518 — Julie Dirksen takeover: What does it mean to be 'the human in the loop'?

Hello and welcome to the MindTools podcast. I'm your guest host, Julie Dirkson, and we're here to talk about some of the challenging human behaviors associated with AIUs. I wanted to have some conversations around a deep dive into these behaviors, and Rosticki and the MindTools crew kindly allowed me to guest host this podcast to make space for a few of those conversations. I'm joined by my co-host Claire Gibson, and we are talking to Marcus Bernhardt from Endeavor Intelligence, who specializes in AI strategy and organizational transformation. Thanks for joining us today, Marcus. Yeah, great to be here, and it's the second time for me on this podcast. So, racking up the numbers. I return contributor. Lovely to see you again, Marcus. Absolutely. So this whole podcast series kind of came out of the fact that the area that I'm really interested in is what do we think the humans need to be doing when it comes to AIUs? And Marcus and I have had conversations in the past about this, and so I'm confident that we're both prepared to sort of do a nerdy deep dive. But what I really wanted to do was kind of dig into some of these underlying behaviors, and the specific area of focus that I wanted to do for this episode was we know that one of the most important things that you can do around AI is sort of ensure that you're checking the outputs. It's that validation of what the AI is giving you. And any responsible person talking about AIUs always brings that up. Well, of course, you need to check the outputs. But they say that like, it's not the hard part, which is where I get really interested in it. So what do you both think? What are like the really crucial things that we need people to be doing around AIUs, and specifically the large language model AI, but really any of the any of the AIUs around checking our puts around validating things? Yeah. So let me zoom out a tiny bit before I dive in. I think this is a this is a really, really interesting and important question right now. We've been we've been blessed with the with the human in the loop governance model. We've been blessed with that because it's the easiest thing ever to say we will say to the person who copy and pasted and we find out later that they just copy and pasted without looking at it. And that's a fantastic governance model because we know exactly whom to whom to slap on the hand. And and there we are. The the issue with this is twofold. A, as you already mentioned, this this is not just having a quick look and seeing if things are correct checking outputs is a highly concentrated energy draining work if it's done properly. And at the same time, we're facing a narrative that now in the more agentic world, we can we can trust the outputs more. And so your your question is even more important right now because we're being told that we we will need to do less of the checking going forward. And I'm not saying we're you know, it's not ones and zeros. It's not either completely right that statement or completely wrong. We're we'll be we'll be somewhere in the middle here for a very long time. And we need to avoid saying we can we can trust the outputs or we can never trust the outputs. It's it's a sliding scale, depending on the context. And so yeah, that would be that would be my immediate reaction saying this is this is more important than it's ever been because we're being told by the marketing functions that this this work is less important. It's also sort of gives you what looks like on this surface very good stuff. I think the like the level of polish, even just formatting, I guess like the content itself obviously is brown to hallucations and errors. But I think for me that the real headspace I've had to get into because I'm very sort of design layout UX driven as a as a learning professional. When it gives me a really beautiful table, I I do have to kind of fight against the the bias in my own brain of going oh well some you know it's really beautifully formatted. So I'm almost legitimizing the content because the output is so well done. Almost to the point where I'm thinking out loud maybe I need to sort of say like don't put it into such a beautiful end state because I do think I give it inherently just more credibility because it's because it's beautiful. We're diving into the weeds here. Straight away. But I love this clear. I love this because this is a conversation also that Julia and I have had many many months ago for the first time where we looked at the fact that as humans we look at an output and we are convinced by the polishedness of it that the due diligence has been done. A slap dashed draft looks like a slap dashed draft and when a human has done one we don't blame them for having done a slap dashed draft. We we simply know that is what it is. And so we're not trained that the fully polished output is could be just as awful as what used to look non polished. So really really good addition you've thrown in there loving. Yeah I was reading a sub-stack I think it was sub-stack or blogger something like that sometime last year and it was a piece about the use of humor and learning right and and it had the research citations and it had the reference section at the end and it you know it it was like a decent you know use of the research where they would actually explain something a little bit about it and all this kind of stuff. I did notice that the research references were all a little old like you know there was the you know and I don't I don't necessarily mind that but when they're all old I'm always like you know um but I realized I'm like oh this could easily have been AI generated. And so the markers of trustworthiness so like if I read a similar article that is written by Patty Shank or Clark Quinn or you know somebody who I know and I know you know I know the trustworthiness of the work that they've done and all this kind of stuff but I I know how I can evaluate it because I know the level of rigor that they bring to the stuff that they do and that all of these would be signals like research or incitation you know right up of the thing you know reasonable you know sounding way that they drew a conclusion. So like I have all these things that I look for like for example if somebody's writing a thing and they say research says but they they cite you know a book that already summarized the research that's always a little bit of a red flag for me where I'm like oh just is that you know is this telephone effect that's happening and sort of diluted the message of it yeah and you know stuff like that but this has all the signals that I look for if I'm going to go trust something and I have my own set of rules like if if I'm gonna cite a research study I always have to go pull up the study and read the method section and you know the conclusions and stuff like that to make sure that I understand what it actually was but like if I'm just gonna sort of take this information away and kind of add it to my pile of knowledge and things like that this has all the markers of a trustworthy source except of course somebody could have just pressed a button and generated this whole thing right and that doesn't even mean that it's bad right that it still could be a good and useful article but does that I can't use these signals that I used to use of trustworthyness anymore because I know you you can just magically you know ask them to be inserted into your article and there it is right so it's it's an unsettling place to be when we can't win the things that tell us that this is valid or this is accurate aren't necessarily trustworthy anymore too 12 citations used to be a a hallmark of properly done work and and we would we would see these people who do this and you mentioned patty and and clock there and we would say absolutely absolutely they've read every single word in all of those and they've done some real work here and this is a this is a fantastic piece and very often we did in our minds think thank you for sharing this free on the web you've you've given me some really good stuff here to work with and yeah anyone anyone can produce something that looks like that in five minutes right now so yeah that Polishness is a real problem not simply from a checking point of view but also what about work that I'm not checking what a world work that I'm just just coming across I'm sometimes in my job I'm the person who needs to check these things but in my normal life that come across things that I'm not meant to double and triple check everything you know we come across news and we say okay this is a reasonable source of of news this is a reasonably newspaper I have a certain level of trust but there's so much flying around right now it it adds it adds to the cognitive load in the day today for for sure yeah and so I think you know one of the things that I wanted to do was sort of dig into the model that I usually use for behavioral analysis is comby which just sort of asks at the top level is somebody capable of doing the thing do they have the opportunity both sort of in terms of the system or the environment and the social environment and are they motivated to do the thing and so we know that people aren't necessarily validating outputs because we know that things have made it all the way through. I have a research paper that I'll sometimes show when I'm talking about this where the very first line of the research paper is certainly I can generate that introduction for you. And it made it into, you know, a paper, like an academic paper that presumably made it, it was, and it was a, you know, legitimate publisher. And it also, presumably made it past whoever was supposed to be reviewing the paper before it made it into this journal. And it's been redacted now. But it's, you know, the, the, the actual, like, thing got cut in pace. So somebody wasn't reading, you know, like several people weren't reading this as we go across. And so what are the capabilities that people need to have? And I think we've already talked about this issue of kind of trust calibration, you know, like being, stopping and asking myself what trust level do I have with this material before I, you know, copy and paste it and pass it on. But what other capabilities do people typically do you think people need to have to do this well to, to sort of validate the outputs and make sure? Evaluation is its own skill with its own responsibilities. And I think the role would include finding emissions and finding wrong statements. But also asking further questions before making a decision. There was a research paper from a couple of years ago now, I think, were nurses and doctors asked follow-up questions that the model did not ask before coming to a conclusion. So let's imagine the same model as a second opinion. If that second opinion immediately jumps to its conclusion without asking follow-up questions, then we simply know that the depth level of the context wasn't as deep. And so when we're looking at an output, it is both about finding out whether it is correct. But also looking for emissions. And this brings me back to your point, Claire. The more polished it looks, the less likely I'm probably to think there are emissions here. And I think that's the real human value ad is both a keen eye on validating. That in itself is a huge skill. Julie, you always say copywriters have a job for a reason. And you can't do eight hours of that work a day. And then it gets even more complicated by you, you should have maybe asked additional question. A copywriter doesn't also try to be a subject matter expert at the same time, do they? Yeah, because one of the things that shows up under capability when we're doing the kind of behavioral analysis is the term psychological stamina. So even if I can do this thing, can I continue to do this thing for long periods of time? And that's what the job roles have switched to quite a lot. And especially with things like computer programming and coders and so forth. And so I've been learning all sorts of new terms like verification burden. And you know, attention decrement and all of this kind of stuff, where you know, if somebody could write code for eight hours a day, it doesn't automatically translate into them being able to evaluate written code for eight hours a day. I was looking at some of the research into airport screeners, the security screeners who, you know, what's the expectation for that level of vigilance for that role? And I think the EU limit is nobody does it for more than a half an hour at a time before they swap people out. And some of the research was sort of suggesting maybe 45 minutes or an hour is reasonable in some other instances. But certainly nobody expects you to be able to closely scrutinize something at a really high level of detail for, you know, more than like 60 minutes at a go. But are we looking for that from, you know, computer programmers who are having to port through code all day long and things like that? And that kind of goes into like what is our systems? What needs to be present kind of in our systems to support this? I mean, already the most of the large language models have improved quite a lot from some of the early versions in that they will insert the reference for, you know, a number of things into the thing. So you can, you know, it'll say something about this research here, but it'll give me the research citation. I can easily click out to go to the paper and, you know, take a look at that and make sure that I agree with, you know, kind of how it's interpreting it or things like that. But what else kind of needs to be present in some of our systems to support, you know, people not automatically accepting things without, you know, kind of thinking about it or something that, you know, really helps them kind of, you know, have some sensors, some gauge of how much should I trust this? Or, you know, those kinds of, those kinds of questions. Have you seen anything, Marcus, or, or Claire, that you feel like, you know, kind of is improving the system around supporting that? It's not, not explicitly something that I have seen, but something as I was doing a bit of prep, I think it may be dovetails, capabilities and systems sort of together. And as I weirdly think that we are actually quite a privileged generation in terms of AI usage, because we know a little before times, you know, we are experts on this call right now. We're experts in our field, we know what good looks like, we put the working, we built that from the ground up, you know, you publish papers without AI use and you will continue to do so. So I think in that, that sort of capability systems thinking is, yes, in the immediate term, what can we put in place in terms of making sure people have the skills to check, etc, but also kind of looking down the road when, what do we do when the people join the workforce to have maybe never published a paper without, you know, in the era pre-AI. What happens when the learning designer in my team has never worked as a professional without AI assistance? So I think that kind of vigilance burnout gets even more dangerous when that's all you've ever done is check something. You'll continuously just checking for something else that you've never actually quite had a hand in building. So that's often something I think about in terms of what companies should be just having on the horizon, I suppose, in terms of the workforce that's coming through. I feel like I've taken that off in a completely different time jump, but I open the floor to either of you to explore that. I've said in a few different places that one of the things that I think is an issue is that all of the systems that we're putting in place around this AI usage is in this window where most people know how to do their jobs and we're moving out of that window, you know, quite quickly where there's going to be people coming into these roles with these systems in place and they won't already know how to do their job and, you know, the methods by which they learned how to do their jobs are clearly changing quite a lot. So I don't know, Marcus, what have you seen in that regard? Very little. We're new to this, all of us and no one's putting their hand up in an organization and saying we need to reorganize how we do things. Because that's tough and we don't know what it should look like possibly. I don't know the answer, but I'm observing, I'm observing the slog of the hamster wheel that we're all running, becoming tougher and tougher. Example, 10 years ago, I would get to my desk and I would read through the emails that arrived since the last time I checked. While I'm reading those emails, I'm kind of in my head mapping out how those different conversations are going and what's happening. I might have two ideas. I might jot down, get back to this person and ask one question. So I'm already thinking, but I'm doing the boring work. One of those silly things that we've been saying and I've been saying them as well is, oh, I will take over the boring work, we'll get the interesting stuff. Well, by going through my emails, I've just done quite a lot of thinking and remapping and re-engaging with my workday, it's sort of the warm-up. In the gym, that would be the warm-up phase. I'm not ready to rock yet, but I've got myself back in and I've kind of started to prioritise in my head and I've got an idea of how the day should work out in an ideal world. There are stressful days and there are curve balls and let's not get into the full details. This isn't an academic paper, but I've got myself up and running and now I decide my day. If I'm now expected that two agents have already pre-triaged my inbox and decided what I need to decide and how and I'm being faced immediately with three big decisions and three things that I need to answer to. I've had no warm-up, I have had no re-engagement cognitively with the topic and the expectation is that I can be thrown back in immediately and basically we're going from zero to high performance mode with no warm-up and no engagement or re-engagement time for our brains and so I'm finding that extremely difficult to think about how we could solve that because from there on your day will continue and that's the hamster wheel right from there on there will be more and more and more and also remember the time when we had the problem with the reply all emails if everyone is churning out stuff at the speed of an agent has pre-thought this through for me I'm just putting the I'm just putting the finer details on this I'm a project manager I've thought this through for 45 seconds deep after my agent gave me a brief and now boom and this goes out to five people those five people just got something back and then they fire back and then I fire back and so it's gosh this whole idea of the the booring stuff doesn't work for me and also this idea of the time we gained back doesn't work the time we gained back is if I'm working on a silo and everyone else is working at normal pace if all of my colleagues are working agent supported we're all working at hyper pace more comes in more needs triaging and I'm being I'm being pushed into a hamster wheel of only high-level decisions continuously non-stop from the start till the end of the day sorry if I sound a little bit worried or or or I'm not I'm not offering the solution but just observing what is happening right now goodness me the I'd if your example Julie is beautiful that they say that if you're if you're looking at the extra machine at an airport you can only do that for 30 minutes and there's a reason we say that we don't blame the people that they can't do to it was we say there is there is some kind of a limit there that we have all accepted we're not going down any route like that and the the time gained back there are very few places where I can actually see that um so just yeah I think I think as well like that's yeah I sort of I guess thought about that in my own mind but I think you just laid out really nicely Marcus so thank you and I think as well we are possibly living in a time where that speed is being celebrated and people are being asked to do a lot more faster with less and there's this it's almost kind of like I'm like a binary fish and of emails and tasks and that just seem to kind of uh you know in that endless loop where possibly you are being um you know praised for getting stuff out getting that email you know um getting that reply sent and not necessarily being praised for pausing taking a moment quality checking and making sure that you you know the email you send is defensible so that system for me is um I guess the top down on a bottom up that would be great to to start seeing in organisations that are um asking people to pump the brakes a little bit I'm not just um like gold star you've completed all of your projects but to what and you know for impact and one more I'd like to throw in there more of a romantic thought that the workflow is not the completion of tasks it could be it could be in the most boring of worlds a workflow is the completion of all tasks but I would like to say innovators and innovative companies and people who try to reinvent the wheel in their organisation had become better and stay in the flow with the times every team every function had ideas along the way they didn't just complete tasks if if they just completed tasks we wouldn't have a term for the water cooler conversation and so we are going down a route of thinking that everything we do is just a task completion and if if if something can complete the task we're good and we're praying that the thinking will happen alongside and my example would be in the organisations that I've worked when or that I've led when someone has put a report together for me they didn't just put the report together and present a final piece of output they were able to tell a story of that report they were able to tell a story of where friction putting that report together came from and where conversations led us to have to maybe rethink one or two elements of the report and for me that's where that's where the renewal where the innovation happens it comes out of those things if those things don't happen we don't need to romanticise the water cooler because because we'll get to the water cooler completely overloaded and we'll try to get back to our desk as quickly as possible because five agents are waiting on and on something I don't see how the most innovative and great companies in the world would have would have worked that way if we had just completed all the tasks yeah no I like that framing I think that's really interesting you know that that about work being more about things that are more important than just task completion and you know things like that I think that's great because I mean obviously the last piece of the behavioral stuff and then we'll kind of wrap this up but the last piece of the behavioral stuff is what's the motivation right which kind of goes to a lot of things like what are the incentives what are we you know and obviously it's the AI usage in a lot of organizations has been the Wild West in terms of you know ungoverned and you know trying all sorts of different things for for the last few years but we're going to have to be a little bit more deliberate about it you need going forward but you know the concern is of course that like we are inciting the behavior of getting more things done faster and you know not necessarily looking at the quality issue not necessarily ensuring that the answer that looks good but doesn't really fit the context or you know any of those kinds of things and so you know what what our organization is going to need to think about when they think about incentives or they think about goals or they think about any of these kinds of things to sort of guard against you know just turning into you know a sort of hamster wheel of task completion without any of these other things that we we think have value if we keep the end goal in mind and the end goal is more than task completion I think I think that's where the answer somewhere lies I don't have it right now but I think in every team in every function we we know what work we want to get done and we know where the frictions are and where the pain points are and the bottlenecks and so we can in a team or in a function think about what would improve our work and what would improve our workflow and I call this the the bottom up momentum in an organization where the frontline people know how to use a tool hopefully to improve how they're doing the work how they change their workflow and how they operate better that can only in most cases be done properly by the people who've been doing the work for the last three to five years and know it inside out if they're trained on the tool and leadership needs to be open to that as part of the new way of working you can't govern that from above and say this is how we're all going to change so there needs to be a dialogue but under a governance system and so if a if a if a team whatever they're doing says we can we can do this a bit faster we can do this a bit better and we can work work more optimized here they they have a way of doing that and if that dialogue is kept and we move away from how how many tokens are they burning through and then we ask five months later well actually are they burning through them cleverly and then and then we keep changing the parameters it's almost like like you know the the smile sheets in lnd it's it's what are we actually looking for in measuring and one of the one of the big four companies made huge headlines by just measuring who's burning through the most tokens and they won the you know they they were top of the leaderboard for burning through tokens wow yeah give me give me give me five minutes I'll be competing let me let me let me answer the computer system with that company and give me five minutes I'll set something up that puts me very close to or at the top of the leaderboard within a week and and will have nothing to do with the work I'm doing for the organization but but my name will get noticed by everyone so no there there has to be that balance and the focus on what is the work actually about and I'm purposefully not mentioning ROI I'm purposefully saying there is a there is a goal that a team or a function has in certain workflows and they know how they should or maybe could improve if they have the right advice and guidance technologically paired up with their tacit knowledge around how things work and the right mixture of how they've always been done and how we might innovate yeah yeah nice yeah now I like that because I think that's really interesting because I mean so much of this is kind of focused on productivity but productivity and service of what right yeah when someone invented the iPhone what was what was the productivity piece there right Yeah. Absolutely. And now they're struggling because the iPhone, you know, like to continue to find value in the iPhone, they made it fold. So okay. But the, you know, what they really needed is, you know, the whole next kind of, you know, what, what is the transformational technology, you know, going to be going forward and things like that. Not how do I fold my iPhone? You know, for $2,000 or something I thought, I don't know. It's a, it's a thing. But I mean, the, some of the most impressive organizations I've ever interacted with are the ones where they actually will invoke the values of the organization in the decision-making process around, oh, should we create this training program? Well, how does that fit in with our integrity value, you know, over here? And it's super rare. But I think that that's a really interesting point because we know that the incentive structure of a lot of this is we can spin this, we can spin these wheels really, really fast in, in a direction. But who is doing a good job of ensuring that this direction isn't just about, you know, ROI. But also in, you know, in service of these bigger, these bigger goals are these bigger values that we have either as professionals or as organizations or, you know, things like that. Yeah, to maybe throw in an example from, honestly, this week, where in my job, I get given a lot of stuff to digest. And then, you know, I decide on the behaviors that we want to change. And I develop stuff that helps people change their behaviors. I would argue that using AI in that initial process, controversial, it's going to make me a worse learning designer. Like, I need to be close to that content. I need to understand what is happening. I need to put myself in that position of empathy for somebody who just wants to do their job better so that then I can be good at what I am doing. So maybe there's something there around getting people to, or organizations to really think about what bit of your job is crucial and critical. And you, and motivates me and allows me to be creative, allows me to stretch my expertise, allows me to innovate, fine, and then I'm going to protect that with all I can, and then potentially identify, you know, when I'm pulling together. My goals, maybe actually, that's when I use Claude and he gets, you know, he interviews me and we can kind of, you know, I can use it to speed that up. But where actually do we need to get close to stuff, be able to stand up and defend something or query it and be that sort of temporary expert. I think for me, there's a huge part of my job that actually I would be very uncomfortable, you know, giving over to AI. Yeah, and I know that, you know, as somebody who just finished a book this year, you know, the thought process of working through something is part of how we come to understand it. And so skipping over that feels to me like that's really an issue. But obviously, again, when we start to look at the innovation piece, we're looking at all of these incentives, and you know, is that going to be a hard sell in certain organizations to argue, no, no, no, I should really work through this myself because of this sort of great understanding that I have of it versus the I could press a button and, you know, just create this. You know, that becomes part of that whole conversation too. Yeah. In my head, I said earlier, I have this thought process of checking outputs or asking questions, and Claire said she queries stuff when she thinks it through. And so there's a, for me, in my simplistic model that I use there, there's a real thought process around, am I diving straight in and wanting to get to the final output? Yeah. Or am I doing that phase where I'm zooming out a little bit, asking more questions and finding myself and then narrowing it down to the solution. And yeah, that's, I guess that's what it will come down to in this, in this question, how much time is there to get close to the content and query things and, and make decisions before we then, we then start to dive, dive in and finalize and hone in on the, on the, on the final output piece. And the first, like you said, the faster we do that, the less we can defend what we've said, I think you said. That's, that's also a part of having thought it through. If I, if I do five high, high level, very important decisions right now because five agents surfaced me something, what am I in three days going to remember about the thought process around what just happened? Bzilch. Absolutely nothing. And, and it's the same, it's the same with, with meeting summaries. If I've suffered through three really difficult long meetings, I remember what was said. And, and, and, and we won't bring up the same points again next week because we were all there. But if I read meetings summaries, I can get through them. I can immediately think, oh, one and two need actioning off we go. But in a week, I have no recollection of how we got there because there was no time spent in the meeting, in, in the thought process, in the getting there. So, yeah, that, that for me is the question, how much, how much time is there to do it? And if we, I think the, this brings me back to the solution is if we focus on the real stuff. If we focus on a team improving, not just, not just hitting number targets. Well, and it suggests that what we really should be doing is having some kind of triaging of the tasks that I'm putting through the AI, right? Is this something where I don't need to learn this, right? I was using it to review some contract stuff, you know, the other day. And I don't need to build my expertise in that. I just needed somebody to explain to me what this language meant, you know, in a way that I could understand it. And actually, that's a mediocre example because I'm sure there's reasons why it would have been better for me to understand it, you know, a little bit more, but whatever. But it's still, it's something, it's not something I need to build my expertise in. I'm not going to necessarily, you know, lean into that. That's not a core competency for me. It's not, you know, whatever, it's a marketing copy or something like that, you know, where it's good if I have decent marketing copy for an event, but it's also not the end of the world if it's just okay, you know, as opposed to, you know, something that I'm writing for, you know, my own, like writing a book, writing an article or, you know, something like that, where it's not okay if it's just a mediocre as far as I'm concerned. And so there's that kind of how close is it to the core, I think might be, you know, one of the ways that we kind of gauge the things that we're asking it for, how much is this something that I need to continue to develop as a professional to get better at, you know, like some of those kinds of triage things about where does this sit in, in like, you know, my unique offering, my values, you know, the organizational unique offering, the values, you know, all of those kinds of things, I think become part of that conversation. I like that. I think that's a, that's a brilliant approach. And if we, if we allow, if we allow the time for the triage, we've got to match a step in, in the direction of finding the solution to this, if we're, you know, there's, there's the example always of how many tickets can IT close versus are we improving how this organization works and how many tickets there are? Yeah, right. Right. If, if we're just looking at can IT close more tickets and, and that is the success factor, we know, we know we've been game theory set up the wrong game. Yeah. And so, and so I think you're playing the right game. You've put your finger on the, on the pulse there, I think, Julie, I think that triage with the right time, because it's important that we do this correctly. That would, maybe remove us one step out of the hamster wheel. And, and help, and help the organization as a whole. Yeah. And unless we're seeing, like I said, unless we're seeing all work, just the completion of one task after another. Now, I think, I think the, I think that focus, I think that, you know, being able to kind of think about upfront, you know, the purpose of the task and how does that fit in with values and goals and, you know, personally, professionally, organizationally, all of those kinds of things. I think there's something really interesting there. So that, I think that's a good moment. We're at a, at a good time to kind of wrap it up. So before we wrap up, we're going to do the, the regular feature for the Vine Tools podcast. What I learned this week. So have a clear do you want to start with something that you learned this week? I would love to, very, very small set up to this story. I went out for some drinks for some friends on Saturday, got to talk in as you do, and a friend of a friend is having an absolute nightmare with her bathroom renovation. Honestly, one thing after another. And my friend, here is a sort of scuba diver scientist by trade, sort of said, "Oh, yeah, it's an incident pit." And I was like, "What's an incident pit?" And she basically explained it. You might already know what this is, but it was new to me. It's where one small issue, basically escalates and escalates and escalates, and escalates. So she gave an example of when you were diving. Maybe your mask has got a small crack in it, and it, you know, water starts coming into your goggles. That's a relatively low level incident. But if you panic and you go, "Oh, my mask, something wrong with my mask," then potentially you start to sink or you lose concentration, and it's the sort of knock-on effect where you basically end up in an incident pit. And they suggest that actually it's almost a bit like an event horizon. It could just potentially just keep going and going and going for other but people are just, yeah, start off with something that's relatively easy to solve. So I don't know if it's thematic for what we've been speaking about today, but I thought it was really interesting and something I will be using to describe nightmare situations going forward. I love that term, the incident pit. I'm totally going to use that too. So, okay, fantastic. I think that's fantastic. Absolutely not that the incident pit. And I do, I do think it, it relates to what we've said because I made the comparison earlier with the reply all emails. This is the, is this the agentic pit? Yes. More and more and faster and more and even faster and even more. Possibly yes. So see what I did there. Mine is going to be closer to home. I had a little bit of a shocker learning this week of a friend of mine works at a large organization. And I've been helping them a little bit in their in their co-pilot studio to start setting up some useful things for themselves that can that can run, you know, at 4 a.m. automatically and give them a helping hand in what they might find useful and they've set up several nice things. And we were progressing really well. And yeah, I was just doing this as a friend's favor and the friends rang me and said, I've just lost all access to co-pilot studio and I contacted IT and because there's been a token use certificate change with Microsoft, this is now, this is now all changing. And because many of these things were timed, they have to, they have to run in an environment where they can run on a clock. And right now, they don't know when they'll get access back and there might be access to something called a sandbox or not. But my goodness, talk about talk about someone trying to improve, trying to work with the systems and making some really good steps and then without any warning, the platform vanishes and the access is gone. Talk about motivation. So, someone's been pulled out of the pit very drastically there. Oh my goodness. Oh wow. Well, mine is also, I think, relevant to the conversation. I've been listening to the book The Score by, I'm going to probably not do justice to his name. I think it's C type when and he is a game, a game design designer slash philosopher and looks at how game design, you know, the philosophy behind game design basically and things like that. And so he's written a book called The Score and I think a lot of these issues, you know, this sort of, I think the subtitle is something like how to make sure you're not playing somebody else's game or how to make sure that you're, you know, playing the game that you actually want to be playing and things like that and looking at how data systems can often kind of lead lead you down possibly but not useful angles on it. And I was also listening to another podcast right after that. He was talking about rock climbing in the book and how the grade system and rock climbing, you know, kind of guided his things and so I was actually listening to another podcast about rock climbing grade systems and how basically subjective they are but also, you know, in different countries they really measure different things. So the American system kind of really goes kind of straight to difficulty and it has a number system and the British system is more of an assessment of how risky a particular climb might be or something like that. Yeah, and so that there are these underlying assumptions behind all of these numbers and all of these things and so the idea of being able to kind of really surface those up and, you know, examine them and make sure he was talking about, he works in this philosophy department and he was talking about the reasons that he loved philosophy and got into it in the first place and how those all got kind of a little bit tweaked in the wrong way when they started ranking the US, I think it's the US news and reports would rank different departments and so then philosophy departments were suddenly ranking and how like law schools, for example, used to have like specialty areas and now it's very driven by like are you in the top 10 are you in the second tier or third tier, you know, there's kinds of things as opposed to this is a good school because it's in the region where I want to wind up and I'm going to build those connections or this is a good school because it specializes in, you know, human rights law or contracts or something that I really am interested in, not is now it's just are you 14 or 15 or 16 on the list or, you know, things like that. And so, yeah, I do recommend it. I think it's I think it's very germane to the whole conversation. So, yeah, love that. Overclassification or simplifying of data and and and the interpretation thereof is an endlessly interesting topic, I love it. Yes, podcasts, we can definitely all reunite to do that that topic. Yeah, absolutely. So, I think we're at a good time to wrap this up. Marcus, where can people find you? You can find me on LinkedIn or on endeavourintel.com. Is there anything that you've got coming up that you're excited about? Next week, I'm going to Ed Tech Week in New York and also heading out to the southeast summit ATD in Atlanta. Okay, great. I've got a week of travel coming up and looking forward to both events. Okay, fantastic. What about you guys? Yeah, Claire, how about you? Working people find you and what's something that you're working on these days? Oh, what a great question. You can find me on LinkedIn and depending on when this goes out, I think I'm looking forward to the learning tech awards. So obviously, that's kind of a mid-November. So, I'll be heading along there so I might see some slightly faces at that. All right, fantastic. People can find me at usablelearning.com or on LinkedIn and I am excited because I've got a new book on skills development coming out, which is a super interesting topic in the age of AI. And I'll be doing some launch activities at DevLearn in a few weeks. So, thank you so much. I love the conversation. I feel like we surface really important stuff and I very much appreciate it, so thanks everyone. Thanks for having me on. Thank you. Yeah, good to see you, Beth. Take care.