Trascrizione completa di Why AI Is Breaking Team Collaboration (And How to Fix It)

kind of stuff, and thinking that they were like, you know, at the, at the forefront of creativity. 00:04:02.585 --> 00:04:21.341 It's, I think it's quite easy to look back at that time and go, you know, "Oh my God," you know, "AI bashed into creativity and created this sort of nuclear blast." But actually, you know, the, the positive side for me has always been that what we saw there actually was people who don't have necessarily particular skills starting to be able to express their creativity in different ways, right? 00:04:21.446 --> 00:04:29.425 And there's an interesting, quite recent review that was done by a couple people at, at MIT, where they looked at some recent studies around creativity. 00:04:29.593 --> 00:04:33.723 It's called The Hidden Cost of aiAI-Assisted Creativity, if you wanna look it up. 00:04:33.742 --> 00:04:34.290 It's really good. 00:04:34.396 --> 00:04:41.298 And they found … They looked at these four studies, and they found that there's, there's quite a lot of evidence that AI sort of enhances individual creativity. 00:04:41.468 --> 00:04:56.242 So if you give people a creative task, and you let people talk through it with AI, you let people sort of brainstorm, you let people, perhaps people who aren't any good at sketching, start to sketch out their ideas and think visually where they haven't been able to do that before, you get some really interesting effects. 00:04:56.259 --> 00:04:58.026 So that's definitely something I've seen. 00:04:58.184 --> 00:05:14.687 On the other hand, 100%, as soon as that leaves an individual person, as soon as it enters like a workplace or a project where creativity is not there for its own sake, but it's there to act as an input or act as a part of a project or deliverable, you get some horrible effects. 00:05:14.699 --> 00:05:15.747 So you get the work slop. 00:05:15.747 --> 00:05:19.894 Work slop is, you know, the, the… I'm sure all your listeners will have, will have heard this term. 00:05:20.134 --> 00:05:26.966 So it's when you're like, "Hi, I need one idea for a brainstorm," or one idea for a, a, a project, creative project of some kind. 00:05:27.067 --> 00:05:35.434 And instead of that person giving you one idea, they're like, "Well, here's 700 that I've generated with my AI," and "Here's 20 pages for each one that are like visuals." That's lovely. 00:05:35.721 --> 00:05:46.505 In, in the previous work, we'd be like, "Wow, you put so much effort in." What it actually shows us, I think in this case, is people are just not spending the time and applying the creative filters that you need to apply, right? 00:05:46.507 --> 00:05:56.718 So whatever your creative management process is, most people, if they're managing creative processes, will have some kind of filter built in, whether it's like the double diamond or, or something like that, right, framework. 00:05:57.079 --> 00:06:00.331 That goes out the window just because there's so much velocity. 00:06:00.384 --> 00:06:02.149 There's so much ability to generate. 00:06:02.451 --> 00:06:02.975 So there's that. 00:06:02.975 --> 00:06:10.896 There's also the thing that you alluded to earlier, which we- I've seen, as you say, in my previous life as an AI consultant, when I've gone into organizations, I've seen this a lot. 00:06:11.036 --> 00:06:17.483 The most senior people often are prone to this, which is a kind of AI psychosis, because they're talking to an AI. 00:06:17.490 --> 00:06:18.970 The AI is feeding back, "This is amazing. 00:06:18.970 --> 00:06:19.485 This is clever. 00:06:19.485 --> 00:06:20.134 This is incredible. 00:06:20.141 --> 00:06:20.584 Keep going. 00:06:20.584 --> 00:06:20.928 Keep going. 00:06:20.928 --> 00:06:26.709 Keep going." And they're going down rabbit holes, and I think they know individually that they're not doing their best work necessarily. 00:06:26.764 --> 00:06:27.995 But it's, it's addictive. 00:06:28.339 --> 00:06:36.596 And what happens then when they bring it to a group is that sense of unease, the unhappiness they feel that they are… They know they're not doing it properly. 00:06:36.781 --> 00:06:38.930 They bring it to a group, and then they become super defensive. 00:06:39.204 --> 00:06:47.185 So you get this terrible effect where people take something away, do a job, and actually go off on a real tangent, maybe do something they're only 50% sure of. 00:06:47.192 --> 00:06:52.995 But when they come back to the group, they're so keen to make people know that this is still their work that they overcompensate. 00:06:53.175 --> 00:07:03.879 So you get this terrible effect where, you know, unless you are building spaces for collaboration that actually take into account the way people work with AI, all you end up doing is, is breaking teams. 00:07:03.880 --> 00:07:11.944 And I have seen, sadly, you know, a couple of occasions, genuine occasions, teams that worked really, really well together that have been like, it's like a divorce. 00:07:12.055 --> 00:07:14.483 Like the AI has broken their creative process. 00:07:14.555 --> 00:07:38.625 And so that's why, you know, I'm going to talk about it maybe that's the sort of what we're trying to solve with, with Amble, which is the product that we're building is to just reset that and say"We need to start thinking in different ways of collaborating and this, this can work well." And just on this last point, I know I've gone on for ages, but on the last point, another study which is, it's called the Cybernetic Teammate, and I, I think a lot of people have s- have seen it, so it's this one was I think a Harvard Business School survey. 00:07:38.858 --> 00:07:43.190 This was they, they basically went to Procter & Gamble and said, "Can we study your teams? 00:07:43.555 --> 00:07:59.564 And we'll study some people individually without AI, some people individually with AI, some teams without AI, and some teams with AI." And they found that the individual performance, so individuals with AI suddenly

bumped up on like knowledge tasks, so some creative, but mostly just sort of knowledge tasks. 00:07:59.829 --> 00:08:11.817 And they found that the teams with AI elements, they were the most variable, and they were the ones that, that you saw the, the biggest improvements at like the top 10% tier of quality of work. 00:08:12.132 --> 00:08:23.601 So there's definitely some evidence here that if you get it right, where you've got teams working with AI in the right way, you can produce like some, you know, work where the it's greater than sum of its parts, basically. 00:08:23.617 --> 00:08:26.694 It's just a question of how you do it, and you have to do it in a quite considered, deliberate way. 00:08:26.918 --> 00:08:27.641 I really like that. 00:08:27.653 --> 00:08:29.737 Honestly, there's so much I wanna unpack there. 00:08:29.891 --> 00:08:37.370 The first thing, though, is, that I love is like that sort of theme of democratization is, is sort of coming up and, and the reasons why we need to put some effort into it. 00:08:37.600 --> 00:08:42.702 Like I, I really like this notion that, you know, now you have the opportunity to be, to be creative. 00:08:42.758 --> 00:08:46.100 And from a democratization lens, you know, you see this in sports, right? 00:08:46.100 --> 00:08:52.052 You're like, "Okay, well, you know, folks who got good at hockey first had to be able to afford hockey gear." You know what I mean? 00:08:52.241 --> 00:08:55.610 And so like there's this sort of, "Okay, well, now we can all play hockey," sort of thing. 00:08:55.841 --> 00:08:59.897 And then the, the, the natural thing, but we're not all good at hockey immediately. 00:09:00.269 --> 00:09:01.875 So we need to still figure it out. 00:09:02.136 --> 00:09:10.359 You know, we still need to develop, you know, our, our taste and our ability to curate what comes out when we can suddenly play hockey, you know, at the NHL level or whatever, right? 00:09:10.596 --> 00:09:11.695 It's it's a whole different thing. 00:09:11.798 --> 00:09:12.365 You're totally right. 00:09:12.365 --> 00:09:17.261 And for me, so the doctorization, I'm, I'm so passionate about AI as an enabling technology. 00:09:17.298 --> 00:09:25.596 And you know, I have, like a lot of people, I suspect listening to this podcast, if you're in a creative space, you have probably some people around you who really, really do not like the idea of AI. 00:09:25.720 --> 00:09:27.590 So I always have to be like very aware of that. 00:09:27.594 --> 00:09:33.371 And, and my big, my big piece, and I really genuinely believe this, is that it changes, it levels the playing field in a lot of different ways. 00:09:33.693 --> 00:09:35.453 But, but… And the hockey analogy is perfect. 00:09:35.524 --> 00:09:57.923 The thing that I always look out for is it's not whether someone was necessarily gonna be great at a particular thing, but it's the point in the process where someone says"Let me show you what I mean." And that point is where if you are someone who come, is very naturally is good at drawing, good at putting this together, good at putting words together, you have this ability to basically persuade people that your idea is better than it is. 00:09:58.178 --> 00:09:58.899 Kind of, right? 00:09:58.899 --> 00:10:07.267 You're, you're a very persuasive person, or you can draw beautifully, and so the s- the scamps look amazing, and maybe it wasn't the best creative idea, but it was, like, the most beautifully expressed one. 00:10:07.384 --> 00:10:12.317 And, and that's fine, and that's good, and there's a place for that, and, and part, that is part of creativity, and I'm not knocking that. 00:10:12.619 --> 00:10:17.918 But it means that someone who couldn't, who couldn't do that, who was sitting there with an amazing idea but couldn't put the words together. 00:10:17.918 --> 00:10:25.395 Or maybe we're talking about someone who's like neurodivergent in some way, someone who's, is, you know, speaking a second language, like barriers of all kinds. 00:10:25.581 --> 00:10:34.311 That's the point where you can say, "Look, this levels the playing field." It means that if you get 10 people in the room and you give it a chance, you're more likely overall to get a great idea out of that room. 00:10:34.482 --> 00:10:39.672 And I think that's the optimistic opportunity here, taking your other point, that we still need to get there, right? 00:10:39.672 --> 00:10:59.673 We still need to get to this point where, you know, someone using AI doesn't, doesn't make them hyper defensive, doesn't make their, you know, team get into like a divorce situation you know, isn't turning people off and is a way that we can actually collaborate instead of this like murky I don't know, if I say I used AI, then people are gonna think it's bad and then I need to defend it. 00:10:59.996 --> 00:11:04.989 Or people love that I have 700 ideas, but you know, it, we're, we're not further ahead. 00:11:04.990 --> 00:11:06.335 This is just more stuff. 00:11:06.595 --> 00:11:09.144 And I honestly, I see it a lot even where I'm working right now, right? 00:11:09.144 --> 00:11:10.489 Where there is tension. 00:11:10.528 --> 00:11:18.359 There's tension around the like you know, even, even the people who have trained their LLMs to sound like them, to think like them, you know, they're putting information into it. 00:11:18.359 --> 00:11:48.135 They're, they're putting in a lot of energy to make sure that their ideas can come out faithfully through an LLM, but then everyone looks at it and goes, "Well, that looks like, you know… This person would never write full paragraphs, and now I'm getting full paragraphs that sound like them, but it's clearly AI, so now I'm annoyed because now I feel like this person didn't wanna take the time to give me feedback or to write their feedback to me, and I'm insulted." Meanwhile, the thing you're saying, which is maybe that person wasn't really good at formulating their feedback, and now

this is helping. 00:11:48.298 --> 00:11:49.945 This could actually be good. 00:11:50.376 --> 00:11:57.846 Not just, you know, not sort of getting past that sort of sting that, "Oh, they just threw my deliverable into an LLM and said, 'I don't know. 00:11:57.846 --> 00:12:01.743 What do you think?' And it came back, and then they copied and pasted it and sent it," is how it feels. 00:12:01.743 --> 00:12:07.945 But when you think about the, the level of effort and, you know, have some empathy for the person who is doing this, yeah, maybe they're really time poor. 00:12:07.971 --> 00:12:09.996 And it's like you got some feedback instead of none. 00:12:10.259 --> 00:12:14.620 Or maybe they're just, you know, really self-conscious about the way they communicate and they feel like this is better. 00:12:14.740 --> 00:12:21.168 Even if it's not quite them, they were like, "This is an improvement. People should like this." And either way, the, the intention there is actually, it's quite good. 00:12:21.221 --> 00:12:27.975 It's like I do care enough to provide feedback, and I wanna get it in a, in a format that people can understand in action. 00:12:28.225 --> 00:12:33.258 And me saying, "Make better, please. Make it pop," is probably, you know, not what they wanna hear either. 00:12:33.557 --> 00:12:44.634 So I, I think that's it's such an interesting stage and like of the, the human dynamic because of AI, not the AI dynamic necessarily, which, I mean, maybe is a good segue. 00:12:44.836 --> 00:13:09.634 You brought up Amble and this is clearly something you are, you're very passionate about, like taking the AI experience from the sort of like localized individual thing that supercharges the individual, like you said in the sort of HBR report- And taking it into sort of a team level where teams are working together, AI's in the room and no one's, you know, feeling necessarily self-conscious about using AI 'cause it's there. 00:13:09.666 --> 00:13:10.611 It's standing there. 00:13:10.614 --> 00:13:11.780 It's, it's helping out. 00:13:11.840 --> 00:13:32.323 You touched on this a bit, but I'm just wondering what was the thing that took you down this path to build an entire platform that is, you know, collaborative AI, and what's different about it you know, compared to having our own sort of private conversations with AI and now having, I guess, kind of like, you know, a group chat with AI, a meeting with AI? 00:13:32.389 --> 00:13:41.523 We're building Ample, so, the company that we, we started a few months ago, and, and we is me and a couple of, of, of old friends, so people I've worked with for 20 years or so. 00:13:41.798 --> 00:13:57.418 The more we talked, we realized we were seeing the same thing, and it, it is this sort of AI psychosis thing, this idea that people who are really good are kind of falling into bad habits, but also the, the democratization that we thought would happen in the workplace where everyone gets to contribute maybe a little more wasn't happening. 00:13:57.437 --> 00:14:09.914 So we started to look at that, and we thought, "Well, what's, what's a, a way that would work for us?" And we, we played around with a few things, actually, and one of the things we found really powerful was, was putting the AI kind of in the middle, so putting it where we could s- all see it. 00:14:10.087 --> 00:14:22.732 If you can see some- If you sat next to someone, and I used to do this quite a lot, if you sat next to someone and watched them use a normal you know, ChatGPT or Claude or something, you can quite quickly see that the people who are very good at their jobs use it in a more advanced way. 00:14:22.879 --> 00:14:23.566 I mean, that's obvious. 00:14:23.573 --> 00:14:28.285 Okay, so your prompt writing, the things you ask about and care about are gonna make a big difference. 00:14:28.559 --> 00:14:32.947 And that's why, you know, in, in software engineering AI has got so good at coding. 00:14:32.967 --> 00:14:37.695 Writing the code is to some extent, this, this will get me in trouble, to some extent a solved problem. 00:14:37.888 --> 00:14:38.216 You know? 00:14:38.258 --> 00:14:42.053 You know, it, it, it is no longer the case that you need to sit there and hand write these lines of code. 00:14:42.120 --> 00:14:50.125 But the difference between someone who's never, ever, ever coded and someone who's spent 20 years building software, they will just write totally different things. 00:14:50.187 --> 00:14:56.394 They, they are wr- They are essentially using the same tool in two s- such two completely different ways, they might as well be different tools. 00:14:56.563 --> 00:14:57.950 And that's also true of creativity. 00:14:57.951 --> 00:15:01.125 It's also true of, of well, lots of places where people use AI. 00:15:01.557 --> 00:15:09.711 So if you put AI in the middle, you get the benefits of that because you've got people who can use… who bring different things to the table, and they are collaborating together. 00:15:09.712 --> 00:15:17.856 They are basically, you know, this, this is oversimplifying it, but they're kind of co-writing prompts or they're, one of them says, "Here's something I want you to do," and the other person goes, "Oh, actually, hold on a second. 00:15:17.859 --> 00:15:19.033 Make sure you do this," whatever. 00:15:19.511 --> 00:15:21.793 That automatically raises the bar. 00:15:22.144 --> 00:15:32.211 But also because everyone's seen it happen no one gets defensive because everyone saw them use the AI and everyone's happy with that because that's, that's what it's there for. 00:15:32.339 --> 00:15:39.575 And equally, in some cases, you might want to say,"I don't think the AI's appropriate for this." Like we're writing, I wanna write something in my voice. 00:15:39.610 --> 00:15:41.075 I don't think the AI's appropriate for this. 00:15:41.241 --> 00:15:48.340 And that's a decision that you can state in a really useful, collaborative way, rather than just like

being, being, being obstinate, you know. 00:15:48.745 --> 00:15:50.385 So that definitely formed it. 00:15:50.385 --> 00:16:12.735 I think when we started looking at it as well, we were seeing this tendency towards like meat proxying, which is a horrible term, sorry, but the idea that people just sort of like act as, as sort of messengers between LLMs, and we're like, "That is silly." You know, that is that's a really silly way to build organizations where you've got this sort of wall of, you know, of human, and all it does is pass things forward. 00:16:12.748 --> 00:16:16.863 And you can get rid of that just by not needing it because you're putting the AI in the middle. 00:16:17.186 --> 00:16:17.799 So there's that. 00:16:17.816 --> 00:16:20.932 We then thought, okay, what… Well, that, that's great. 00:16:21.232 --> 00:16:22.102 That's a starting point. 00:16:22.102 --> 00:16:23.092 But does that solve it? 00:16:23.094 --> 00:16:23.692 And no, it doesn't. 00:16:23.693 --> 00:16:28.625 And there's a few things we then went on to think about, and one of those things is it needs to be proactive. 00:16:28.634 --> 00:16:30.389 It needs to know what the project is. 00:16:30.389 --> 00:16:31.764 It needs to know things. 00:16:31.917 --> 00:16:36.052 It needs to know what everyone should know, and it needs to be able to say, "Hold on a second. 00:16:36.195 --> 00:16:46.153 You could spend the next six months learning how to use AI for this project, or I can just give you some things that might be useful." So you've done the last three projects. 00:16:46.179 --> 00:16:49.861 You always leave writing the web copy to three days before it's needed. 00:16:50.083 --> 00:16:53.254 This has always caused a problem with knock on blah, blah, blah, blah. 00:16:53.628 --> 00:17:00.250 How about I do a draft, I, the AI, do you a draft now six months before it's needed, and you can just feed it in as you go. 00:17:00.619 --> 00:17:12.979 So rather than you having to be like, "I'm gonna use AI for this," it's able to say, "Hold on, I'm… I know all this stuff about project management and all this stuff about how to build the things you're trying to build or do the things you're trying to do. 00:17:13.336 --> 00:17:22.166 I'm not gonna run this for you, and that's not the right route, but I can absolutely give you some, some helpful suggestions during the process." So we looked at that. 00:17:22.173 --> 00:17:33.162 And then the other part of it is just, you know, that kind of coding the, the democratization of coding experience, which, which we, we really loved, and we're like,"Is that something useful?" We're like, "It definitely is something useful." And I can talk about a little bit. 00:17:33.193 --> 00:17:53.454 This is not super relevant, but being able to generate interfaces that teams can use without them having to buy software, suddenly that totally boosted everyone because you can have an interface where people don't have to learn how to use, say, Excel or whatever it is, or Photoshop, but you can just boil it down to what the team needs, put that in front of everyone, and then it becomes collaborative. 00:17:53.721 --> 00:17:57.401 'Cause again, you're breaking down those boundaries and you're getting, you're democratizing the process. 00:17:57.678 --> 00:17:58.204 I love that. 00:17:58.240 --> 00:17:59.830 That's so cool, and it's blowing my mind. 00:17:59.839 --> 00:18:07.049 And, you know, I, I, I'm not gonna lie, when you started, I was picturing this like, you know, the Polycom in the conference room in the middle, right? 00:18:07.050 --> 00:18:11.125 You know, AI's in the middle, it's just standing on the table, and it will pipe up sometimes. 00:18:11.380 --> 00:18:16.736 You know, we'll ask it things, but you know, fundamentally it is, it is your… It's just a tool w- that's listening. 00:18:17.032 --> 00:18:22.951 And then as you went, I was like, "Wait a minute, this is, this is different." A, I really like that sort of pair programming mindset. 00:18:22.951 --> 00:18:30.150 It's like, well, if we are prompting together or asking together or, you know, creating together, we're learning from one another, e- each other as humans. 00:18:30.183 --> 00:18:37.457 And also, you know, taking away this sort of layer of trying to have to figure out AI alone, I guess, is the other thing. 00:18:37.616 --> 00:18:48.310 I, I, I had this thing in my head where I'm like, oh, y- you know, sometimes I say, I'm like,"AI is, you know, the big thing right now, but eventually it'll be as commonplace as typing." And then I'm like Oh, what if I'm a slow typist in the room? 00:18:48.635 --> 00:18:52.859 Am I-- What if I'm, like, showing that I'm bad at prompting as, as feeding into that psychosis. 00:18:52.872 --> 00:18:54.153 But I, I won't go there right now. 00:18:54.153 --> 00:18:59.002 But I- what I really like is that notion that it's it's not even just like a teammate necessarily. 00:18:59.003 --> 00:18:59.901 It's definitely not just a tool. 00:18:59.901 --> 00:19:07.122 But the interface thing you said, you know, it blows my mind because, you know, how many times have I spent 15 minutes looking through the menus of whatever. 00:19:07.294 --> 00:19:08.778 I do some light video editing. 00:19:08.779 --> 00:19:17.282 I use DaVinci Resolve which is could do much heavier duty editing, and I'm looking for just that one feature to whatever, crop a circle, right? 00:19:17.394 --> 00:19:23.508 You know, I'm like, gosh, if only this interface was just dumbed down to what I need, then I would just get on with it. 00:19:23.539 --> 00:19:25.757 And I, I love that idea that that could be that thing. 00:19:25.844 --> 00:19:33.846 I was just gonna ask you if you could walk us through what the experience is like of having AI participate in a collaborative session with the team. 00:19:33.929 --> 00:19:42.094 You painted a picture of how it could sort of, be helpful or intervene but I'm just wondering, you know, what's a

scenario where it can help the ideation flow instead of I don't know, getting, getting in the way. 00:19:42.132 --> 00:19:43.113 It… And it can get in the way. 00:19:43.115 --> 00:19:45.380 Let-- You know, just be, be, be honest about it as well. 00:19:45.380 --> 00:19:49.763 You know, there's no part of this which is let's put AI in there and it will just magically make things better. 00:19:49.803 --> 00:19:57.357 And I think it's interesting, if you look at things like OpenClaw and Hermes and Claude Tag, which are sort of… They're, they are a similar principle. 00:19:57.365 --> 00:19:58.820 Here's an AI that we can put somewhere. 00:19:59.002 --> 00:20:04.256 In raw forms, they can actually really… Well, nothing worse than they can distract people. 00:20:04.444 --> 00:20:10.174 And particularly, you know, when we did some initial tests with, I'll talk about brainstorms because it's quite an interesting one. 00:20:10.175 --> 00:20:22.477 But if, if you, if you do a brainstorm and you say,"Let's bring an AI into a brainstorm," and you just say "Come be a brainstorm partner," what happens when people start sort of using it is they sort of defer to it, which is just what you don't want in a brainstorm, right? 00:20:22.708 --> 00:20:27.498 And part of that is the automation bias part, is the people sort of, you know, demurring to the, the AI. 00:20:27.733 --> 00:20:30.373 But also it's, it's just bad user experience. 00:20:30.701 --> 00:20:39.348 AI has still, when it's voice-based, a kind of a lag, and there's always a sort of… You'll see this when you u- anyone using AI, they're like, "Am I using it right? 00:20:39.615 --> 00:20:40.595 Is this correct? 00:20:40.606 --> 00:20:41.529 Is this… Am I doing it? 00:20:41.633 --> 00:20:42.135 Oh, oh wait. 00:20:42.190 --> 00:20:45.772 Oh, is it, is it gonna… Has it got it?" You know, all that kind of stuff. 00:20:46.028 --> 00:20:47.997 So anyway, so let me take you through, like, how it could work then. 00:20:48.001 --> 00:20:54.515 So, we very early on we were like, "Okay let's… It's all about eliminating friction, colla-collaboration," blah, blah, blah. 00:20:54.526 --> 00:20:59.804 And we started talking about this and we went, "Hold on a second. That's just wrong, isn't it?" It's not actually about c- eliminating friction. 00:20:59.806 --> 00:21:03.993 There's loads of friction you do wanna remove you know, finding a time in people's diaries, sure, whatever. 00:21:04.310 --> 00:21:08.551 But actually a lot of creativity comes from friction of some kind. 00:21:08.615 --> 00:21:17.134 People disagreeing and then the disagreement leads to a really good idea, or people getting something wrong and the wrong thing turns out to be actually what, what you want in the end. 00:21:17.144 --> 00:21:24.707 All of these things are like, you know… The, the, the sort of using AI just to smooth things, so coordination friction, it turned out to be the wrong route. 00:21:25.002 --> 00:21:38.439 So we were like, "Okay,

when we go through collaborating, we need to make sure that when it's, when it needs to reduce friction, it can, and when it needs to up the friction or turn the friction into something useful, it can do that as well." So let's take brainstorms as a good example of this. 00:21:38.682 --> 00:21:41.959 So we built a few brainstorming tools. 00:21:41.961 --> 00:21:44.711 We did some experiments and we discovered loads. 00:21:44.750 --> 00:21:53.058 We discovered, like I mentioned earlier, that people like defer to it we discovered that people generally have bad ideas what brainstorms are and what they're used for. 00:21:53.179 --> 00:21:56.722 So there's kind of four like learnings, I guess, that we came from. 00:21:57.074 --> 00:22:00.903 One was like, you know, there's plenty of evidence anyway that the brainstorms are not great. 00:22:01.055 --> 00:22:06.161 And actually let people come up with their own ideas separately creates better results, but then coming together to share them. 00:22:06.162 --> 00:22:17.741 So we're like, okay, let's make it so that if the AI is seeing what everyone's doing for the last few weeks or months, whatever, it should be able to pull some good ideas out of that and put them in the brainstorm to get the conversation started. 00:22:17.979 --> 00:22:20.508 And then we were like, okay, no bad ideas in a brainstorm. 00:22:20.525 --> 00:22:21.376 Let's interrogate that. 00:22:21.376 --> 00:22:28.563 Well, it turns out actually that's a really bad way to run a brainstorm because all you do is you get people just agreeing at the start, and it's really interesting. 00:22:28.598 --> 00:22:36.842 I was at a talk, a lady called Cathy Peach at Nesta in, in the UK, and she was saying that when she runs… She's about the collaborative intelligence. 00:22:37.270 --> 00:22:41.152 When she runs sessions, she asks room-splitting questions at the beginning. 00:22:41.332 --> 00:22:44.430 And those are questions where she, she knows that the room will split. 00:22:44.778 --> 00:22:46.881 And basically what they do is they ask these questions. 00:22:47.123 --> 00:22:49.320 If they get consensus, they just get rid of the question. 00:22:49.455 --> 00:22:50.284 I'm oversimplifying. 00:22:50.288 --> 00:22:51.228 Sorry, Cathy, if you're listening. 00:22:51.473 --> 00:22:54.736 But they got get rid of, because there's no point asking a room of people to agree on something. 00:22:54.808 --> 00:22:57.403 You only really want to ask them where you're gonna get diversity of thought. 00:22:57.698 --> 00:23:00.118 So getting to a point where you're splitting the room is quite important. 00:23:00.396 --> 00:23:02.856 So that was another idea, like getting the bad ideas out. 00:23:02.856 --> 00:23:07.976 And one of the ways that we came up with for AI to help with that is we already know a lot about the project. 00:23:07.976 --> 00:23:09.770 We know about the audiences for the project. 00:23:09.773 --> 00:23:15.299 So if you're designing a label for a, a new oat milk or something, you know who your audiences are. 00:23:15.431 --> 00:23:16.344 So why not bring them in? 00:23:16.355 --> 00:23:22.526 Bring them in as like digital twins or personas, whatever, and have them there just gently voting on ideas. 00:23:22.684 --> 00:23:33.295 So if all three of your personas don't love what you're talking about, you know, the AI doesn't shut you down or press a big buzzer, but you do… You can see like it's getting down votes. 00:23:33.396 --> 00:23:36.019 And because it's the AI doing it, you don't take it as personally. 00:23:36.029 --> 00:23:51.298 And that's a, a big finding that we found is that if you get just even the three of us talking about what to name our company, things like that, actually having the AI as a, not a judge, but as a sort of a, a facilitator to keep track of things means that you end up offending each other less basically. 00:23:51.324 --> 00:23:51.933 Or not offending each other. 00:23:51.933 --> 00:23:56.184 It's fine to offend each other, but you end up being able to sort of de-risk the conversation. 00:23:56.464 --> 00:24:03.765 So this, this, this whole idea of like psychological safety is something that we've been trying to harness basically. 00:24:04.113 --> 00:24:09.156 And it's interesting, there's something called the Habermas Machine, which i- if you've not seen it, you should have a look at it. 00:24:09.167 --> 00:24:29.468 So it's just a, an experiment that some people put together, which is basically when you're doing negotiations or you've got like a contentious concept, rather than have someone facilitate, the AI sits there and kind of rewrites the same statement, you know, to allow for all the different people in the room, and it's much more effective at producing a consensus than a human is. 00:24:29.469 --> 00:24:36.982 It, it's just human nature, I guess, just psychological, but, but that's a good way that you can use the kind of AI psychologist idea to, to, to get there basically. 00:24:37.144 --> 00:24:39.579 So there's loads, I mean- There, there's a whole bunch of things. 00:24:39.579 --> 00:24:42.874 So the, the brainstorming tool we built builds on all of these and, and you join it. 00:24:43.164 --> 00:24:45.280 As you chat, it just, it doesn't ever talk to you. 00:24:45.762 --> 00:24:54.444 It just puts Post-its up as you say things, puts them into categories, does what a, like a, a someone leading the, the, the process would have to do normally. 00:24:54.663 --> 00:24:56.644 Keeps everyone focused on the actual task. 00:24:56.951 --> 00:25:00.704 Maybe comes up and says that you've only got three ideas, gotta move on, that sort of thing. 00:25:01.144 --> 00:25:05.199 And that sort of facilitating approach has been r- really effective and really fun as well. 00:25:05.416 --> 00:25:05.973 People love it. 00:25:06.269 --> 00:25:06.769 I like that. 00:25:06.769 --> 00:25:10.152 And it's, you know, it is the sort of paradox of a human facilitator,

right? 00:25:10.153 --> 00:25:22.510 You're trying to play that role of someone neutral and, you know, quasi objective and making people feel heard, but also, you know, making sure that not every, you know, idea makes it to the final cut. 00:25:22.540 --> 00:25:23.866 And it, it is, it's political. 00:25:23.866 --> 00:25:28.122 It's so funny when you're talking about these sort of layers of… I'm gonna frame it as layers of psychosis, right? 00:25:28.122 --> 00:25:31.801 We're talking about the psychosis of okay, well, am I using AI well enough? 00:25:31.854 --> 00:25:36.180 The psychosis of, well, AI just told me that these are great ideas, all 700 of them. 00:25:36.231 --> 00:25:38.806 I, you know, it said it in such a way that maybe I should believe them. 00:25:38.900 --> 00:25:52.705 There's the like, oh, us working together and, you know, trying not to, you know, hurt each other's feelings in an unproductive way but still introduce enough friction that we're not just nodding at everything and that, you know, we're not making decisions. 00:25:53.013 --> 00:26:22.128 And I really like that idea that, okay, well, actually, you know, even scientifically so far or at least the, the research is showing and your research is showing that in, as a fac- as a facilitator role, you know, AI can be quite effective at just kind of keeping us on task and formulating our ideas without necessarily removing friction, and ideally, without necessarily adding more friction than necessary Outside of our own sort of individual or cultural psychosis and attitudes towards AI. 00:26:22.709 --> 00:26:23.930 Yeah, I think that's it exactly. 00:26:23.950 --> 00:26:29.923 I mean, and worth saying, I think, you know, there is very much still a role for human facilitators in this process. 00:26:29.924 --> 00:26:37.241 But the reality is in most day-to-day workplaces, you don't have trained facilitators who are gonna be effective at moving this stuff on. 00:26:37.601 --> 00:26:45.824 So it's another, it's another way you can kind of bring some of the benefits of that, I guess, using technology rather than having to re-engineer your whole processes in a way that will never really happen. 00:26:46.480 --> 00:26:57.026 Yeah, and I, I love that for my audience as well because, you know, in my community of, you know, digital project managers I've been saying and others have been saying, we're like,"Well, you know, facilitation skills are gonna be really important right now." You know? 00:26:57.041 --> 00:27:03.678 You know, instead of the things that are getting automated this is a very human thing that we can do that helps drive towards goals. 00:27:03.895 --> 00:27:06.378 It is in some ways management of a, of a project. 00:27:06.380 --> 00:27:07.352 It's leading us through. 00:27:07.567 --> 00:27:10.548 But we're meant to be q- quasi-neutral as well, right? 00:27:10.548 --> 00:27:13.621 We're not like shot callers necessarily at the project level. 00:27:13.873 --> 00:27:16.419

We've got stakeholders in the room, and we should be good at that. 00:27:16.549 --> 00:27:21.762 But I also like the flip of but if someone's taking care of that, then we can be contributing in a completely different way. 00:27:21.763 --> 00:27:23.348 And I think that's true of anyone in any craft. 00:27:23.538 --> 00:27:26.092 If you're, if you're not taking notes, you can participate. 00:27:26.385 --> 00:27:30.983 And then the question is, okay, well, what are you bringing to the table when you're participating if you're not taking notes? 00:27:30.983 --> 00:27:38.122 And I think that's it's an interesting question that gets into how a, a team collaborates which allows everyone to be more present. 00:27:38.123 --> 00:27:40.540 I think it's, yeah, I think it's, I think it's fantastic. 00:27:40.807 --> 00:27:46.462 You had mentioned in its current form, at least your brainstorm sort of, platform, the AI doesn't speak. 00:27:46.580 --> 00:27:50.074 Earlier we were talking about maybe it will come in with suggestions and, you know. 00:27:50.275 --> 00:27:59.351 And you had mentioned, right, quite rightly that the tone, right, the, the sort of ethos of a lot of these LLMs is convincing or authoritative enough that we kind of defer to it. 00:27:59.352 --> 00:28:00.227 We're a bit demure. 00:28:00.246 --> 00:28:07.905 We're like, "Okay, well, you clearly have trained on everything available on the internet, and I have only read, you know, seven books in the last, you know, as many years. 00:28:08.125 --> 00:28:16.750 So I guess AI, you're right." But I'm wondering as you've been sort of creating these experiences and, and when AI is talking out loud has this introduced, like, more friction? 00:28:16.801 --> 00:28:20.954 As you've been building this out, has anyone sort of like gotten in a fight with AI? 00:28:21.181 --> 00:28:23.583 And if so, like, how did it go down? 00:28:23.634 --> 00:28:24.576 And, and who won? 00:28:26.188 --> 00:28:27.356 I mean, 100%, yeah. 00:28:27.357 --> 00:28:34.625 So we have so for the brainstorm tool in its current format that we're, that we're playing with, the AI doesn't speak for those other reason. 00:28:34.986 --> 00:28:40.084 We have a separate, a different tool that we, within our product that we've been working on, which is all around briefings. 00:28:40.156 --> 00:28:45.565 So your audience will know full well the importance of a, a good briefing, right, in a creative process. 00:28:45.801 --> 00:28:48.451 So we, we thought, "Okay, this is really important. 00:28:48.471 --> 00:28:51.352 We've, we've worked in these industries for years and years and years. 00:28:51.352 --> 00:28:52.483 We know how important briefing is. 00:28:52.483 --> 00:28:53.487 People often get it wrong. 00:28:53.632 --> 00:29:15.371 Let's just have the AI take your briefing." So we set it up and we said, "Okay, you're gonna chat to a person, you're gonna ask them these questions, and while you're doing it, you're gonna draw a mind map so

they can easily see where you've got to, and they can correct you." So that kind of worked okay But oh my gosh, the AI did, there was something about the way it was asking that was just incredibly annoying. 00:29:15.632 --> 00:29:18.212 So when any… And it, and I, I don't know what it is. 00:29:18.218 --> 00:29:20.122 It wasn't being aggressive, obviously, 'cause it's AI. 00:29:20.144 --> 00:29:28.273 Maybe it was being too unaggressive, but it was, it was sort of fi- fixating on the points that didn't matter, and it was going,"Okay, what about this person? 00:29:28.273 --> 00:29:35.029 What are they doing?" You're like, "Well, I don't, I don't, I don't wanna tell you about that." People felt they were being sort of led by it, which it should be the opposite way around, right? 00:29:35.279 --> 00:29:36.414 And you, I mean, you'll know, right? 00:29:36.445 --> 00:29:39.708 You, you know, the, the skill of taking good briefing is a skill. 00:29:39.953 --> 00:29:50.571 So yeah, we did have a few occasions on that, on that front where people were, you know, not sort of shouting at it, but getting pretty, pretty pissed off with the AI just sort of wasting their time, and I think that's the main thing. 00:29:50.673 --> 00:30:01.513 People are so quick to move from, you know, the first two or three seconds using any of these tools if people haven't encountered them before, they're like, "This is magic." But it really doesn't take long before they're starting to go like, "Well, why, why can't it do this? 00:30:01.528 --> 00:30:03.305 Why is it, why is it asking me this three times? 00:30:03.305 --> 00:30:06.609 Why doesn't it, why does it keep misspelling this person's name?" Whatever it might be. 00:30:06.902 --> 00:30:09.286 So yeah, no, we definitely have had that, that situation. 00:30:09.294 --> 00:30:13.654 And it, you know, it's interesting to watch people use, particularly our tool, but, but obviously other tools. 00:30:13.654 --> 00:30:30.375 And when I was, when I was doing implementation, you would have people who… And actually I used to sit, so in one of the offices I used to work with, work in, I used to sit at a desk here, and three desks away there was a, a CFO and three desks away there was one of one of the directors of this company. 00:30:30.567 --> 00:30:33.212 Catherine and Patrick hopefully they won't mind me using their real names. 00:30:33.444 --> 00:30:34.903 But Catherine was all about being nice to the AI. 00:30:35.095 --> 00:30:40.217 So she would always say,"Oh, thank you. Okay." And she would have a full-on conversation with it before she asked her first question. 00:30:40.224 --> 00:30:42.691 She would say, "How are you doing? How's it going? 00:30:42.692 --> 00:30:44.421 What are you…" You know, and discuss things with it. 00:30:44.423 --> 00:30:45.643 And then she would say, "All right. 00:30:45.653 --> 00:30:52.307 The thing I'd like to talk to you about today, copilot, is this." Patrick would be the entire opposite. He'd

be like, "You stupid machine. 00:30:52.310 --> 00:30:56.929 You've got this wrong again." And I was sitting there. I was like, "This is a really good test. 00:30:56.965 --> 00:31:00.086 This is a really interesting case," 'cause who's getting the best answers? 00:31:00.410 --> 00:31:02.615 And actually, interestingly, it varied. 00:31:02.673 --> 00:31:13.134 There were definitely points where a really sort of quite aggressive approach meant you were being direct, you were being clear, you were stating your point, you were expressing why you were in control, and that worked really well. 00:31:13.327 --> 00:31:19.996 There were some points when being friendly and nice and, and, and gentle with it led to a more human response for copywriting. 00:31:20.017 --> 00:31:20.991 So it's actually interesting. 00:31:20.992 --> 00:31:25.760 I think there actually, oddly enough, may be an argument for occasionally getting into a fight with your AI. 00:31:25.785 --> 00:31:27.884 That's not necessarily a bad thing, if you know what I mean. 00:31:28.045 --> 00:31:30.796 I think that's f- I mean, honestly, that's fair as, as I think about it. 00:31:30.796 --> 00:31:46.289 So I mean my, my biggest corner turn as I was sort of starting to use AI, like LLMs specifically in my day-to-day, was like, "Okay, just pretend you're in Slack chatting with your teammate." Not, I don't, you know, not to the extent that Catherine was you know, having a conversation you know, I, I wouldn't ask it about the weather. 00:31:46.511 --> 00:31:50.746 But I would sort of address it like a teammate and like any, you know, good teammate. 00:31:50.903 --> 00:31:55.267 The collaboration element of being a human, right, is, is up and down, right? 00:31:55.268 --> 00:32:02.066 It's not like we're always gonna have good conversations where we're pleasant to one another, and sometimes I do need to be more direct with one of my employees, right? 00:32:02.066 --> 00:32:05.826 To, to kind of make sure I'm, I'm, you know, they're, they're staying on, on course, right? 00:32:05.837 --> 00:32:06.621 Not getting distracted. 00:32:06.621 --> 00:32:10.431 And these things that we do to sort of adjust the way we work together. 00:32:10.528 --> 00:32:13.542 And yeah, they… You know, in, in reality it's not always one or the other. 00:32:13.593 --> 00:32:15.664 Definitely people have styles, right? 00:32:15.673 --> 00:32:20.203 Someone might have a more direct style, but doesn't mean that every conversation is going to feel like you're being yelled at. 00:32:20.318 --> 00:32:25.910 As I say that I'm like, "Okay, I know a few people who, you know, maybe that's not so true." It always feels like you're getting yelled at by them. 00:32:25.911 --> 00:32:36.258 But you know, I think there's nuance to the human experience and, and how we experience other humans that, yeah, sometimes don't show up in the same way when we're training AI. 00:32:36.271 --> 00:32:39.448 Because AI's like, "Okay, I'm learning about you and I've put you in a box. 00:32:39.448 --> 00:32:45.644 I think I've figured you out now, and I'm only gonna give you that." But i- with the human experience like it, it might be the case. 00:32:45.718 --> 00:32:50.902 Yes, I've seen, you know, what you said briefing is an art, and yes, I've definitely seen that with humans and AI. 00:32:50.931 --> 00:32:57.761 I've gotten frustrated with AI because it's like asking me these verbose questions, like going down this other rabbit hole that I don't wanna go down right now, like you said. 00:32:57.999 --> 00:33:00.066 And also I've seen humans do that too, right? 00:33:00.281 --> 00:33:07.364 And they're like, a, a, a good human will be able to read the room and be like, "Okay, I'm going the wrong direction. We'll circle back." You know, that's what the parking lot is for. 00:33:07.364 --> 00:33:18.795 All those other, you know, semantic things we say at work to be like, "Ooh, okay, well, we'll ta- we'll talk about that later because clearly you don't wanna talk about it now." And it's interesting that you know, we think of it as so binary for AI. 00:33:18.906 --> 00:33:25.521 Not getting down the rabbit hole with the anthropic, you know, is AI conscious and does it have feelings thing, but yeah, it's like it's, it's also nuanced. 00:33:25.630 --> 00:33:26.559 That, that's really interesting. 00:33:26.560 --> 00:33:28.557 That's a really, really good point and I, I like that. 00:33:28.606 --> 00:33:32.391 Being able to take a breath is something humans can do and, and AI can't as well. 00:33:32.402 --> 00:33:44.778 So if you go to a colleague and you say "I think this," and they disagree, you can both take a step back, reevaluate your thoughts and come back and, and have a… You can agree with each other or disagree or whatever or come up with a third route. 00:33:45.218 --> 00:33:47.464 The AI doesn't do that, but it acts like it might. 00:33:47.848 --> 00:33:56.378 So I think it can be quite dangerous I guess to have a situation where you've got you know, a conversation with an AI where you're o- you're overly convinced it's a human. 00:33:56.698 --> 00:33:57.414 It's very interesting. 00:33:57.445 --> 00:34:06.323 I think something that's come up a lot with the latest anthropic models, which are super sophisticated obviously, is that previously if you sort of said, particularly when you're coding, and I've had this. 00:34:06.398 --> 00:34:09.152 When you're coding and you say,"No, I don't wanna do that. 00:34:09.704 --> 00:34:14.855 Let's take a different route and we'll do this instead," less sophisticated models would say, "Okay, this is fine. 00:34:14.855 --> 00:34:15.567 I take this. 00:34:15.567 --> 00:34:44.436 We will then do what you've said." Latest kind of, particularly the, the latest papal and, and 8+ models, I know plenty of people who have had the situation where they, they've sort of said, "No, we're not gonna do this," and the AI's kind of- Agreed but done it anyway, or agreed

but said,"You're kind of wrong, so I'm gonna, I'm gonna take a third route that we haven't agreed." I think the, the model of treat it like a, used to be treat it like a, an intern or treat it like a, a colleague who does, just onboarding onto a project, whatever that is, I don't think that works as well as it used to anymore. 00:34:44.436 --> 00:34:52.528 I think we need to develop a new way of talking to our AIs and, and, and understanding what kind of things they're gonna listen to and what kind of things they need more convincing about. 00:34:52.739 --> 00:34:55.282 It's a hugely good point that this is a moving target, right? 00:34:55.282 --> 00:35:09.135 We, we… It takes us so long to be like, "This is the way we should use AI," and by the time we get there, the models are, like, seven steps ahead and it doesn't apply anymore, and we have to kind of, you know… In the process of us, you know, we're talking about this sort of psychosis, right? 00:35:09.135 --> 00:35:15.856 And everyone kind of needs to, to get there culturally you know, within their team environment of how we actually feel with AI, about AI, and how we work with it. 00:35:16.023 --> 00:35:18.516 Meanwhile, it's still moving forward, right? 00:35:18.527 --> 00:35:22.443 So by the time we all agree, and we're like, "Yeah, okay, this is how we're gonna use it. 00:35:22.458 --> 00:35:23.320 We've written it down. 00:35:23.320 --> 00:35:28.177 We've got our sort of, you know, our, our model or our charter." Oh yeah, it's, it's, it's already different. 00:35:28.226 --> 00:35:33.043 And we need, we need to have that in our minds that it's, you know, it is going to continue to be different. 00:35:33.286 --> 00:35:37.875 But also, you know, to your point arguably getting more human characteristics. 00:35:38.143 --> 00:35:47.447 Like, how many times have, have I asked someone to do something, we're like, "No, no, no, we're gonna go this direction," and they're like,"Okay," and then they'll do the, the third option that we hadn't talked about because, you know, they have conviction. 00:35:47.447 --> 00:35:48.824 They think it's the right thing to do. 00:35:48.958 --> 00:35:54.543 And so might our tools now, and it's this interesting quandary where it's like, is it a tool? 00:35:54.572 --> 00:35:55.585 Is it a teammate? 00:35:55.623 --> 00:35:56.630 Is it an intern? 00:35:56.655 --> 00:35:57.596 Is it an expert? 00:35:57.630 --> 00:36:06.646 Is it, you know… And then it, it, it's very easy to fall into that hole of "Okay, well, I guess we better just believe it because, you know, maybe it knows better." 00:36:06.947 --> 00:36:07.757 It, it might do. 00:36:07.806 --> 00:36:08.454 It might know better. 00:36:08.454 --> 00:36:08.927 That's the thing. 00:36:08.929 --> 00:36:20.663 You know, it, in… Again, in software engineering, which I think is where it's, it's most obvious at the moment, a lot of software engineering is about Understanding trade-offs between capability and risk. 00:36:20.813 --> 00:36:26.891 Should I open this surface up it, which enables all these capabilities, but obviously gives it another surface for attacks or whatever it might be. 00:36:27.255 --> 00:36:34.221 Your AI will, will, will advise you, and you see this with a lot of people who don't normally develop software starting to vibe code. 00:36:34.340 --> 00:36:43.894 The AI will, will talk them into this incredibly sophisticated security setup, for example, which for something that may never leave their computer. 00:36:44.094 --> 00:36:48.092 And on one hand, that's clearly ridiculous, blah, blah, blah, and it wastes huge of time. 00:36:48.286 --> 00:36:53.210 On the other hand, the moment they do then want to put it live in some form, the AI was right. 00:36:53.289 --> 00:37:03.989 So it's very hard to you, you don't wanna totally close yourself off and say,"We're in charge." Because one of the benefits to these systems, they do have all this embodied knowledge, and that's genuinely useful for people. 00:37:04.146 --> 00:37:07.391 I honestly, I think it's like the context judgment, I guess. 00:37:07.418 --> 00:37:13.722 I don't know how else to phrase it, but, you know, and in some cases I'm like,"Yeah, yeah, AI, but like you don't know all the context. 00:37:13.724 --> 00:37:15.199 Like I know this person. 00:37:15.199 --> 00:37:16.481 I've worked with them for 10 years. 00:37:16.672 --> 00:37:18.463 You know, I'm not gonna train you on that. 00:37:18.472 --> 00:37:30.867 It's gonna take too long." I'm like, "This is, this is just, you know, from my context, I, I think I have more context on this to make the call than you do," versus, "Yeah, I'm not a cybersecurity expert, so actually, you know what? 00:37:31.123 --> 00:37:33.320 That's probably fair." And we do this with humans too, right? 00:37:33.331 --> 00:37:37.239 It's like, okay, well, you know, listen, like like you and I have probably got it a lot, right? 00:37:37.262 --> 00:37:38.045 "Oh, you're the creative. 00:37:38.045 --> 00:37:38.775 You're just the creative. 00:37:38.969 --> 00:37:39.851 You're just the project manager. 00:37:39.851 --> 00:37:42.054 You don't have all the context here." And, and maybe we don't, right? 00:37:42.056 --> 00:37:51.527 And again, we have to sort of be able to sort of, you know, take a step back as well based on how much knowledge and, and, and the quality of the judgment of the person who is making the call. 00:37:51.659 --> 00:38:01.728 We won't always agree with it and it could still go sideways, but like that's, I think how at least how I look at it is okay, well yeah, you've, you've seen this a lot and yeah, I trust your judgment because you've seen this a lot. 00:38:02.041 --> 00:38:04.690 And AI has seen a lot because it's trained on a lot. 00:38:04.967 --> 00:38:07.119 But you know, it's trained on a lot of information. 00:38:07.501 --> 00:38:10.802 So security is a great one versus stakeholder politics. 00:38:10.902 --> 00:38:12.327 They

might not be able to make that call. 00:38:12.347 --> 00:38:25.550 But I, I mean, that all being said, I do love that idea of like having, you know, this sort of synthetic audience or like a, you know, the avatars of people who, you know, are stakeholders and what they might say, 'cause I think that's that's just good customer experience, you know? 00:38:25.561 --> 00:38:33.315 Like user-led thinking, you know, not just my idea's great and who cares what the end recipient of this, you know, cares about. 00:38:33.358 --> 00:38:37.585 But it's it, yeah, it's a really interesting one to sort of navigate that. 00:38:37.663 --> 00:38:48.067 I, I wonder, I mean- This is sort of related, but I did wanna get in there because we were talking about AI sort of facilitating, it's a good facilitator, and we were talking about, well, sometimes a human might still facilitate. 00:38:48.398 --> 00:38:52.557 And it stands to reason that if a human's still facilitating, sometimes AI will be a participant. 00:38:52.659 --> 00:38:56.681 And I was curious, you know, as you've been sort of designing this i- is it different? 00:38:56.704 --> 00:38:58.846 Do, do facilitators need different skills? 00:38:58.847 --> 00:39:04.117 Do human facilitators need different skills to facilitate a session where AI is a participant? 00:39:04.755 --> 00:39:05.525 Yeah, I think so. 00:39:05.643 --> 00:39:11.878 Well, I mean, just the most obvious one, and I, I, I'd be very interested to know whether, you know, you and your, your listeners have encountered this. 00:39:12.146 --> 00:39:24.128 Any situation now where people are sort of in a group session where they have access to their laptops, nothing else, people are a lot more distracted by the fact that they can do things in the room or ask questions in the room. 00:39:24.128 --> 00:39:27.068 I mean, people always used to sit there Googling stuff, that's fine. 00:39:27.208 --> 00:39:42.226 But I think at the point where people are going off and really engaging another participant in the room who, who isn't physically there, where the AI is like, "And I jump to them," where you end up with the defensiveness what I talked about before, where people are swearing blind that they're, you know, it's their idea and it isn't. 00:39:42.521 --> 00:39:44.655 It's like a, something that Claude's come up with. 00:39:44.778 --> 00:39:48.456 The facilitator's job there is really hard 'cause you, A, you don't want to diminish anyone. 00:39:48.471 --> 00:39:52.692 B, there might be some politics in the room where you can't speak against people who are more senior or whatever. 00:39:53.032 --> 00:40:00.213 But you do want to try and try and get across the idea that we're all here in this very valuable time, physically in one place. 00:40:00.213 --> 00:40:02.594 Maybe you're all on a Zoom call, but it's all valuable time. 00:40:02.815 --> 00:40:12.755 Is… If there's something here that can be done outside of the, the room, that should be done outside the room, and what we're here, we're here to do is to have this conversation first and foremost. 00:40:13.293 --> 00:40:15.820 But also accepting sometimes there may be really useful points. 00:40:15.822 --> 00:40:26.731 You know, we've had sessions recently where you start Claude off at the start of the session looking at some things, and halfway through it comes back and it changes the, the tone of the conversation, changes the… And that can be really, really valuable. 00:40:26.836 --> 00:40:37.387 It's just building it in and building in a sort of understanding that you have access to these intelligences but the intelligences are there to do particular things, I guess. 00:40:37.391 --> 00:40:38.927 I think it's a really open, open problem. 00:40:39.132 --> 00:40:57.179 I mean, I've definitely seen more recently people sort of, sit in some kind of creative session, and really they'll start, like they'll have gone, they'll have heard the idea they like, and they will have then gone past it and started what would've been a process they couldn't start until they were like back at their desks or whatever. 00:40:57.597 --> 00:41:09.987 And they'll be like, "All right, let's build it. Let's design it. Let's…" And that, that seems lovely, but it destroys the point of having a process, the point of having these gates, the point of having these abilities for people to feed in, and that's really powerful. 00:41:10.001 --> 00:41:22.538 We think we took a, a website project which is a classic example where you have Typically, traditionally, a kind of creative or designer or someone who's gonna care deeply about the user experience, whatever. 00:41:22.829 --> 00:41:25.233 And then you've got developers who care deeply about how it's built. 00:41:25.677 --> 00:41:51.626 And the fact that those two people can now be the same to some extent, you can have someone who's never done any formal training in design or UX sit there and say, "Okay, so I think what we do is this," and you can have the UX design person say, "Okay, so I turn that into a website." You very quickly get to the point where unless there is a very effective facilitator in that process, people are stepping on each other's toes, they're annoying each other, but also they're not, they're not taking, they're not taking advantage of having a team. 00:41:51.916 --> 00:41:56.018 At that point, just set those two people up on two different projects if they're that amazing, right? 00:41:56.023 --> 00:42:02.871 So actually it's, it's making… It's part of, I think, a facilitator's approach in this new era is, is being champions for the human. 00:42:03.148 --> 00:42:08.277 So what, what bit of this human that I've physically got here is gonna be really valuable to this project? 00:42:08.429 --> 00:42:16.202 I mean, there's loads of industries that we're talking to at the moment where the apparent value of someone was the 300-page report they produced. 00:42:16.228 --> 00:42:17.688 But of course, that was

never their value. 00:42:17.757 --> 00:42:21.429 Their value was what they knew to go into the report, or the questions they asked or whatever. 00:42:21.621 --> 00:42:33.335 So at the point where you can say that you don't need to worry about this big chunk of work, and we can just have your Present, like your, your brain fully engaged in this for five minutes that can end up being more valuable, right? 00:42:33.411 --> 00:42:44.195 So, I think, I think it's a really interesting time to, to be a, a project manager, but, but also just to be working in spaces with humans who are at very different stages of their use of AI and, and trying to navigate that, I think. 00:42:44.268 --> 00:42:44.811 I love that. 00:42:44.822 --> 00:42:55.943 Actually, you just blew my mind because that idea that it's if we get people's competency to a certain level where they don't need to be working together on the project team, we can just put them onto other projects. 00:42:56.206 --> 00:43:04.137 And I know some people listening are like, "Wait a minute, are you just, are you getting rid of jobs?" And I'm like, "Well, no, not really." We get… We're, we're… Like you said about the pair programming, you could cross-train. 00:43:04.197 --> 00:43:13.514 And I think you said earlier, it's when you're working with someone who's coded for a decade, and you're, like, new to coding, and you both have access to the same tool, but we're gonna be using it different ways. 00:43:13.527 --> 00:43:16.028 But if someone was to say, "Oh, you know what I would do with that? 00:43:16.039 --> 00:43:43.458 I would ask this, I would include this detail, here's why." And then suddenly we're getting more of that context from somebody who's been, you know, thinking about coding or developing applications or back-end infrastructure for a decade, and we're getting ourselves to a certain level where, yeah, it might actually be, as a staffing model, I think that's really interesting to say okay, well, at a certain point this UX person could, you know, build the app on this other project, and this is something that we, we could explore, and they might enjoy it, and they might do a really good job. 00:43:43.836 --> 00:43:44.125 Yeah. 00:43:44.125 --> 00:44:00.608 And you know, there's, for, for agencies in particular so for anyone who has a, you know, has clients, you know, you've always wanted to get the, the people who understand the client the best, those who are closest to the client, as close to the work as possible, and that's now possible without really having to… That can be the same person. 00:44:00.733 --> 00:44:04.277 And I'm not saying get, fire developers at all quite the opposite. 00:44:04.306 --> 00:44:05.697 But like here's a good example. 00:44:05.864 --> 00:44:12.315 Most marketing agencies, and I work with a lot of marketing agencies, cannot for the life of them seem to manage building their own website. 00:44:12.501 --> 00:44:14.168 It takes them years and years and years. 00:44:14.172 --> 00:44:15.327 They never get around to it. 00:44:15.391 --> 00:44:17.362 It's never because it's a particularly complex problem. 00:44:17.362 --> 00:44:19.061 If it was a client, they'd solve it immediately. 00:44:19.414 --> 00:44:25.534 It's because of a priority thing, but it's also about this kind of it's ours, I wanna make it perfect, I wanna use all the skills in the company. 00:44:25.661 --> 00:44:37.737 And actually, we did something with… It wasn't an agency I worked at, but it was a suggestion I gave to another agency where I said,"Give the website projects," this was a couple of years ago, so when it was possible to start building websites. 00:44:38.059 --> 00:44:42.992 "Give it to the most junior employee and tell them that you're not gonna say anything mean about it when it comes. 00:44:43.001 --> 00:44:45.476 That you just, you just wanna see what they can do." And you know what? 00:44:45.632 --> 00:44:52.548 They came up with a pretty decent three-page website for this agency that several… And it turns out it's not that hard after all. 00:44:52.800 --> 00:44:54.478 But you can definitely get in your own way. 00:44:54.668 --> 00:45:00.259 So actually some things there is a, a real benefit to having a unity of, of thought approach on it. 00:45:00.744 --> 00:45:09.206 To come back to collaboration though, which brings it kind of full circle, that's where, that's actually a bit of an artifact of, like, where collaboration could go wrong in a pre-AI world. 00:45:09.655 --> 00:45:17.016 And that's when people are just like over-- they're overvaluing all the inputs and they're undervaluing the outputs of a process. 00:45:17.331 --> 00:45:34.840 So again, something where AI can, can do a good job 'cause it can make the outputs free or very cheap So someone come along and go, "Well, let's just have a look what happens if we do build the agency website a bit like this," sketch it out, blah, blah, blah, something like this, and then you can see it rather than feeling like you've wasted everyone's time going around in circles, so. 00:45:35.092 --> 00:45:35.925 Yeah, no, I love that. 00:45:35.925 --> 00:45:40.891 And especially, you know, as someone who I won't notice a typo in something until I've printed it, you know what I mean? 00:45:40.891 --> 00:45:44.313 It's once it's the real McCoy, then you can ha- really feed back on it. 00:45:44.313 --> 00:45:45.907 Even wireframes is a bit tough. 00:45:45.907 --> 00:45:49.010 You're like, "I guess so, kind of." In your head, you're like,"Well, we'll change that later. 00:45:49.010 --> 00:45:50.351 I'm not gonna let that fly," right? 00:45:50.371 --> 00:45:56.864 Whereas if, if you can get f- you know, further faster to build a thing and feed back on that, then yeah, there's… it's more of a conversation. 00:45:56.990 --> 00:45:58.188 That's that's really interesting. 00:45:58.228 --> 00:46:00.637 You know, the, the Amble experience seems amazing. 00:46:00.710 --> 00:46:02.429 You

know, I'm really looking forward to getting my hands on it. 00:46:02.648 --> 00:46:14.911 For teams who, you know, for whatever reason, aren't able to have a look at Amble or use it for their, you know, collaboration, what's something that they can do right away to start using AI a bit more collaboratively instead of privately? 00:46:15.101 --> 00:46:21.887 So one thing I would say there is i- I think a lot of people are tempted to sort of do the sort of put the AI in Slack thing. 00:46:22.039 --> 00:46:28.523 My advice generally isn't, and is until everyone who's kind of using it and on the same page understands it, actually, that doesn't seem to work. 00:46:28.563 --> 00:46:36.474 So I would say rather than take that route, try this exercise, which is something I do with quite a lot of people, is literally you get on a call and you share your screen. 00:46:36.690 --> 00:46:42.456 No one is allowed to edit, say, a Word document or PowerPoint, whatever it might be, some kind of artifact that you're creating. 00:46:43.003 --> 00:46:46.115 All you can do is feed into the AI, and someone has to type it in. 00:46:46.117 --> 00:47:02.985 That's fine, but everyone, you can only prompt, because what you end up doing then is talking about the artifact as something that you all care about and want to do well, rather than getting annoyed that someone's typing slowly or getting annoyed that something's not phrased how you would phrase it, you know? 00:47:03.012 --> 00:47:06.271 Once everyone agrees that it's just not a … It, it … And it can be a draft. 00:47:06.271 --> 00:47:08.333 You, you can say "No one has to care about this at the end. 00:47:08.333 --> 00:47:23.681 We can go revise it, and we'll give it to Bob or whoever is gonna take it away." But that kind of collaborative principle, you'll find that the conversations you have around it, which previously you wouldn't necessarily have been able to, if you think you do a half-hour call, there was no way you'd be able to write a 30-page document. 00:47:24.123 --> 00:47:34.527 But nowadays you can have that being written, and you can just feed into, into … It's, because it's on screen, you can see what's happening, what's changing, and you'll get people say, "Ah, you know what? 00:47:34.710 --> 00:47:35.963 That's not quite right. 00:47:36.128 --> 00:47:44.923 This risk register is it's missing these parts," or, you know, "We haven't really catered for what the client wanted here," or, "No one's mentioned security," or whatever it might be. 00:47:44.987 --> 00:47:46.092 So I would start with that. 00:47:46.092 --> 00:47:49.951 I think that's a really easy way to get started with, with collaborative AI in its most useful case. 00:47:50.318 --> 00:48:03.829 I wouldn't go down the sort of, you know, trying to necessarily put Claude in the middle of things and, and brief it, because without the framework around it of the project, it will give you very weird results, I think, and you'll end up with one person

driving, essentially. 00:48:04.251 --> 00:48:06.365 while everyone else kind of sits back and goes, "Well, I don't really know what's going on." 00:48:06.686 --> 00:48:09.405 I love that because you know, we wouldn't do that to a human either. 00:48:09.449 --> 00:48:15.146 I mean, maybe an intern or a co-op student at an agency, but you know, not on purpose, just out of necessity. 00:48:15.335 --> 00:48:23.644 But normally you'd bring someone in, you'd, you'd give them context around, you know, what their specialization is, and, you know, you'd onboard them, and then, you know, they'd- you'd feed them context as you go. 00:48:23.644 --> 00:48:30.106 And what I like about that model is that instead of, you know, everyone's own way of using AI privately A, everyone's kind of getting on the same page. 00:48:30.106 --> 00:48:38.438 They're getting used to not holding the pen necessarily, and they're also sharing context that the AI is presumably learning from as well, having just been born yesterday. 00:48:38.438 --> 00:48:39.300 Not really, but you know what I mean. 00:48:39.302 --> 00:48:43.163 Just showing up on a project, been like, "Cool," "you're gonna help us write this brief. 00:48:43.204 --> 00:48:49.460 But also I'm gonna take the pen from you every now and again, and I'm not gonna tell you why I made a change." It's never gonna actually figure it out. 00:48:49.543 --> 00:49:01.282 The context point is so important, and it's interesting, like if you look at how humans interact when they're working on a shared project, the context negotiation part, the point where they're deciding what matters and what doesn't, always happens. 00:49:01.360 --> 00:49:02.126 It's, it's crucial. 00:49:02.130 --> 00:49:03.240 People do it without realizing. 00:49:03.326 --> 00:49:05.862 They set the, you know, tone of the meeting or whatever. 00:49:05.867 --> 00:49:09.226 They say, "We're doing this, we're not doing that." They bring people up to speed. 00:49:09.226 --> 00:49:11.369 That's really natural, but they're very quick. 00:49:11.371 --> 00:49:21.989 Humans are very quick at doing this, really quick at identifying what in our vast pool of knowledge in our brains and what's happened in the last 30 years on this project is relevant to this exact moment. 00:49:22.184 --> 00:49:24.393 So that's what-- that's a gift we can give the AIs, right? 00:49:24.393 --> 00:49:25.329 They're, they're bad at that. 00:49:25.590 --> 00:49:36.670 You know, if you just, you know, in Copilot, for example, which you know, Microsoft's AI tool it has the ability to look at everything on a project that you've ever had and any, any email you send and stuff. 00:49:36.956 --> 00:49:43.668 And most people say when they use Copilot that, that they spend so much of their time basically just telling it, "That's not relevant. 00:49:43.686 --> 00:49:44.336 That's not relevant. 00:49:44.606 --> 00:49:46.518 That budget is not for this client," this, you know. 00:49:46.849 --> 00:49:50.943

So, so yeah, use all the human ability to, to do that context filtering up front. 00:49:51.179 --> 00:49:52.125 Luke, thanks so much for this. 00:49:52.161 --> 00:49:54.398 Just for fun, do you have a question you wanna ask me? 00:49:54.523 --> 00:49:55.186 Yeah, there is. 00:49:55.202 --> 00:50:01.097 I mean, so I'm fascinated by project management and particularly digital project management and how it's, it's changed, right? 00:50:01.106 --> 00:50:02.679 So I've been on the sidelines of it. 00:50:02.688 --> 00:50:04.644 I've never been a project manager myself. 00:50:05.072 --> 00:50:15.355 One thing that I've always hit against, and it's something that really feeds into Ample and so I really would like an answer, is if you have a team that does not want to be managed, has AI helped with that at all? 00:50:15.550 --> 00:50:23.333 I'm thinking, for example, like Asana is one of the, you know, is, is seen as a really, really good experience, a beautifully built product. 00:50:23.420 --> 00:50:29.155 I've seen it used in a couple agencies, but al- only ever seen people bounce off it, and it ends up being someone's job. 00:50:29.456 --> 00:50:31.567 So my question is, has anyone solved this? 00:50:31.567 --> 00:50:35.981 Has anyone solved the managing a team that doesn't want to be managed problem with AI? 00:50:36.519 --> 00:50:39.146 I mean, arguably, we've just shifted the problem. 00:50:39.177 --> 00:50:57.826 I mean, I do- know that like the thing I see the most is, right, like on the value prop for a project manager would be like, "Hey, listen, just like a, like your agent is gonna do the follow-ups for you. You don't have to be the person who nags about time sheets or, you know, is this thing on track for, for Friday or what have you," which, you know, in, in, in its- in itself is like, it's more complex than that. 00:50:57.826 --> 00:51:06.016 Sometimes when I'm checking in with somebody, I'm not just managing them, I'm also trying to get information back from them so I have more context to, you know, act in different ways and make decisions. 00:51:06.219 --> 00:51:18.496 But I have seen it take the edge off of that like, "Oh, my project manager is now nagging me. It's Wednesday, it's not due till Friday, leave me alone." These like nudges that come through the system, I think are a- an interesting way to solve the problem. 00:51:18.954 --> 00:51:22.498 I, it's funny because like I was at an Agile conference a, a couple years back. 00:51:22.541 --> 00:51:23.980 You know, Scrum, for example, right? 00:51:23.980 --> 00:51:26.858 This is our methodology where we're gonna move fast and we don't need a project manager. 00:51:26.858 --> 00:51:27.592 We're a self-managing team. 00:51:27.594 --> 00:51:32.340 We're just gonna get rid of that role, product owner, Scrum master, you know, development team, let's go. 00:51:32.670 --> 00:51:49.572 And even, even in that context there was this whole separate movement against that conference to be like, "Don't bring me to a conference. Leave me alone and let me do my work." And it was this sort of oh, not only did the system where we didn't have a project manager not take away that problem of feeling managed and just wanting to get on with your work. 00:51:49.597 --> 00:51:57.181 You know, people are now like,"Okay, well I don't wanna show up at standup because that also doesn't let me work." You know, at a certain level, I mean, like humans who don't wanna be managed don't wanna be managed. 00:51:57.407 --> 00:51:59.407 Like I don't know if there's a solution to that. 00:51:59.735 --> 00:52:02.007 And definitely I, I've seen it go the other way, right? 00:52:02.009 --> 00:52:09.750 Where it's like all the nags from the AI agents get ignored or actually are more annoying and some of the context isn't shared as a result. 00:52:09.804 --> 00:52:12.672 So I would say it's not a problem that's been solved yet. 00:52:12.802 --> 00:52:17.135 And also I, I kind of wonder, you know, you, you mentioned about the friction in the creative process. 00:52:17.137 --> 00:52:29.465 I wonder if sometimes the friction in the being managed process, as someone who myself does not like to be managed, I'm trying to understand the value of it, and I think it is one of those things where, yeah, there's gonna be some friction because we don't all want the same things. 00:52:29.467 --> 00:52:30.161 We don't all agree. 00:52:30.161 --> 00:52:32.795 We don't work the same way, and we have different priorities. 00:52:32.800 --> 00:52:46.454 So, you know, there's just gonna, gonna be this tension of sort of feeling nagged and feeling questioned and feeling, you know, under pressure that maybe is just part of the process, which is- Yeah … yeah, maybe just giving my audience license to nag. 00:52:47.686 --> 00:52:48.802 No, I love that answer though. 00:52:48.844 --> 00:53:01.938 That's really interesting, and it's a really… I was listening to a, a negotiation coach a while ago talking about how you deal with a room full of people who obviously like are never, essentially never gonna agree on a solution, and you, you find something else to agree on. 00:53:02.152 --> 00:53:12.501 So you find like you, you agree on the process instead, or you agree on what doesn't get included, and that, you know, produces an environment where people also feel like there's progress being made. 00:53:12.766 --> 00:53:14.911 And I quite like the idea of like the unmanageable team. 00:53:15.101 --> 00:53:21.349 It's, it's finding something that you can all come together, even if that is how annoying it is to be managed or something. 00:53:21.427 --> 00:53:21.737 I don't know. 00:53:22.634 --> 00:53:23.166 Yeah, totally. 00:53:23.478 --> 00:53:23.867 It's hard. 00:53:24.085 --> 00:53:26.889 I know I'm annoying, but we're gonna get through this. 00:53:27.193 --> 00:53:28.541 We can all agree on that. 00:53:28.659 --> 00:53:30.572 Luke, thanks so much for spending the time with me today. 00:53:30.601 --> 00:53:31.313 I've had so much fun. 00:53:31.313 --> 00:53:31.871 I've learned a lot. 00:53:32.114 --> 00:53:32.561 Yeah, me too. 00:53:32.635 --> 00:53:35.956 Where can people learn more about you and Supervenient and Amble? 00:53:36.021 --> 00:53:41.518 Yeah, so if you go to www.supervenient.ai that's our website, and you can find out more information. 00:53:41.530 --> 00:53:43.990 It's in a really … it's in the MVP stage. 00:53:44.007 --> 00:53:44.867 It's on a waiting list. 00:53:44.867 --> 00:53:50.891 If anyone listens to this and would like to find out more, we are recruiting a very small number of alpha users. 00:53:50.891 --> 00:53:55.322 You get to use it for free for a few months, give us feedback, tell us how useful it is. 00:53:55.583 --> 00:53:58.367 We pay for your tokens, if that's so you get some freebies there. 00:53:58.505 --> 00:54:01.014 And follow me on LinkedIn and, and find out more as it goes. 00:54:01.181 --> 00:54:02.034 Luke, thanks again. 00:54:02.195 --> 00:54:02.989 Thank you so much, Galen. 00:54:03.003 --> 00:54:03.737 Really good chat to you. 00:54:04.005 --> 00:54:04.494 All right, folks. 00:54:04.505 --> 00:54:07.594 That's it for this episode of Digital Project Manager Podcast. 00:54:07.751 --> 00:54:10.742 If you enjoyed this conversation, make sure to subscribe wherever you're listening. 00:54:10.811 --> 00:54:17.235 And if you want even more tactical insights, case studies, and playbooks, create a free account with us at thedigitalprojectmanager.com. 00:54:17.414 --> 00:54:18.909 Until next time, thanks for listening.