Dave West  0:00  
Hello and welcome to the Scrum.org community podcast. I'm your host Dave West at Scrum.org. In today's podcast, we're talking with Daryl Fernandez about his recent paper that was called "The Hidden Management Gap in AI Adoption. Welcome to the podcast, Daryl. Always enjoy these conversations, Dave. Very much looking forward to this topic. I know it's it's it's a pretty. It was actually a bit of a bit of a a surprising paper when you sent it to me to review. I was struck by the irony of the phrase management and AI because on one hand, I'm seeing obviously you know in the press etc. and talking to companies, I'm seeing AI being used as an excuse to lay off lots of middle managers. And then, on the other hand, when you're using AI, I personally see the importance of being more like a manager with it. You know, reviewing work, providing critiques, telling it when it's wrong, telling it when it's right, telling it to lean into this and maybe stop doing that-all traits that that I'm not very good at, but I hear are management traits. The other thing that was interesting is, as you sent me this paper, I just provided a technical review of Mick Kirsten's new book, which is called Output to Outcome, and he in that book there's a section that talks a lot about the need for everybody becoming a manager because of AI. So, considering that, what was your motivation for writing this paper?

Darrell Fernandes  1:40  
You know, taking a step back, it was really interesting to me. I was in a dialog with a software engineer here at Scrum.org, and we were both using AI to augment the things that we were doing. and And the way we were referring to AI was a little bit in conflict. And his perspective was, AI is a tool. It's no different than a hammer. It's no different than a software compiler. It's just another tool that I use in my job, because he he in his role as a senior IC looked at it that way. That's the background he had, and that's the way he looked at all the things that came into his sphere of how he most effectively accomplished his role, I, with a more of an organizational leadership background, was referring to AI as more than a tool, as something that I had to be accountable for. I thought through what that accountability looked like. I thought through my responsibility in making sure I was guiding AI the right way that I was clear on the outputs that AI was giving me, and that there weren't any biases or hallucinations or anything else. Not that he wasn't doing all those things. I certainly believe he was, but we just had such a different perspective on the role that AI was playing, and it dawned on me that depending on how you've grown up through your career, you could very easily have different perspective. You could very easily have different perspectives on how AI fits into your world. And what dawned on me more than anything else was the accountability model, and as a leader of an organization, you always have to be accountable for everything that happens in your organization. There's nowhere to hide. Whatever is going on on a day-to-day basis, you have to own that, and you have to be on top of it enough that you can be aware of the right decisions being made. It's it's where, you know, it's where the empowerment of Scrum is so important. You have to be able to empower people to do what they need to do to get the work done and to deliver product. But you also have to know enough to be accountable to help guide them, to help guide their business partners, to help guide to the outcomes and through the constraints that are inevitable, and so in my world, that's kind of how I saw AI, and and it started with that. Well, we've got to think about accountability, and then as I backed it up and looked at the paper we had previously discussed, which is the hiring process for AI, if you will, all of a sudden it dawned on me that we haven't done a great job of helping our frontline team members become good managers, and we know this. The data tells us this for decades that we, as industry, don't do a good job preparing the most impactful senior individual contributors to be good empowering leaders in their roles, and so I was I was sitting with this, and then I said, "Well, I wonder how the theories and the approaches of Michael Watkins, the the theories and the approaches of Patrick Lenchioni would apply, because those are two people who I certainly pay attention to in my organizational leadership days. So I decided to do a little work there and to do a little. Research and to understand, do the things that they thought about, do the considerations that they had, apply to how AI fits into this this new work model that we have? And the reality is, it's interesting, and you could you kind of have to squint a little bit, but they do. I I saw patterns that led there, as we think about AI.

Dave West  5:24  
So the one thing that I got from actually both the papers you've written recently on AI, whether it be the sort of hiring your AI teammate or this paper, was something that I think a lot of people have a have a problem with, and I want to bring that elephant clear front and center. AI isn't a human. It isn't. You could argue whether it's conscious or whether it's not. That's sort of like I'll leave that to the philosophers and the you know and the New York Times writers and the ethicists, the people that I speak to instantly because they have a bias. We're like, this is a tool. This is not a human being. This is not a team member. They step away from using some really powerful mental models for using AI in an effective way. Is it? Do you agree? Is that something? Is that perhaps the first step to success? Don't worry. Don't bring the politics. Don't bring the philosophy. Don't bring the. Just ignore that. These are just ways of approaching AI with a certain outlook.

Darrell Fernandes  6:30  
Yeah, I I would agree. AI like who cannot agree? AI is not human. AI is not a person, right? It's it is a resource, but it's more than the kinds of tools. Yeah, that we've used in the past. It's not a compiler. It's not a database technology. It's not a portfolio management tool. It's different. It has the ability to consume. It has the ability to infer. So there are characteristics that are in in a gray area that are more than tooling and certainly not human, and I think I think ignoring that gray area is dangerous. I think it can lead to some some ignorant biases that you apply, and ignorant biases probably not the right word, but unconscious outcomes maybe is a better word. So it could lead to some unconscious outcomes because you're not paying attention to that gray area, and you're not paying attention to the the fact that it will infer. Not that it can, but it will infer, and it and it may not give you the same answer multiple times in a row, so you have to be engaged differently than a database technology. When you query a database technology, same input on the same base data is going to give you the same output every time.

Dave West  7:52  
Yeah,

Darrell Fernandes  7:54  
the fact that is that AI is not going to be that, and I think you have to push to the side those debates and be realistic about. I need a new set of tools to run this thing. I need a new set of capabilities to run this thing. And for me, and it's it's all my own personal bias when I write these things. For me, those tools came from being an organizational leader, and so I look through that lens. So my writing and my papers may lean towards that human element, and that may that may push some people away. It may build some detraction from the content. I hope people can put that aside, know that that's just the lens I bring to the problem, bring to the opportunity, and I'm applying that in a way that I hope can help people get to that gray and navigate that gray with tools that, again, most individual contributors have never had the opportunity to be exposed to. We've not done a good job as organizational leaders of sharing some of those techniques, some of those capabilities, broadly across the organization, and to to the point in Mick's book, we all have to upskill here. We all have to step into that gray with new tools, and maybe maybe there are brand new tools that we have yet to define that are that are better than anything we've had before. But for now, what I see is the application of those leadership tools can be valuable in how you think about how you consider your approach with AI. I

Dave West  9:32  
think the one thing that both the papers have done, and this one in particular, is it's made me just step back, and it made me just use a different, you know, sort of like point of view to look at something, which I think is really, really exciting. And you get the ability to experiment with these ideas, and and just ultimately get the AI to be able to help you do more stuff, which is all. So you talked a little bit about Watkins and some. The other research that you base this on, so tell me a little bit about this capability gap and the research that Watkins did on the subject. This capability gap around management, etc.

Darrell Fernandes  10:11  
Yeah, so so you know you can look at Watkins' work through a lot of different lenses, and and he's got many management approaches out there. I'm going to jump into the first 90 days, Dave, and start with with one of the key points that Watkins makes in the first 90 days. It's one of my absolute baseline foundational books of of a being a manager, right? And so the first thing is what got you here won't get you there, right? That's one of his core principles in the first 90 days, and when you think about it, letting go of some of the things of how you used tools in the past is no longer going to work with how you use AI, and you've got to recognize that you need a different approach, right? And that's just that's just one one lens that Watkins brings to to this this area of how he thinks about human management, and how you can take how he thinks about human management and roll it back into working with AI, and it's I think it's just a really interesting, powerful, and simple example of how the theory can be applied in that gray area between the two.

Dave West  11:39  
So let's choose a real example. So an example would be I've always written code myself using IntelliSense, etc. So I'm not going to do that now. What I'm going to do actually is I'm going to instead of writing the code and writing the test harnesses and going through that whole process, I'm going to step back and I'm going to literally just give it an intent, and instead of going in and making changes to the code after that first time it fixed it, I'm instead I'm going to hold myself back, and instead I'm going to tell Claude Code what I think's wrong, and I'm going to let it fix it. So I'm going to basically stop my normal approach to writing code or writing a user story or whatever, and I'm going to only I'm going to interact not with the artifact, but I'm going to interact with the agent that's creating, or the the prompt that's creating that artifact.

Darrell Fernandes  12:38  
And I think that's a great example, Dave. And the interesting thing there is, while you do all that, while you make that shift, you can't lose accountability. It's still your code.

Dave West  12:50  
Yeah,

Darrell Fernandes  12:51  
it's you own that code. So not only do you have to do everything you just did, but you have to understand every call that that code makes and why every error condition the code handles and why every algorithm it runs and why like you have to be conscious about all of it and I think this is where it gets very interesting and we've seen the the data unrelated to this paper of yeah you may get code produced faster but the learning curve is significantly diminished if you don't take steps to own the code. Right? You can produce the code, but in producing the code, are you actually owning the code? And I think that's a different line here. And again, a manager doesn't write code in many cases, some managers do. You you at Scrum.org still tend to at times dabble in some code. I believe. I think

Dave West  13:47  
dabble, but don't don't tell the engineers because they'll be like when very sensibly. But I'm I'm yeah.

Darrell Fernandes  13:53  
But for the most part, you're not you're not a developer. You're not a software engineer. That that's not your accountability at this point, but if the the technology team, business technology team at Scrum.org, were to produce something that that caused a customer impact, you would certainly stand up and take accountability for it, and you would understand what happened, and you would understand how it happened, and that I think is the interesting piece, and that's where Watkins, with his with his approach is really clear that as you step up, your ability to understand everything changes. But that doesn't mean

Dave West  14:30  
you can ignore it. You have to find a mechanism to be accountable for it. And those mechanisms could be something which we do is we ask, and so it's very similar to the way which I work with human beings. Actually, walk me through the key parts of this code. Build a video that that so that when we're maintaining it, we can pick it up, etc. Or build build a a description, or write, do a UML diagram, or sequence flow, or a find. State machine or whatever. I remember when I was helping junior developers, and it was last century, which is worrying to say, that one of the instead of always being in the code, I would use other ways of exposing the code to me that are maybe a higher level of abstraction, etc. And but what I never did when I was trying to help somebody get better at that craft was I never wrote the code myself. I always just helped them through that. Okay, so that makes sense. That makes that makes sense.

Darrell Fernandes  15:35  
And the beauty of AI is we all have different. We don't. There's a standard set, but there are different learning styles. Some people learn visually. Some people learn through reading, hearing, touching. With AI, you can actually have it create a learning journey for you on the code that fits your learning style.

Dave West  15:58  
It's funny you should say that. I just been with my kids, we've been trying to. They've all got their summer reading books, and I know that neither of my kids have paid. Even though they've read the books, they've paid zero attention to those books. So I was just playing with Claude actually, and said, "Hey, let's write a song about this book, and because I'm like, there must be a mechanism to get this book into their heads, and so AI'S really good at those sort of different ways of looking at the same information, and and as a manager, you realize very early on in your management career that not everybody's like you, and it's so annoying, or actually, in my case, rewarding because you don't want people like me, and so you have to be very creative in the approaches. And it's, and this is the other thing you have to AI doesn't get things wrong. It's like there's no such thing as a bad pet. There's only a thing as a bad pet owner, and that that and I'm not equating AI to being a pet, though sometimes it feels that way. That you got to. It's how creative are you at finding different ways to help it get it right. And so you know when when you think about back to walk-ins in the broad management capability gap. Like AI has a need to be governed, and when

Darrell Fernandes  17:30  
you look at, and the paper does a good job of looking at what effective AI governance can look like and what many frontline workers lack, and it's it's just a really compelling grid of wow we need to get in front of this as leaders if we want if we believe AI is going to impact our business the way we're saying it is we need to get in front of this and make sure that we've got the structures in place for people to be successful as they use AI on our on the organization on the enterprise's behalf to deliver value, whatever that looks like in their role in their industry. It's a really interesting opportunity set, but I don't know that all organizations appreciate it at this point. I think there's so much energy and momentum happening that some of this is lost in the churn.

Dave West  18:33  
Probably. So let's lean in a little bit on that. That you wrote about six things that effective AI governance or management requires: role clarity, evaluating against strategic content, comfort of ambiguity, etc. Which one of these do you think that organizations, most organizations, are already doing quite well, or are all companies in sort of a bit of a mess? I know ambiguity, role ambiguity, and clarity was something that you used to press me on all the time. Your racy charts everywhere. Whenever we did so, for the listeners, I just want to call out that before I was working directly with Daryl, I occasionally did work for him in his companies, and every time we did it, he made me write a racy chart of who's doing what and where. You know who's responsible, who's accountable, who's involved, who's community. And I'm like, really, no. I'm just doing a workshop. He goes, no, very important. You're very into that. So maybe that's something we're quite good at. But in general, what? Which one do most organizations get right? Which ones should people actually think, oh, that's a surprise.

Darrell Fernandes  19:43  
Yeah. So, so I actually I've thought about this quite a bit because Scrum.org kind of talks to impact such a broad array. My background, obviously, financial services, but I kind of put a two by two together when I think about this this area, Dave, and the. Governance, effective governance, and where companies are, and the two by two is really on the one axis. You've got how big is the organization? Because smaller organizations have different problems than large scaled organizations. A racy and a small organization is usually implied. Everybody knows because we're all talking all the time, and if there's a a gray area between what what Rob's doing versus versus what Kristen's doing, it's it can be cleared up relatively easily, and nobody gets too far out of line with that. In a large organization, when there's so many different things going on and so much scale, and and the accountability model is so much different, it's harder. And I think things like races become more important, which is why I always pressed on them. Because at the end of the day, project delivery challenges usually come down to a communication gap somewhere in the life cycle.

Dave West  20:52  
Yeah,

Darrell Fernandes  20:52  
and so understanding how the communication was supposed to happen and having the forethought to be clear about how the the communication was supposed to happen is really important. So you've got size of the organization because the problems are different, and then you've got industries, and I don't mean all industries, but you've got things like highly regulated industries: pharma, financial services, medical care, health, if you will, to more manufacturing side, which is less regulated, although more and more regulated as time goes on, and technology plays and automation plays a bigger, bigger role, but but less regulation in certain industries, and so the amount of regulation and discipline also has an impact on which of these. So so with that being said, I think there's a there's kind of a two by two here of a high regulated, high scale industry, I think that the you start to see a loss of a degradation, if you will, in the ambiguity space. Those industries are not comfortable with ambiguity. They want clear. They want specific. They want definitive and and understanding how AI kind of infers and predicts answers-that's new, right? And and from a from a tool, that's a skill that needs to be built. If I go into the small organization, less regulation-I won't say no regulation, but less regulation-I think the the challenge there comes down to the accountability because usually it's not that people aren't accountable in those organizations, but understanding how to how to be accountable for the output of AI is a more of a gap in those areas, and that's a gap across the board. And honestly, I could put that gap, but I don't think it's a gap in strategic intent. Everybody generally knows the strategic intent in a small organization. It's not as as confusing. There aren't multiple competing intents. In a in a general one, I would say is willingness to course correct and adapt. And that's a human condition. Yeah, and I see it more and more in these economic times as I talk to companies that they want to stay the course, and it's really hard right now because there's so much uncertainty in the environment. But it's a very difficult time, and I think AI exacerbates that because of the the nature of the inference and the nature of the response set that you can get,

Dave West  23:28  
yeah, I think it's funny. Clear strategic intent is something having a clarity of why we're doing something has such a big impact on AI and the ability to ask that over and over again, I think is is crucial. All right. So that

Darrell Fernandes  23:48  
before I'm sorry, Dave. Before I move on, I I think it kind when I think about working with AI, like the the the seven Ws of journalism, right? Who, what, where, when, why, how. If you can't answer those when you're interacting, like what AI am I? Who? Which AI am I using? What am I asking? Why am I asking it? How do I want it to be produced? When is the and when can have multiple variants here? Like when is the learning training data as of? When is the input as of? Like there's so many pieces that come back to something as simple as who, what, when, why, how, and we forget those simple lessons. I know races are a pain, but they're a simple tool that has so much power. But sometimes they're so simple that we want to ignore them because we're supposed to be better than that.

Dave West  24:39  
Yeah. All right. Let's talk about failure modes then. Let's move to failure modes. What's this automation deference person? I, I, you know, it sort of stuck. It sort of stood out. The lovable digital slacker. How will leaders discover these failure modes in or? Organization and I love that automation deference person. That sounds like a strange political party. I'm a member of the automation deference group. So,

Darrell Fernandes  25:08  
so this is this is your cruise control person, right? They fit in the organization. They're generally liked by the organization. They do good work. They they produce just enough, but there's always something that could have been better. There's always something that could have been more. There's always there's always another step that should have been done. And so when we think about automation deference, that's the person who comes and does a wonderful prompt to AI. This is where you get the the art artifact effect comes into play where AI produced a wonderful, wonderful word summary of a prioritization with all the pretty formatting and the good descriptors, but because you didn't review it very well, there was three hallucinations in there of customer quotes that actually never happened, so it's it's it's deferring to the the output as a as a so it's deferring to the output as a gold standard and a final version without taking the extra step as a leader to ensure that the output is accurate to ensure that you're comfortable being accountable for that outcome. And in many cases, in a large organization, those things can just float for a while before somebody asks a hard question, and then all of a sudden, people say, "Wait a second! Where did we get that? Why did we do that? Why? Where did that come from? And then you start to peel that onion back, and you realize that lovable slacker, if you will, who's using AI fairly effectively with the good prompts, but not the follow through on the accountability side, can certainly have a big impact. And it's one thing to do it in a backlog prioritization way; it's a very different thing to do it in a revenue reporting or a invoicing scenario.

Dave West  27:05  
Interesting. And how does the leader sort of find? How how do you discover? I said leader, but how does the person that's using it? How do you discover these these modes and then help these people?

Darrell Fernandes  27:19  
I think as a leader, you have to ask hard questions, right? You have to be willing. So I worked for a phenomenal leader at one point in my career, and and his mantra was trust but verify. You certainly have to trust that your team did the right thing, but you also have to verify that your team did the right

Speaker 1  27:37  
thing,

Darrell Fernandes  27:37  
right? And and so it comes down to a trust but verify model of because that person produced it, because that person presented it. If they're in your organization, and and whether that they is an AI instance or whether that they is a human being, you're accountable for it. So if you haven't taken the steps to ask some hard questions about it and verify what's there. You're you're accountable for something that may be incomplete, and and that's why I came back to the Watkins the Lenchioni models because I think a lot of the lessons we've learned over time are still applicable when it comes to the AI gray area that we're working in today.

Dave West  28:25  
Yeah, you know I think so too, and in fact they're just more obvious now, perhaps even though you have to approach it with a sort of different mindset. Well,

Darrell Fernandes  28:35  
well, obvious and potentially carry more risk.

Dave West  28:38  
Yeah, because human beings. Yeah, because you have, yeah, and because human beings aren't as readily in the loop, things can slip by, right?

Darrell Fernandes  28:47  
Correct. So I think it's it's it's very important to be even more tuned into it today, and the pace, the pace of output, and the pace of of generation is so much different today than it was even a year ago than it was two years ago. In our tools are not keeping up with the pace. Our tools to to to scale with the pace as humans, yeah,

Dave West  29:13  
understand aren't

Darrell Fernandes  29:14  
keeping up with the core tools.

Dave West  29:16  
And that I think brings us to a really interesting question: How do you balance the bottleneck and governance? Most organizations I see, in fact, even Scrum.org, dare I say it? So I'm with the help of AI producing more and more good content. Now, don't worry, listeners. I act. I'm actually involved in every single part of the content, but it's nice. I've now got a partner that I work with, so the the empty page is not a problem anymore. That I can produce so much more content than I've ever done, and yeah, it's yes, I have to still do quite a lot of work, and in fact, I'm even doing more work because of it, because I'm actually more interested. Because I'm like, oh, I didn't know that, oh. That's a piece of interesting research. Let's go and look at it. Yeah, AI is a fabulous partner in content development, but we have a governance bottleneck here because we do not put things onto the site unless another set of eyes has looked at it that we've looked at against style guides. And yes, we've built skills that check all those things, but I've forced in our company, which is probably wrong, that things don't get posted without a final set of eyes on them, and that that that can be a bottleneck.

Darrell Fernandes  30:33  
I think it is absolutely a bottleneck. I don't think it can be, and I think it's a bottleneck in development teams. I think it's a bottleneck in marketing teams. I think it's a bottleneck across the board, and I I think the whole point that the bottleneck can't be you elevate to senior lead more senior leaders all of those decisions. You have to the only way to scale with the tools is to prepare the teams to take on a lot of that validation, to take on a lot of that accountability in understanding, verifying within the constraints that your organization has the appropriateness of AI, and there is a bit of art to that. That there there isn't a recipe that you can just apply, which is why I went to the the Lencioni, the the Watkins frameworks, because much like scaling frontline managers, you have to now scale everybody. That's the answer, Dave. Honestly, is you have to prepare everybody in the organization to be ready to do these things, and you're gonna you're gonna have moments of learning. Some would call those failures, but you're gonna have moments where you're gonna you as the CEO are gonna say, "How did we let that go to the website? Now it's not catastrophic, but it may not be ideally on brand. It may not be exactly where we wanted it to be. Whatever the case may be, there are going to be learning opportunities, and you're going to have to coach through those so that everyone in the organization can build this muscle. Because because it's a muscle that most in the organization, by count, don't have. If you were to weight it by seniority in the organization and accountability in the organization inherently, I think the organization would probably have a good amount of it. But by pure count of heads who are ready to take this on, most of the organization does not have the full set of tools to fully appreciate how AI, good, bad, indifferent, will impact the outcomes of the organization,

Dave West  32:49  
and that's the reason why we have to, at the moment, put in place governance processes that ensure that those people are not left out to dry because suddenly they're delivering a lot more content.

Darrell Fernandes  33:00  
Well, well, and I think back to the the regulated versus more regulated versus less regulated. The more regulated is always going to need the evidence of that governance, but it

Dave West  33:10  
means that we need to be very explicit on what parts of our business need that governance and what parts don't. Correct. And historically, we've done a very poor show. When in doubt, particularly if you're a large organization with a lot to lose, you tend to assume everything does. Yeah, one size fits all, and it's the easiest model. And it's the easiest model. And what AI is teaching us is that you don't need to do that. So we could talk for hours about the paper, and I I do recommend everybody sort of reads it, or at least gives it to their AI bot to read and highlight the summary that's in need for for their context. As we leave, what should be our listeners take away? What would be the one or two things that they should take away from this paper or this discussion about the paper that they could start doing today, that will have an impact on their use of AI.

Darrell Fernandes  34:07  
So you tipped my hand a little bit here, Dave, earlier in the call. I I think the easiest place to start to to force yourself as an individual to reconcile where you are on this journey is as you're about to leverage AI in a workflow, in a code development model, in a code generation model, in a test generation, in a marketing campaign, whatever the case may be, as you're about to engage with the AI capability of your choice, write a Racy. Write write a racy on who what or who is responsible for the action. What or who is accountable? What or who is consulted? It will force you to answer questions like on the consulted. What inputs should I be bringing to the table for AI to consume to ensure the context is a. Appropriate,

Dave West  35:00  
yeah.

Darrell Fernandes  35:01  
On the accountability, am I ready to take this on? On the responsibility, who who are my who are my peer set that's covering different skill sets, different roles, different points of view in the organization that complement me? That I need to make sure weigh in on whatever I'm doing, and and so it sounds so simple and trivial, but it forces a level of thoughtfulness as you engage with the tool that I think will help people appreciate where they may need to think a little bit more than I have a great prompt. Let me go ahead and see what AI has to say. There's nothing wrong with that, by the way. As a learning journey, I think that's a phenomenal place to start. But if you're actually going to go impact your business or impact your enterprise, your organization, your customers with the output of AI, I think it's really important to to build that racy and make sure you have all your ducks in a row, so that you can feel great about how you're leveraging AI in that capacity.

Dave West  36:10  
Awesome. Well, thanks, Daryl. Thanks for spending the time with us today on the Scrum.org community podcast. It's always a pleasure to to have these conversations recorded. Obviously, listeners, we have many of these conversations anyway, so it's always good when we get a third set of ears, or hopefully more than just one set, listening as well. Thanks for being here today, Dal. Look forward to the feedback, which I know we'll get, Dave. So it's it's always fun. It is good, and thank you for listening to today's Scrum.org community podcast, everybody. Today we were very fortunate after Darryl Fernandez talking about his recent white paper, the hidden management gap in AI adoption. An interesting metaphor analogy. I don't know what the correct word is, but that really lets us step back a little bit from how we're using AI and apply some of those tried and trusted middle management management techniques and approaches, like the work of Watkins, etc. to AI, treating it in that way to at least approach it in a different way and hopefully improve the outcomes because of that. If you liked what you heard, please subscribe, share with friends, and of course come back and listen to some more. I'm lucky enough to have a variety of guests talking about everything in the area of professional Scrum product thinking, and of course, Agile. Thanks, everybody, and Scrumm on.

Transcribed by https://otter.ai