Why Your AI Problem Is Really a Leadership Problem with Minyang Jiang

Full Episode Transcript

This is the full recorded conversation from The Human Protocol Podcast, published by The Human Protocol Group. © 2026 The Human Protocol Group. Please link to and attribute the original episode when referencing this conversation.

[0:01] We're going to be AI first and we're going to be AI forward and that's how we're going to accomplish more. The company strategy should have been set. The north star should be existing prior to any tools that you implement. >> Everyone wants to get better. Everyone wants to feel motivated to embrace AI. But you need to have in place the right space where we celebrate people coming into not knowing anything and learning, right? And it's a phase approach. We cannot expect anyone to get in one day and become an expert. If you truly have a labor cost issue, you have a labor

[0:32] a labor cost issue, you have a labor cost issue with or without AI. So just address that head on, right? If you anticipate that AI is going to come and defay all of these human capital, then you have to be really honest with yourself and say like, well, my goal truly is to cost cut my way into competitive advantage and differentiation, which is very hard to do because like I said, that can be copied by your competitors.

[1:02] Companies write algorithms. Governments write laws. Someone has to write the human story. People don't resist AI because they are irrational. They resist when the story is unclear. When the purpose feels threatening and leaders demand adoption without building trust first. This episode is about the leadership work that comes before anything else. Welcome to the human protocol. The podcast about staying human while building their officer. I am your host, Michael Solomon. Most leaders

[1:33] your host, Michael Solomon. Most leaders think their AI problem is a technology problem. My guest today is going to tell you it is a change management problem and that the way that you frame AI decides everything that happens next. Call it costcutting and your people hear layoffs and quietly stop trusting you. call it growth and they start building with you. MJ Mingan Yen is the chief strategy officer at the fintech leader credibly where she leads company's gen

[2:03] credibly where she leads company's gen AI transformation. She's a Ford veteran who built the startup and spun it out and watched it fell then rebuilt. She's a Yale digital edict fellow and my favorite detail a literature major from Harvard now leading an AI overhaul. MJ, welcome to the human protocol. >> Michael, I'm so glad to be here. Thank you so much for having me. >> It is an absolute pleasure. I'm very excited about this conversation because I know this is a topic that you are passionate about as well. And when we

[2:33] passionate about as well. And when we had our intro call, you mentioned something that really stuck with me since then, which is this concept of lazy leadership, right? And when we're talking about lazy leadership for our audience, what do you have in mind? Well, I think uh that's a great question. There's a a few things I want to touch on, right? One is I often see change management being delegated or outsourced. So, a lot of top level

[3:03] outsourced. So, a lot of top level leaders will say, "Oh, we have something really difficult we need to accomplish such as the adoption of a new technology. Hello, VP of whatever. You go execute and implement that." I've heard a lot of leaders talk about how they're re middle management especially receiving direction to say you're going to you're going to be responsible for making sure that your team implements and adopts AI but also you're responsible for delivering everything else that you're delivering for the company at the same time. So to me that is lazy because you're taking the

[3:33] is lazy because you're taking the hardest part about leadership and you are delegating it to someone who doesn't have necessarily the resources or the bandwidth to do that. The second part of being lazy in my mind is when you take strategy and you outsource it to AI. When people ask, well, what is this transformation that you're leading? And the answer is, well, we're going to be AI first and we're going to be AI forward and that's how we're going to accomplish more. Well, the company strategy should have been set. The north

[4:04] strategy should have been set. The north star should be existing prior to any tools that you implement. So this tells me that there was a gap in thinking that is now being proposed with AI, but it's actually covering up the fact that maybe the company didn't have a north star or a believable strategy to begin with, but because AI is such a hot buzzword, it has taken the place of actual deep thinking. So those are kind of the two examples. And the last one I would say is um I will also often see leaders who

[4:35] is um I will also often see leaders who will mandate AI usage without actually doing it themselves. They come to the table with very ambitious proposals of what it can do and hey why can't we cut headcount here? Oh, we should be able to cut cost here. But they've never actually really played with and understood the constraints of AI in a real way. They've only heard about what it can be done. So not understanding the leadership failures and then reflecting that failing with AI for example I think

[5:06] that failing with AI for example I think is great. I think leaders should share stories of when they experimented and failed with AI and that I think will build more credibility than the three things that I mentioned earlier. >> Yeah. And this last part that you touch on is extremely important, right? Because how are you going to make people to follow your message and follow your mission and feel empowered about adopting AI if you haven't had the time to adopt it first, right? And it's not that we're going to get into this topic anyways or any shape, but it's it

[5:37] anyways or any shape, but it's it it makes me think about how do we think our government or governments are going to make laws, right, to make sure that um AI is there for us if they don't even understand the technology behind it. So everything should start at the top you understanding what the technology is and the strategy should follow after but it's not that we as leaders can remove ourselves from this equation when we need our people to follow our direction that start with us agree >> yeah and I think you know there's a

[6:07] >> yeah and I think you know there's a difference between the word empower and mandate right so AI empowerment is a very different tone than AI mandate because mandate seems to me like you're not actually helping people use this technology in a way that makes them feel capable. You're just saying they need to use it or else, which I'm not saying doesn't work, but I think that there are risks and longer term repercussions. And the way you would execute that strategy would be very different than if your

[6:37] would be very different than if your mindset was, no, I actually want to empower my organization with AI, not I want to mandate that people in our organization use AI because I'm frustrated by the lack of adoption or progress. Then I would say you as a leader that is your job to solve for that frustration and lack of adoption right and there's a better way to do that than just say well everyone has to do it or else >> absolutely all this conversation it comes around a topic that I also know that you're passionate about which is

[7:08] that you're passionate about which is when we talk about AI just from the lenses that we're going to implement it or we're going to adopt it just to cut cost right Um, and you mentioned something to me that it makes a lot of sense. It's it's all we do. It's implementing AI to cause cost. That's what competitors can do. They can copy that. That's not a strategy in a way. Right. >> Right. Yeah. And I think, you know, there's been a lot of literature lately

[7:38] there's been a lot of literature lately about how maybe it's not cutting cost super effectively. And I will give you very concrete specific examples. Right. So when we first started implementing AI adoption at credibly that was what the executives were excited about. Oh, we can increase employee productivity, we can cut down time, maybe we can eventually think about, you know, shrinking these teams. These are all fun and interesting things to entertain. But if that is your only benchmark, a few things very quickly start to happen. One

[8:08] things very quickly start to happen. One is I didn't see the initial wave of employee excitement around AI. But we know everybody was using it, right? They were using it to make their own work efficient, but they didn't feel sanctioned to say they were doing that as part of a company permissions. So, you know, you had shadow sort of IT situations pop up. You had the fact that people felt very hesitant and also um distrusted the company message because the first thing they're going to think about is like, "What about my job? Is it coming after my job?" And no one's going

[8:38] coming after my job?" And no one's going to say it, but it's you can clearly hear you can hear it in the silence, right? And then I think the other thing that happens is you start to remove really good ideas off the table because the first set of ideas are so what everybody wants to chase because they are lowhanging fruit and that distracts you from maybe doing the more value added thing. Um, and then the last thing is like you really have no better ROI because everything is going to be down to well, how much does this cost me and have I implemented the right cost

[9:09] have I implemented the right cost measures? I will tell you that our API costs have gone up. >> I'm on the billing. So, I'm seeing the cost these costs go up right now. We're being questioned like, hey, are you guys spending like too much with open AI and yes, there are some things we have to do there, but these costs don't necessarily disappear. They just get shifted in different ways. And so if your only benchmark for AI is cost and how much you're not taking out, then you're not looking at other benchmarks like is your organization more nimble? Are your people more capable working with AI? Are

[9:41] people more capable working with AI? Are you changing and adapting faster? Are you creating and doing projects that are worthy for the company that have a longer term return on investment that you are not seeing right now because they're using AI to do it? There has to be at least five metrics that companies look at when they're thinking about AI adoption. and so much only just look at have I saved human capital without looking at the other factors some of which are qualitative so even in my own personal experience it's only when we got into the second wave where we had concrete use cases where training was

[10:11] concrete use cases where training was implemented where we specifically told the employees no this is about growth and new ideas and not about cost cutting and efficiency that I really see another generation of adopters begin and be excited and confident enough to work with this technology >> and I'm sure that was in part MJ because of the strategy that you put together was for people to feel like AI is here to augment what we do not to really erase their jobs or anyway right because really we are human beings if we think

[10:41] really we are human beings if we think something is coming for our jobs we're not going to adopt something that is going to hurt us right it's a natural reaction to it but how we frame it how we implement it and the strategy and the communication that we put around that it makes total difference in in these projects because at the end of the day, these are still human beings implementing this. >> Yes. And I think leaders have to make tough decisions and I've had to make tough decisions. But making tough decisions to use AI as a scapegoat for

[11:15] decisions to use AI as a scapegoat for prior bad decisions or for suboptimal decisions is not necessary. Right? So you if you if you truly have a labor cost issue, you have a labor cost issue with or without AI. So just address that head-on, right? If you anticipate that AI is going to come and defay all of these human capital, then you have to be really honest with yourself and say like, well, my goal truly is to cost cut my way into competitive advantage and differentiation, which is very hard to do because like I said, that can be

[11:45] do because like I said, that can be copied um by your competitors. So I think that when people are talking about AI strategy and I've spent quite a bit of time on this. I will say I did not start with a strategy. We started with building culture and excitement and training. Then we built a um in-house skunk works lab that works with AI agents and then we rotated people through it. I've done nine plus hours of training with the entire company. I've gone to classes myself and brought that back. Then we said maybe we should have a strategy, right? Because now you've

[12:16] a strategy, right? Because now you've learned enough. you see how people work when you're seeing some of the gaps and now we said what is the future of work look like I developed an internal framework and one of the things that I think is also really important and leaders often eclipse this when they're thinking about AI adoption is that AI is really good at introducing complexity and so just as with change management which by the way everyone is terrible I don't want to hear anyone say they're good at change management they're not just because you've done it many times every times it looks different the

[12:46] every times it looks different the people who are best at change management are really humble about the fact that it's it it's it's hard every time because it's unpredictable every time you're dealing with people. So, first if you think you're good at it, you're already off to a bad start. Um, with AI adoption, when a AI is really good at introducing complexity and what that means is like I'll give you an example in underwriting, right? When we do underwriting and risk assessment, you can have a human decide the definition of is a business seasonal. Okay. Well,

[13:17] of is a business seasonal. Okay. Well, what does that mean? That means that for some businesses, their their high months are the summer season because they're in a tourist place like Virginia Beach or whatever and their low months are in the winter. Okay. Um for college towns, maybe it's the opposite, right? They have students back in the school, but then the summer they're out. Okay? So humans can understand some variation of this and they have big pictures in their head, but an AI can understand down to the individual business in that exact

[13:47] the individual business in that exact location. And we've tested this, right? And it's something that takes someone a lot of time to train to develop rules that they can then execute by. an AI can actually come up with a very dynamic model that is very specific to that business, which is great because it actually delivers more customer value because there's more knowledge and precision around the health of that business in the business of lending. However, it's complex because now you

[14:17] However, it's complex because now you don't have rules you can just apply. You don't have things you can just automate. Now you have to really think about, well, what is the prompt that I'm going to really work with AI to generate this? And by the way, like does do I agree with the output? Oh, it looks normal for this business in this area, but just a few zip codes over, it's going to look very different. So when you take that example and you sort of compound it over business cases, it's very important for leaders to understand when AI starts to change work, if you're using the power of the technology, you're likely making

[14:48] of the technology, you're likely making a lot of work streams much more granular and complex, which means people end up managing more work streams, even if they're agents, right? versus true automation which is you can just process the same thing over and over >> one and it's one process and a lot and and I've seen a lot of companies use jet AI to do what is actually just automation which also means you're not taking advantage of the full power of these models. So there are a lot of like if you're a leader trying to implement AI and you're not thinking about how it's

[15:19] and you're not thinking about how it's going to increase complexity, increase decision uh decision flow, increase the smaller granular branches of workstream and taking one big workstream and starting to create these tiny branches which are then more complicated and varied to manage. I don't think you're using the power of Genai in the right way. >> So many layers. some layered, right? And and I love the first one- which I think a lot of people still make that mistake. >> Genai is not an automation bundle up

[15:51] >> Genai is not an automation bundle up with a chatbot or with an open AI API in the back end, right? There's so many layers to it and that representation of what leaders needs to think about when the whole picture is in place is really extremely important because it's a there's so many layers. The more you work on it, the more your people will have to analyze different outputs and different ways. And you know, human beings, our brain is limited in a way when we are analyzing, we usually focus on one thing at a time. With the help of

[16:21] on one thing at a time. With the help of AI, we're probably getting multiple things to take a look at. So that whole idea that the work is getting synchron is really getting a lot from us. I see human beings still doing more with the help of AI. >> Yeah. And I I think you know I see a lot of LinkedIn posts where people are saying like oh you know now my teams are managing you know dozens or hundreds of agents and I have to really ask is that realistic right is that is that should your team be managing hundreds of agents

[16:52] your team be managing hundreds of agents or are you truly talking about monitoring erroneous output or something where that you're seeing you're seeing a model and you're seeing um anomalies. So, you're doing an anomaly monitoring and detection, which is not the same thing as managing thousands of agents. That's not that's not that's not um sincere. That's not an accurate description of what's going on because I I can't I can't realistically say like that's a good use of people's time and nor do I believe any person can actually manage >> monitor all that. >> That's crazy. >> It's not possible. But if you're talking

[17:22] >> It's not possible. But if you're talking about anomaly detection where things are running autonomously, but you have a way to look at something is >> where things go wrong. Yeah. Exactly. >> These are getting too long. We need to investigate. That's great. And I think that's that's we should be upskilling people to do that. That to me is a system AI as orchestration deployed the right way. But if you just have a bunch of workflows and humans are just actively looking at all of them, I think we need to rethink the sort of the design of the overall system. One thing I've learned through going through all this and I'm sure you had as well is

[17:55] this and I'm sure you had as well is there is no oneizefits-all for this conversation. is not because people are saying on LinkedIn or whatever this is what's happening that this is really what happening. I really loved the approach that you follow um at your organization which was let us try let us learn from our organization what works what doesn't and then we can continue you know to refine the message and see where we want to go and I think that's the absolute best way of handling this

[18:25] the absolute best way of handling this because we don't know everything and we are all learning in a way and when you have the whole organization that is composed of people with multiple roles and and different cultures trying to embrace this. Of course, you need to you need the organization need to have the time to really understand how our people are responding to it and then you keep going and you keep going. I think that's the absolute best way to really um understand how we can best implement AI

[18:56] understand how we can best implement AI in any organization because all organizations are different from one each other. >> Yeah. And I think you know there's not enough credit or conversation being really given to the actual experience of learning with AI right because so much of the authority and ex subject matter experts is on like I've implemented this great thing look at the capabilities of a it's like we I get it I understand these models are very capable I think everybody understands that how many of us are PhDs in Sanskrit

[19:26] us are PhDs in Sanskrit 1.2% 2% of the population and this thing can do, you know, better than most humans on really hard math math exams. And I I get it. I get it. Okay. 95% of that capability is irrelevant for the work that you're doing on a day-to-day basis because you're not using those capabilities. So, because AI operates on an abundance model, because it is so intelligent, it's actually harder for humans to harness that intelligence and figure out how it works for them. And so learning is such a crucial part and

[19:57] learning is such a crucial part and that's why I think there's so much conversation with AI around education, right? Like people just don't give themselves enough credit and organizations have to often lack patience to really learn. I when I was in a class at Wharton, one of the um people in my class talked about he came from a very large company and he was just you can just see he was just sitting there like I'm hearing you guys say all these amazing things about AI. All of my databases are separate. The same piece of data that lives in these

[20:28] same piece of data that lives in these different databases have a different label and mean different things to different teams. So the first thing I have to do is to invest millions of dollars in data engineering so that I can connect these databases and at least apply some semblance of a common label so I can run an MCP server so that I can actually give data to AI to analyze. And by the way, we have to then figure out because I can't do it all at once. What is the most valuable piece of data or

[20:59] is the most valuable piece of data or or clusters or groups or parts of that database that I can do that to begin with, right? And you hear these problems and they're very real and that's one company. And so the fact that companies are not acting or are later in adoption is not because people are just waiting around being like we don't believe in the capability of these models, right? I I don't think people I think people are generally at this point understand these models are very capable but you have to give the organization time

[21:30] you have to give the organization time and process and investment to actually make change happen. This is not unlike when you invite like like for example if you have a brilliant PhD who comes into your company and gives a and delivers to your manager just like an amazing lecture on how you should run a company and optimize for its profits and how you can turn this into a billion-doll venture and that person is 100% right and spoton. How long would it take you to actually get to that result? It's not

[22:00] to actually get to that result? It's not tomorrow. you have to actually do all those things that the person is recommending. It doesn't make the person less intelligent. It doesn't make your staff, you know, not recognize that this person is brilliant and saying all the right things, but the execution still takes time and there's still change management and learning and failing and restarting in all of that, right? It's not one happy path. It's very messy and that is how learning actually happens. It doesn't happen in a contained predictable way. It happens with

[22:31] predictable way. It happens with detours. It happens through failure. It happens through variability. That's actually what learning is. >> And I think people don't recognize that in adopting new technology. >> Yeah. I'm wondering if part of it is people, you know, being worried of trying and failing. We don't give enough credit for those that try and fail. We are a society that we are usually just you know acknowledging the good things when something when people get something right but behind the scene as you were

[23:02] right but behind the scene as you were saying the field and the journey it's a lot messier than just the destination when you reach out there. So this is what I also believe. We need to empower our people to be okay with trying and failing and recognizing when something didn't go the way we expected because that's how we learn what works for us. What happens in organization is when you amplify that to the whole organization that can be hundreds, thousands of people. Of course, you need to take all

[23:32] people. Of course, you need to take all that into consideration and that's why it this thing takes time. If you really want to implement it the right way, it's about empowering your people to give it a try to use it and make sure that you get data from them. What is working, what's not working and be open to feedback at the end of the day. >> Yeah, I think so. there's some practical things we can do here, right? So, I think first of all, I was telling someone else the other day, I think especially like America has this very interesting

[24:02] interesting I I I almost think like just in relationship with the word failure, right? It's like we are always told you learn through failure and failure happens in everyone's lives and it's really important and overcoming adversity and all that and then we celebrate the people who are really really successful and often in organizations the language around failure and the artifact around failure are different which means which

[24:33] failure are different which means which means like people say they celebrate failure but they don't actually celebrate failure. So there's a usually there's a gap because now it's like oh you cost us money and blah blah blah, right? Like I I'm lucky enough to work for an organization where this is not the case. I cannot tell you how many projects I have failed at and spend hundreds of thousands of dollars of the company's money and I got no backlash for it because I learned. So with AI there has to be different metrics, right? So first of all, I would expect 90% of your tiny project to fail. That should be a very flat expectation. I can

[25:04] should be a very flat expectation. I can name off the top of my head multiple agents I've had sales reps and other people deploy that didn't work and it was fine because they learned how to use AI. The second has to be there has to be a metric around AI proficiency for the company which means you do this thing you launch this little agent. You build this gem. You wrote this piece of code. You built a UI dashboard. It didn't go anywhere. No one adopted it. You're not even using it for yourself. Okay. But

[25:34] even using it for yourself. Okay. But did you get better at working with AI? And if the answer is yes, then we should celebrate that. Right? So much of our metrics are outcome based because we've been taught to think about that in a business context. But in learning and adopting new technology, there has to be metrics that are learning based. And learning is is qualitative and subjective and difficult to measure. But it doesn't mean it's impossible to measure. I've done surveys in our company around employees expectations and satisfaction and usage of AI saying

[26:06] and satisfaction and usage of AI saying nothing about did you produce something that the company is actually using which by the way we give people prices if they do but just seeing that percent of people who are whose comfort and ability to work with AI has gone up that's an overall corporate success metric right so I would suggest that if we shifted or at least added one metric of success on to learning. Are your people learning? And are they because if they're learning, they're also more adaptable to change,

[26:37] they're also more adaptable to change, right? Which is what you need that. You need that in a resilient organization. Then keep going. Maybe tweak how much you're spending, how you're incentivizing people. We had a very recent debate where we were talking about opening up Claude to anybody in the company who was interested. and our CFO and our executive team, part of them were like, "Well, that's a that's going to be quite a bit of money. So, we need to get people to write a business case and suggest why they want to use it and what problem they're trying to solve." And immediately myself and the VP of

[27:09] And immediately myself and the VP of product was like, "Nope, we're not going to do that because you're setting up a barrier and we want to be AI forward. So, what we need to do is actually give people free access, let them play with it for 30 days, and then have them tell us what they tried to do with it and was it interesting or not. And guess what? People got access, they played with it. A lot of them ended up not retaining the access, but we learned one very important thing, which is what I just told you. My my own team played with Claude, right? And we had a meeting where they had access to Claude and they said and I said, "What kind of problems

[27:39] said and I said, "What kind of problems are you trying to solve with Jai?" and they gave me well I want to run this report at this time of the day and I want to send it to like this merchant and blah blah blah and then I said that's not a geni problem that's an automation problem so you're trying to solve for a different problem not using this tool okay so now we have our data analytics person who is learning to build automation and now we're creating these tools that allow these people to be better they learned what is genai use case what is not genai use case and they didn't end up using the more

[28:09] didn't end up using the more sophisticated models after 30 days it actually got shut off for them. That is all success because look how much you learned in the process, right? Without creating that barrier to entry and you learn things you never thought you would learn. That's that's the power of opening it up and encouraging people and not putting a limit on it but also letting them know like there are off ramps. You can still solve your problems and do better but you don't want to prevent people from trying new things. I I love that angle of, you know, turning people from passive recipients

[28:41] turning people from passive recipients into really builders because that's what we really need. We need to give people the tools and, you know, the education and the ability to go there and try see how it can help you and how you can learn. And we are human beings at the end of the day. We know human beings. We love when we learn and when we feel ourself useful. Right? That's how we should be looking for. That's why our people, everyone wants to get better. Everyone wants to feel motivated to embrace AI. But we need to have in place

[29:11] embrace AI. But we need to have in place the right space where we celebrate people coming into not knowing anything and learning, right? And it's a phase approach. It cannot be we cannot expect anyone from getting one day and become an expert. we all learning on this very quick and fast-paced environment which is uh this whole conversation around AI and how we can we can embrace it. Um but one thing I love um about the things that you're doing um MJ is um I think

[29:44] that you're doing um MJ is um I think that you mentioned to me that you have stood up an internal academy right and you're running pots of people building real agents and I'm wondering if you can share with us how this activated people and how it's motivated them so people And you know they came in and they embraced this instead of feeling that this was a mandate for them to do it. >> Yes. So uh this is now I would say round

[30:15] >> Yes. So uh this is now I would say round probably three. So I would say um what I'm about to describe to you is probably four years in the making now right. So it's not something you achieve on day one. So I run the internal innovation incubation team called skunk works which is modeled after Loheed Martin's um very famous skunk works project where they took a bunch of really talented crossunctional people gave them a goal gave them very little resources and said go solve this problem and because we need you to solve this problem and so I often joke uh when I talk about the skunk works team because I said you know

[30:45] skunk works team because I said you know hey I'm not buying anyone lunch there's no t-shirts the work and the companion the the work and the company you keep which is the people who you work with those are your rewards okay so the first iteration of that team we were always giving them this is preai pre-Genai um unsolvable problems right so the question so immediately you set the culture of these problems no one has easy answers to in fact no one has answers to so you don't come in with

[31:16] answers to so you don't come in with your expert hat on thinking you know everything you come in ready to challenge you come in ready to voice your opinions you come in ready to fight for your beliefs and it gets better. What's interesting is part of the automation problem in the beginning I knew from the very beginning it was going to be a failure because people like I said um when we did end up introducing chatd and generative AI it was a very human construct to look at the best underwriter the best processes we have in place and say how do we take

[31:46] we have in place and say how do we take this general intelligence tool and then how do we mimic all of that how do we make it do exactly what this is doing and by the way we're going to compare the answers against our best underwriters answers. Oh, and it's not working. It's not even coming close to humans. Okay, great. Is that how you automate? Because that's general intelligence for very specific tasks and these tasks are optimized for humans who can only remember 17 things versus a really intelligent model. So, the first wave of that, which by the way was a year, okay? Because this is how long it

[32:17] year, okay? Because this is how long it takes people to change how they think and how they actually work, failed. Failed. All right? And so second round, same people, same question, right? Now the people were like, great, we just failed. Okay, now they started asking question differently. Okay, why are we doing it this way? What's the risk if we didn't do it this way? Okay, now we have some real answers. Now we have two patents from that part time of how we worked. And you can see this is for year

[32:47] worked. And you can see this is for year one, we produce nothing. Nothing. All right. Year two, two things. Year three, six things. your four which is where right now multiple teams working at the same time. I've cycled people through. I've done multiple trainings. Now the people who are building are level like like junior level people in the company who have never even before touched AI but because their managers had been on the team and now empowers them and feel confident that confidence and inspiration and yes you can do this and

[33:17] inspiration and yes you can do this and by the way I know how to do this and I'm supportive and I've done this before trickles down to them and they feel like they can. So now I have literally two agents that are built by junior level people in the company who are not technical who have never been on Skunk Works. They were pulled in and I asked their manager and I said how do they feel about working with AI and he's like they love it they're having a great experience right so building the right culture actually being okay with failing and then knowing that this cycle is going to come but it takes a little bit

[33:48] going to come but it takes a little bit of investment in time upfront. So if anyone is going to build this incubator, I would suggest that you start the same way we did. Small elite team, give it unsolvable problems. Make sure though that the first round is like usually the company goes in with like we need to have some quick wins. >> If we can just show that we got some quick wins and then we can show that this team is worth its value. It's like no, your your output is not did you deliver the quick win. Your output is did you change how people think about how they're supposed to work? And that

[34:19] how they're supposed to work? And that then meant that this becomes a self- sustaining cycle to the extent that I don't really run that team anymore because I just sit in the meetings but now people are just running these things by themselves right so now it's a truly a self-sustaining body that can take new technology bring other team members in as needed train them push them out and then the next version of that is one of the agents that we are building it's not an agent is a skills repository and what he's actually doing is taking all the different pieces of knowledge around the

[34:49] different pieces of knowledge around the company and putting them in one central area but in a way that anyone can go in access what they need to access and then build agents on their own. So that is now what we're trying to do and that's only because people have gotten to a point where like people are comfortable first of all even giving you the information for you to work with right because if you think AI is going to take your job why would I give up my secret sauce of like manuals or stuff that I'm doing so now everything's in one place right I'm seeing the Q&As and I'm seeing the feedback loops and I'm seeing like MCT servers being built so those are the

[35:21] MCT servers being built so those are the kind of things that will come and they will happen faster and faster but that initial phase the learning phase is is going to take a little bit of time. It probably took me three years to find a version of AI human hybrid models that work >> that works for you guys. >> It took me three years to do it and and if you take less than three years, like congratulations. But if you told me you found it in year one, >> I will have a hard time believing you. >> I would too. It doesn't happen from night to day. It it's really doesn't.

[35:51] night to day. It it's really doesn't. and and and you know it's a that's what I want people to really understand especially on organizations implementing this the way we do things matters and it takes time right the quick ROI that people are looking for that's not going to materialize tomorrow right you really need to give time and you need it really needs to be part of the overall strategy when I listen to what you were saying a few things comes to mind one it takes time Right? It takes effort. Second, the

[36:24] time Right? It takes effort. Second, the training creates awareness, but really the building with it is what creates the belief that people have on on on working with this. And that's where the magic happens because once you get people to feel part of this mission and they like what they're doing, they like what they're building, they will continue to embrace it, right? And the more they work with it, probably they will continue to find more use cases where this can get better with the help of AI. And to me, that's really kind of what we

[36:55] And to me, that's really kind of what we all need to think around, how we can get people to jump on it with us so they feel like they're part of the mission and they believe on it. Because once people believe on it, we know what happens. They keep going and they follow they follow the direction. >> Yeah. I think working with AI is very much like a lot of the intangible things that we often talk about in businesses but we give we understand those things take time. So for example, a business building trust that takes time. A

[37:27] building trust that takes time. A business building customer loyalty that takes time, right? You may want to have the trust and the loyalty and you may think you earn it and you deserve it and if people can just do this faster, but I think leaders understand that building the trust in a brand and building the tr and the the loyalty of customers there time has to be a component because you need to be stress tested, right? Those things are the product of having been tested and so AI adoption is the same

[37:59] tested and so AI adoption is the same thing. You want your people to trust using AI. You want AI to become truly embedded in the work at your company so that it's not just functioning on a mandate which then kind of goes away or even worse people are actively sabotaging you. You want it to be truly adopted as a tool of excitement and people are working with it and want to work with it and are advocating for it and being evangelists for it. Then that is like building trust and loyalty. It's going to take some time and that's okay because when you get there then it's

[38:29] because when you get there then it's self- sustaining. >> Uh it is fascinating to hear from you know real builders like yourself how you know how things started working for you the time that it took. It's always, you know, it's always to me fascinating to hear really the honest take on what is happening with AI out there, right? Because the more we hear people in the news, some people may think because they, you know, everybody just has a recipe for implementing AI in three days and the ROI in a week, which this is not

[39:01] and the ROI in a week, which this is not real life. Yeah. And we need people to really understand and what is real in all this conversation which it takes me to to my next to our next segment. MJ, if I recall from our previous call, I think you mentioned to me that you're you stop hiring junior engineers, right? And now you have a need for more senior engineers that are stuck with reviewing complex AI written code that demands debugging.

[39:31] debugging. Things have changed, right? The work has become more complex. Um, and I wonder if you can work me through on when did you see that shift happening? >> Yeah. Well, we didn't stop hiring junior engineers. It's more that I saw the the we have junior engineers and senior engineers. And the way that our tech department started with working with AI and how they asked people to work with AI was which is very common, right? In the beginning, the junior engineers are

[40:02] the beginning, the junior engineers are like, "Great, I'm going to use AI to help me write my code." And so they would deliver and try to and code for code review for senior engineers who then spend a lot of time looking at this junior engineers work. A because it's complex and b they don't even know how to debug it because they can't even find where the error is. So then they're using AI to try to it becomes a maddening because they feel like they're not really um exercising their value as engineers. And also the code that they're given doesn't allow them to

[40:32] they're given doesn't allow them to coach and develop people because it's all prepackaged and written by claw code. Right? So eventually we kind of had to really think about how are we asking people to actually work in a way where the senior engineers can still mentor the junior engineers. The junior engineers have to be able to explain why the code is written in a certain way. How do they think about it? Right? understanding and explaining how they're thinking about it is very very important. And so it's not that the team

[41:02] important. And so it's not that the team stopped working with AI. It's how you work with AI as we all know has different consequences and repercussions. And if you really want to preserve the value of how someone with more technical expertise wants to present themselves and their work, then you need to have them work with AI and their junior developers in a way that sustains that. It is not to cut AI out. It is to use it in a way that develops and continue to develop that human expertise while increasing productivity. But it is not okay to say now that you

[41:32] But it is not okay to say now that you have these tools, I'm going to expect you're going to code three times as more because what you're going to get is three times the amount of garbage that then somebody else has to review and make right because at the end of the day, they're still on the hook for code that makes it to production. So this is where I think like setting arbitrary standards without really understanding how humans are motivated along those work streams, you're going to shift value to different parts that you didn't anticipate. And you may have to learn from a couple of iterations, but I think

[42:03] from a couple of iterations, but I think that's what I see and I've heard people talk about mandating more output for the junior developers and then the developers would do more out work uh uh output that is not of the right quality which then requires more heavy um monitoring and code reviews and in-depth code reviews which then take up people's time who are actually technical experts who scarcity, right? Because you're fixing other problems in the company. So, you're creating more problems in a

[42:33] So, you're creating more problems in a way um depending on how you choose to work with AI. >> Yep. That that makes total sense. And and that's what the shift we are seeing not only with developers and engineers with in any other of of the roles that that our teams have things are changing and the depending on how you work with AI these are different problems and different kind of you know tasks that you are probably going to be responsible for because now it's only it's not only human beings in charge of the whole

[43:03] human beings in charge of the whole process now we have AI in the middle and that itself creates a different set of challenges and potential processes that we need to follow to to kind of manage the whole process. Managing processes now is not only humans, it's humans plus AI and automations, right? And that's the shift. >> Yeah. I I mean I think that's also why like I'm not as concerned as a lot of people about AI taking over creative content or writing because now we are

[43:33] content or writing because now we are all so used to reading AI generated things and there's such a now there's even more demand for original writing and voice and art right for for things that are different there's even more demand for that humans want to go after the thing that is not easily produced that's human nature So, so artists and musicians and others to say like, well, now they can just take over. There's a piece of it that's true, but real higher quality originality and creativity is

[44:04] quality originality and creativity is not going to be easily commoditized because humans will want that authenticity. And so, I think there's a market for that. Um, and because I' I've definitely read enough AI generated stuff that and people I'm starting to hear people complain about it, which means that when you read really good writing, it feels even better. feels good because you're like, man, this is really a genuine voice. >> It's feel like us. >> Feel like a human. They're talking. It's raw. It's authentic. It's vulnerable. Like those are all good things, right? And so I think there's actually um space for creators um going forward as well.

[44:36] for creators um going forward as well. >> Yeah, I I believe that too. Really do. And and I want to ask you something cuz I know that you are a leader um implementing this. Um but not every leader or not all leaders may be in the same type of organization that really embrace uh innovation um risktakers in a way and some organizations MJ they're just laser focused on cutting cost right

[45:07] just laser focused on cutting cost right and creating efficiencies right where is from your point of view where is that tin line or that line between efficiency and try and take the time that you need to really understand how we can implement AI in our organization. >> Yeah, I mean I I actually don't think that efficiency is a wrong thing to pursue, right? My argument is never you

[45:38] pursue, right? My argument is never you should not pursue being more efficient. I just don't think it's the only thing you should pursue. And so I've um written and I'm I I've always thought about if I were to advise those leaders, I would say, okay, write down all the ideas you have that around AI and how you can use it to cut cost and increase efficiency. Okay? Then put those away. And then I want you to write down ideas that it's going to not do that. It's

[46:08] that it's going to not do that. It's going to take you longer to write those ideas down because they're not going to be obvious. But I would challenge them to say at the end of the second round, look at those ideas. Are they more different? Are they more differentiated? Are they more original? Do they grow revenue? Are they things your competitors can't copy? Okay, maybe still do your efficiency and cost cutting measures. Maybe pick one idea from your second list and also execute on that. That's the only thing, right? It's not to replace one side of the

[46:38] It's not to replace one side of the equation with the other, but it is to acknowledge that just because you came up with all these ideas that are lowhanging fruit that you think is really great, don't let that detract you from at least trying one idea from the side where AI is empowered to do something that has nothing to do with efficiency and cost cutting because you might be surprised with what you find there and you might really like what you're writing down there because those probably are things that's not easily duplicable by somebody else in your category or in your competitors set.

[47:08] category or in your competitors set. >> Yeah, I I I agree with that as well. The whole conversation, it's a layer conversation. It's not we can't just focus on that. The pursuit of efficiency, it's one that we all should always be looking for one way or the other, right? How we can make our processes and our people more efficient. But there are so many other layers that we should embrace that it's not only having our you know looking at efficiency at the end of the line when it's actually how you can bring people

[47:39] it's actually how you can bring people together that they really can embrace um AI and and then continue that conversation and and I'm curious um MJ because now we get to a point that we are implementing AI in different parts of the organizations you No, the goal is probably that every department um one way or the other start implementing AI or have AI implemented. What does trust will look like

[48:10] What does trust will look like operationally speaking right once AI start touching customers communication and things related to IP right reputation of the company. curious to see to hear from you you see this going next. >> Yeah, I mean that is a I think very important question and I would say I don't have the answer for that. I have some thoughts around it and I can maybe help people understand. So first of all um in terms of AI being implemented in

[48:41] um in terms of AI being implemented in different parts of the company I think that is absolutely the future. It will not get all implemented at the same time nor in the same industry at the same time. Industries with a lot of compliance and complexity is going to take longer. Industries that are very small but don't have any resources are going to be outsourcing more. And then those in the middle are probably doing a combination where they're also um natively building. Right? So uh so first of all is to recognize that these things are going to happen with different clock speeds for every company that's and that it's going to be unique to the company's culture, their appetite for risk in the

[49:12] culture, their appetite for risk in the industry that they're in, their size, their compliance requirements, etc. a Walmart that's operating globally is going to be very different than a mom and pop shop that's based out of one zip code in in uh in one state. Okay. So, having said that, a couple of things that I've seen is like often companies that are uh you know, sometimes they'll take the most the strictest requirements and they will use that as a way to set up trust, right? To say like, "Hey, we're going to take our compliance requirements. this is really strict and we're just going to apply it across the

[49:42] we're just going to apply it across the board because above that we're fine and we're going to work within those constraints. Um compliance and trust are different things but they kind of build off of each other. The second one is like people try to centralize it right. So there is a um you know you can imagine there is today it's usually rolled up under tech or legal where uh employees are given permissions on how to access certain models the kinds of models they can access the kind of information they can pass into models um the good companies will have very clearly written employee handbooks on

[50:12] clearly written employee handbooks on what's permissible or not but the idea is that like in the future I example see I expect to see organizational models where there's like an AI ethics and safety committee right that like looks at the content look the output. So there's a way to do that there. I think the last thing which I would encourage that people should do regardless of what model you adopt is you have to make sure that every person in the company understands and it's trained on the risk associated with AI so that they can use their best judgment. We have a lot of

[50:43] their best judgment. We have a lot of trainings right around things like um uh you know uh how do you deal with workplace harassment whatever. So there should be trainings like that in the future where you teach employees what does it mean for confidential data what does the impact that you're going to have on customers etc etc right so there's going to be ongoing training around how AI is being used but ultimately it's going to be up to each employee to understand the impact of

[51:14] employee to understand the impact of using AI the wrong way or using AI to generate trust for a lot of companies having that initial ual definition of what does it mean for our customer and stakeholders to trust in our use of AI is already a tough question. What does that mean? What does your content look like? What does your brand look like? Do you disclose you're using AI? Right? What are you using internally? Is it touching customer data? So having even a discussion around that topic alone what does trust in my AI production looks

[51:46] does trust in my AI production looks like written down and having discussed at the executive level is would be a first step and the second step is having an employee handbook that would that documents fair use and making sure that it's updated in a dynamic way. I would I mean like that question I'm not even sure I've taken to our team. I got to really think about this after this podcast. >> I keep thinking about it. That's why I it's it's something that it makes me you know I keep thinking about all this stuff because it's still you know when we look at this and and when we look AI

[52:16] we look at this and when we look AI everywhere how we can keep up with security right and the way I see this going forward it's still the human layer it's kind of the security layer right that will continue to be as we are part of this conversation and and I think we in a way we keep underestimating or or or maybe not having this conversation enough because the more we implement it, the more we need to make sure that we keep like you say employee handbook and our internal processes up to date and updated for

[52:47] processes up to date and updated for this new era that there are so many layers in between. More to come on that I guess right as we continue to learn on it. >> Absolutely. >> Yeah. and MJ, we are approaching the last segment of our conversation which is really one general questions intended to be asked to multiple groups. So, I'm going to start with the first one, which is what would be your piece of advice for those individuals that are trying

[53:18] for those individuals that are trying just to figure it out how to use. >> My piece of advice is if you're trying as an individual, um, think about it as something that you're playing with and not something that you're trying to control or something that you're trying to understand. Because the more you can take an interactive and play mentality to working with it, the more fun you're going to have and the more you're going to want to do it. If you're trying to implement it in the at like an inner organization then those three components that I talked about earlier you need to have the crowd which is um you know you

[53:48] have the crowd which is um you know you need to have like people individuals building that culture who are excited you need to have an innovation lab the crowd the lab and then the strategy which is coming from the leadership and that comes from Ethan Mollik's co-intelligence that framework so every organization trying to implement it should at least think about having those three pieces and how they could work together. All right. For a CEO thinking on implementing AI in his organization, what he should stop doing immediately? >> Um, stop delegating that to someone else, right? Like this is going

[54:19] else, right? Like this is this is going to if you truly think AI is the most important strategic work your company is going to do, you should do it yourself or at least get some people to work with you. But don't just ask your competent people to execute. they already have a lot on their plate and they're probably doing all kinds of other things. So the first thing I would encourage people to do is to play with it themselves and get a really good understanding of what it can and cannot do and then work with the smart people in your organization. But if you don't interact with it yourself, then we go it goes back to the lazy

[54:50] then we go it goes back to the lazy leadership idea. >> Okay. Love it. To policy makers, what safeguards matter as AI reshapes works and trust? >> Wow. policy makers. I would say it would be great if all of you could get together and establish one standard for everyone, which I know is really difficult. Um, but that is what most companies would like, right? We have global standards or regional >> support that I >> but I mean like come on like that would be nirvana if people can actually do it. Uh, but I find that to be very

[55:20] Uh, but I I find that to be very difficult. I would say like having some sort of um and Europe has the AI act, right? I think having some sort of federal level view on what this means that is meant to spur innovation but also takes into account real issues with data privacy trust things like that if if there's something an umbrella direction of what things mean um and that you can evolve over time so that there is a bit of guidance uh for companies right I mean we know cyber security is huge that's already being discussed in the news right but the idea

[55:51] discussed in the news right but the idea of fair use the idea of using AI to create deep fakes and things like that there there's going to be some societal backlash from that. So, the sooner that policy makers can at least start to develop a point of view on it before, you know, I would say the um the court of public opinion decides for themselves. >> Love it. One last more, I promise. What would be your piece of advice for a younger version of MJ watching her today getting ready to start transitioning to

[56:21] getting ready to start transitioning to this future with AI? Huh. I would just say maybe to other younger people how about that um I would say I mean this is very basic advice but I would say like one read okay don't give up reading because I've heard a lot about how people are trying to take shortcuts with just you know taking the easy knowledge head first and two remember that learning comes with friction and it's important that you have that and if you are trying to learn

[56:51] have that and if you are trying to learn without friction then it is not learning And learning and learning how to learn is one of the most important things in life and AI cannot teach you how to do that. >> My friend, what a beautiful advice. Thank you very much for dedicating part of your day to be with us today. Um I have really enjoyed this conversation. So I cannot appreciate you more. >> Thank you so much. I'm really happy to be here. Thanks for spending an afternoon with me. >> Absolutely. And for everyone watching out from home, uh there's going to be

[57:21] out from home, uh there's going to be another opportunity next week. Share this with your friends, with your colleagues and talk soon. Take care everyone.