00:25:22,240 - > 00:25:25,519 seeing the emergence of these permissionless, again, we use 532 00:25:25,519 - > 00:25:30,079 that word open source models, which are are lagging behind for 533 00:25:30,079 - > 00:25:32,799 the moment that the big players, but they're open 534 00:25:32,799 - > 00:25:33,200 source. 535 00:25:33,200 - > 00:25:36,000 And this this brings some really interesting perspectives 536 00:25:36,000 - > 00:25:39,839 around data sovereignty and AI sovereignty and what you trust 537 00:25:39,839 - > 00:25:40,799 and what you don't trust. 538 00:25:40,799 - > 00:25:46,240 So give a really good example, the best models always come out 539 00:25:46,240 - > 00:25:50,720 at a global level, which means that we will have to use global 540 00:25:50,720 - > 00:25:53,279 processing for the model to work. 541 00:25:53,279 - > 00:25:56,720 Whereas the open source models, we can run them locally, you 542 00:25:56,720 - > 00:25:59,119 can run them on your laptop if you wanted to, which means the 543 00:25:59,119 - > 00:26:00,240 data goes nowhere. 544 00:26:00,240 - > 00:26:03,680 But the interesting bit is how comfortable we are using models 545 00:26:03,680 - > 00:26:05,920 that have originated from different countries. 546 00:26:05,920 - > 00:26:09,359 So uh you end up with all these different perspectives of um 547 00:26:09,359 - > 00:26:11,920 governance and control that you've got to grapple with to 548 00:26:11,920 - > 00:26:14,160 make sure you're making the right decisions at the right 549 00:26:14,160 - > 00:26:17,440 time to kind of get the most out of these models as they as they 550 00:26:17,440 - > 00:26:17,920 come online. 551 00:26:18,319 - > 00:26:20,799 Sinéad Hammond: Just as a bit of a definition thing there, 552 00:26:20,799 - > 00:26:23,599 because I'm you know that I'm not technical at all. 553 00:26:23,599 - > 00:26:24,960 So just a few quick ones. 554 00:26:24,960 - > 00:26:26,799 First of all, open source. 555 00:26:26,799 - > 00:26:28,000 What's open source? 556 00:26:28,000 - > 00:26:29,200 What was an example of that? 557 00:26:29,200 - > 00:26:31,119 What's what does that look like? 558 00:26:31,519 - > 00:26:33,440 Mark Rotheram: Yeah, so we've we've used the term a couple of 559 00:26:33,440 - > 00:26:34,160 times today. 560 00:26:34,160 - > 00:26:38,240 So open source is where people develop something and they make 561 00:26:38,240 - > 00:26:39,839 it publicly accessible. 562 00:26:39,839 - > 00:26:44,559 So people can take it, they can rewrite it in some cases. 563 00:26:44,559 - > 00:26:47,599 So it's it's kind of like the Komodo is open, yeah? 564 00:26:47,599 - > 00:26:49,440 They show you exactly how it works. 565 00:26:49,440 - > 00:26:54,240 The open AI and anthropic approach is private, so we can 566 00:26:54,240 - > 00:26:57,920 access the model, but how the model is built, how it operates, 567 00:26:57,920 - > 00:27:00,000 how it works is their secret source. 568 00:27:00,000 - > 00:27:00,400 Right. 569 00:27:00,400 - > 00:27:05,279 Um, so the the big West players have all gone down this route 570 00:27:05,279 - > 00:27:08,480 of we are going to keep our secrets to ourselves. 571 00:27:08,480 - > 00:27:12,319 Whereas China and a few of the other open source kind of 572 00:27:12,319 - > 00:27:14,960 leaders have gone the other way and said, actually, we think we 573 00:27:14,960 - > 00:27:18,559 will be able to go faster, quicker by making all of our 574 00:27:18,559 - > 00:27:21,519 models open source because we'll get everyone in the world to 575 00:27:21,519 - > 00:27:23,759 potentially contribute and improve them over time. 576 00:27:23,759 - > 00:27:28,640 So two very different approaches to um how AI is being 577 00:27:28,640 - > 00:27:30,720 developed and published. 578 00:27:30,720 - > 00:27:34,240 Now, like I said, for the moment, the private kind of 579 00:27:34,240 - > 00:27:36,720 approach appears to be leading the way. 580 00:27:36,720 - > 00:27:40,559 The models from OpenAI and Anthropic and Google are 581 00:27:40,559 - > 00:27:44,400 superior to the open source ones, but it's definitely one to 582 00:27:44,400 - > 00:27:48,160 watch as the open source ones just keep gaining momentum, you 583 00:27:48,160 - > 00:27:48,400 know. 584 00:27:48,400 - > 00:27:51,920 If you go back probably six months, they were further behind 585 00:27:51,920 - > 00:27:52,799 than they are now. 586 00:27:52,799 - > 00:27:55,519 So there's an element of them catching up. 587 00:27:55,519 - > 00:27:59,839 And it's it's part of a global AI arms race that's going on 588 00:27:59,839 - > 00:28:04,720 across, you know, the the main US labs and China mainly, around 589 00:28:04,720 - > 00:28:08,240 who can ship the best models quickest and get the adoption 590 00:28:08,240 - > 00:28:08,559 going. 591 00:28:09,039 - > 00:28:11,279 Sinéad Hammond: I've been looking at the news and seeing 592 00:28:11,279 - > 00:28:15,039 the the difference in the AI robots in China from 2025 to 593 00:28:15,039 - > 00:28:15,680 2026. 594 00:28:15,680 - > 00:28:18,240 I don't know if you saw it was a it was a festival they did and 595 00:28:18,240 - > 00:28:21,519 it is light and night and day, um, which is quite interesting. 596 00:28:21,599 - > 00:28:24,480 Mark Rotheram: So Yeah, the the robots bit is is extremely 597 00:28:24,480 - > 00:28:29,039 interesting because a few years ago all of the robots were 598 00:28:29,039 - > 00:28:32,880 designed to run using code effectively, and then most of 599 00:28:32,880 - > 00:28:34,960 them have pivoted to using neural nets. 600 00:28:34,960 - > 00:28:38,559 So effectively you've got an AI model in the brain of these 601 00:28:38,559 - > 00:28:41,839 robots now that's taking all those inputs and doing all the 602 00:28:41,839 - > 00:28:42,400 outputs. 603 00:28:42,400 - > 00:28:44,640 And yeah, it's it's really interesting to see that you 604 00:28:44,640 - > 00:28:47,759 know, once you've taught one robot one trick, all the robots 605 00:28:47,759 - > 00:28:48,400 get that trick. 606 00:28:48,400 - > 00:28:50,240 So you get instant scalability. 607 00:28:50,240 - > 00:28:54,880 So yeah, the whole robotics being led by AI is really 608 00:28:54,880 - > 00:28:57,519 interesting to watch, and that'll be definitely a the next 609 00:28:57,519 - > 00:29:00,480 disruption curve to come in the next uh probably two years or 610 00:29:00,480 - > 00:29:00,640 so. 611 00:29:00,960 - > 00:29:04,079 Sinéad Hammond: Gosh, lots and lots going on. 612 00:29:04,079 - > 00:29:08,160 Um, for businesses, I suppose, Rowan, probably not talking 613 00:29:08,160 - > 00:29:10,799 about robots here, I'm not sure we're quite there yet with that. 614 00:29:10,799 - > 00:29:15,599 But what are you doing in your team to kind of enable them and 615 00:29:15,599 - > 00:29:18,640 help them to kind of embrace all of this new stuff? 616 00:29:18,640 - > 00:29:22,880 I know that you work a lot with Mark on this, and Mark, you 617 00:29:22,880 - > 00:29:26,160 have so much access to all of this knowledge and all of this 618 00:29:26,160 - > 00:29:28,880 technical expertise and all of these different models and 619 00:29:28,880 - > 00:29:30,319 information and all of that sort of thing. 620 00:29:30,319 - > 00:29:34,799 Rowan, on your side, how as a department lead are you enabling 621 00:29:34,799 - > 00:29:39,119 your team to adopt this and to bring this technology in whilst 622 00:29:39,119 - > 00:29:40,960 you've got all of the other work that's going on? 623 00:29:41,279 - > 00:29:44,480 Rowan Gill: One of the um really big bonuses of working for a 624 00:29:44,480 - > 00:29:48,960 company like BCN is we do, as a marketing function, have access 625 00:29:48,960 - > 00:29:52,880 to amazing technical experts like Mark and the rest of our AI 626 00:29:52,880 - > 00:29:53,839 and innovation team. 627 00:29:53,839 - > 00:29:55,359 And that is absolutely brilliant. 628 00:29:55,359 - > 00:29:58,960 And we can see the art of the possible and everything that AI 629 00:29:58,960 - > 00:30:03,039 can help us do and how it can impact us as a marketing team to 630 00:30:03,039 - > 00:30:04,160 really accelerate. 631 00:30:04,160 - > 00:30:09,519 What I try and do is make sure that my team has time to look 632 00:30:09,519 - > 00:30:14,400 and evaluate the tools and then investigate AI and work out 633 00:30:14,400 - > 00:30:16,240 what's going to make the biggest difference. 634 00:30:16,240 - > 00:30:19,680 And having that space to evaluate everything is really, 635 00:30:19,680 - > 00:30:20,240 really key. 636 00:30:20,240 - > 00:30:23,759 And it allows us to sit in that room together and talk about 637 00:30:23,759 - > 00:30:26,559 well, actually, what are our challenges now? 638 00:30:26,559 - > 00:30:29,119 What are the everyday things that we're actually, if we could 639 00:30:29,119 - > 00:30:32,079 solve, that would make things better for us as a team. 640 00:30:32,079 - > 00:30:36,480 So it's really having the strategy in place and knowing 641 00:30:36,480 - > 00:30:41,039 what we're going to adopt when and why, but also we never want 642 00:30:41,039 - > 00:30:43,839 to stop hearing about these amazing things that are 643 00:30:43,839 - > 00:30:46,480 happening and the latest and greatest things that we can 644 00:30:46,480 - > 00:30:50,880 bring into the way that we work, but we have to bring some sort 645 00:30:50,880 - > 00:30:54,240 of calm to the chaos because things are changing so rapidly, 646 00:30:54,240 - > 00:30:55,759 as March has highlighted. 647 00:30:55,759 - > 00:30:59,200 It is really difficult to keep up, and especially within the 648 00:30:59,200 - > 00:31:00,079 marketing world. 649 00:31:00,079 - > 00:31:03,920 So it's it's having an open mind, being aware of what's 650 00:31:03,920 - > 00:31:06,799 happening, but then really bringing it back down to 651 00:31:06,799 - > 00:31:10,559 evaluate what will make the big difference and back to again the 652 00:31:10,559 - > 00:31:13,599 other point is it safe and is it secure, and can we move 653 00:31:13,599 - > 00:31:14,400 forward with it? 654 00:31:14,640 - > 00:31:15,119 Sinéad Hammond: Yeah, yes. 655 00:31:15,119 - > 00:31:17,440 I think if business leaders want to know what's happening in 656 00:31:17,440 - > 00:31:20,160 AI and how it applies to businesses, is this the right 657 00:31:20,160 - > 00:31:21,920 space for them then going forward? 658 00:31:21,920 - > 00:31:25,200 Just really quickly to finish off, Mark and Marin, both of 659 00:31:25,200 - > 00:31:26,960 you, I'm gonna ask, I'm gonna put you on the spot. 660 00:31:26,960 - > 00:31:31,200 Um, what is the one priority for an organization technology 661 00:31:31,200 - > 00:31:35,440 in terms of AI this sort of next 30 days that businesses should 662 00:31:35,440 - > 00:31:36,160 really be focusing on? 663 00:31:36,160 - > 00:31:37,839 I'm gonna start with you first, Mark. 664 00:31:38,240 - > 00:31:40,480 Mark Rotheram: I think it's ironic that you asked from a 665 00:31:40,480 - > 00:31:43,039 technical perspective, but the the best takeaway is the 666 00:31:43,039 - > 00:31:44,319 technology barrier is gone. 667 00:31:44,319 - > 00:31:44,720 Yeah. 668 00:31:44,720 - > 00:31:46,400 So we can do these things. 669 00:31:46,400 - > 00:31:50,720 The the here, right now, the ability is there. 670 00:31:50,720 - > 00:31:54,319 The barrier is organizational readiness. 671 00:31:54,319 - > 00:31:55,519 Are you ready? 672 00:31:55,519 - > 00:31:57,839 And if you're not ready, how are you going to get ready? 673 00:31:57,839 - > 00:31:58,880 So focus on that. 674 00:31:58,880 - > 00:32:01,599 Focus on do you have an AI policy? 675 00:32:01,599 - > 00:32:04,400 If not, get someone thinking about how to do that. 676 00:32:04,400 - > 00:32:08,000 And then that will lead you into adoption, and then that'll 677 00:32:08,000 - > 00:32:11,039 lead you into right, actually, what outcomes should we be 678 00:32:11,039 - > 00:32:12,319 focusing on that we do today? 679 00:32:12,319 - > 00:32:15,039 And then we get to the fun stuff of ah, I didn't know those 680 00:32:15,039 - > 00:32:17,440 were outcomes, I could even go after, let's go and work on 681 00:32:17,440 - > 00:32:17,680 those. 682 00:32:17,680 - > 00:32:20,720 But I think you know, just think about the technology 683 00:32:20,720 - > 00:32:22,720 barrier is effectively gone. 684 00:32:22,720 - > 00:32:25,759 How do you organise around the capability that's available to 685 00:32:25,759 - > 00:32:26,319 you today? 686 00:32:27,200 - > 00:32:27,759 Sinéad Hammond: And Rowan? 687 00:32:28,000 - > 00:32:31,359 Rowan Gill: Yeah, I think going back to the um to get to that 688 00:32:31,359 - > 00:32:35,759 place, you need the time to actually work out what you want 689 00:32:35,759 - > 00:32:37,519 to implement and how you're going to do it. 690 00:32:37,519 - > 00:32:41,680 So I would say sometimes it's hard to say stop, especially 691 00:32:41,680 - > 00:32:43,839 within some teams that are running quite quickly. 692 00:32:43,839 - > 00:32:47,839 But sometimes you need to stop and evaluate things to then move 693 00:32:47,839 - > 00:32:49,200 forward in the in the best way. 694 00:32:48,970 - > 00:32:50,799 Sinéad Hammond: Thank you very much. 695 00:32:50,799 - > 00:32:52,319 Really interesting conversation. 696 00:32:52,319 - > 00:32:54,319 Loads going on, as always. 697 00:32:54,319 - > 00:32:57,519 There's always new releases, news about AI. 698 00:32:57,519 - > 00:33:01,440 So if you want to get your regular debrief for AI news and 699 00:33:01,440 - > 00:33:03,920 business updates, then this is the place to be. 700 00:33:03,920 - > 00:33:06,720 We will be doing episodes like this regularly. 701 00:33:06,720 - > 00:33:11,359 So make sure you subscribe and visit bcn.co.uk to find out more 702 00:33:11,359 - > 00:33:14,480 about the services we offer around data and AI as well. 703 00:33:14,480 - > 00:33:17,519 Thank you very much, Rome, for joining us and for Mark joining 704 00:33:17,519 - > 00:33:17,920 us as well. 705 00:33:17,920 - > 00:33:19,200 It's been really nice chatting. 706 00:33:19,200 - > 00:33:22,400 I've learned a lot of new things about AI and new terms 707 00:33:22,400 - > 00:33:25,039 that I didn't know about and some exciting stuff that's 708 00:33:25,039 - > 00:33:25,839 coming up in the news. 709 00:33:25,839 - > 00:33:27,279 So, really, really good conversation. 710 00:33:27,279 - > 00:33:27,759 Thank you. 711 00:33:27,759 - > 00:33:28,720 Thanks for watching. 712 00:33:28,720 - > 00:33:31,680 Thanks for listening to AI for business. 713 00:33:31,680 - > 00:33:35,119 If you are interested in finding more out about our AI 714 00:33:35,119 - > 00:33:37,920 services, you're worried about shadow AI, you're trying to 715 00:33:37,920 - > 00:33:40,559 understand how to adopt AI better across your business, 716 00:33:40,559 - > 00:33:42,079 then get in touch at our website. 717 00:33:42,079 - > 00:33:45,440 We're at bcn.co.uk and we'd love to talk to you.