What Netflix got Right and Why Most companies are built WRONG for AI! - Dr. Mik Kersten

0:00What Netflix got right is that alignment of the technological architecture, the team structure, and the leadership structure, and hopefully the org chart. But a lot of companies, those are misaligned. And as soon as those are misaligned, you've got these three or four competing structures, which have to be coordinated all the time, which means there's all these meetings and reviews and documents and presentations and so on. And that's what creates all of those basically weight states upstream and downstream of teams. And it just does not scale.

0:30What would happen if we just told the truth? Welcome to TruthWorks, where we dig into the nitty-gritty of leadership and work. And what needs to change? I'm Jessica Neal. And I'm Patti McCord. Our journey together started in HR, but trust us, it's evolved into something wild, honest, and well, a bit rebellious. So throw out the handbook. We're here to redefine rules to work for us, not against us. Let's dive into another episode.

1:07Hi, everyone. Welcome to another episode of TruthWorks. I'm your host, Jessica Neal, and I'm with my co-host, Peter Clark. Hey, Pete, what's up? Hey, Jessica. How are you? Good. Another new background. Yeah. Where are you? I'm in our Palo Alto office today and then tried to find a room with not too much clutter in the background. Or a big whiteboard. I was going to write some equations on the whiteboard behind me and look like I was smart, but I didn't have time. Well, you know who could actually do that is our guest today. I'm pretty sure he could do that. And then I figured you'd call me on it.

1:43So I figured it was probably safer to not write anything like that. Yeah, pretty, especially with our guests. And today I'm actually very excited. We're talking to Mick Kirsten, who is a founder, an author, a scientist. And he talks and researches a topic that I'm very, very passionate about, as you all know, on this podcast, because we talk about it a lot. But it's really digging into how organizations work and how they get in their own way. And he's writing a book right now, which he's going to tell us a lot about. It's about to launch. But it really tackles a topic that I think is important for all of us to be thinking about.

2:21And with AI, we all talk about AI all the time, but it's a big thing and you're going to need to focus on work and how your organization is structured a lot more. And Mick is here to tell us all about it. Welcome to the show, Mick. I'm so excited to have you. Thank you, Jessica. Thank you, Pete. Great to be here. Well, first, why don't you tell us a little bit about you and how you got into this work and tell us about the book that's getting ready to launch, I think, next week, correct? That's right. Yeah, it's launching on July 14th. So exciting times. Exciting Biomix book. We'll talk about that at the end of the show.

2:59But how did this all start for you? Yeah, so I spent basically a decade as a developer. So just writing software, a lot of open source software as a researcher as well at Xerox PARC near where you are, Pete. Yeah, yeah, yeah. Yeah. And back there, I think the new thing with Xerox Spark is there's always an intersection of people and technology. And I was very interested in that and really worked a lot on connecting the way that we build software with the way that people work. And I ended up doing my PhD in it. For my PhD, I founded a company called TaskDop, was CEO of that for 17 years. And in that commercial context, we built these tools that connected the way people work across organizations.

3:41And what I realized is just how disconnected that work across organizations is, because we saw the flow of data and collaboration and things like tasks and projects and products and initiatives across all these tools. And so we studied that data with kind of the research backgrounds that we had on my leadership team. And we realized that there was just these massive disconnects in how organizations are wired that really get in the way of productivity, of employee engagement and of, in the end, customer outcomes. So I wrote a first book about that called Project to Product, which took that initial data and said, okay, we need to actually not structure things around project work where the business is throwing stuff over the fence to technology and the way these enterprises work.

4:21Because we already had that whole generation of startups, unicorns, cloud native companies that figured out better ways of working. But somehow, a lot of large enterprises never actually made that shift. And then we did another round of research on that with fresh data, with additional data from even the initial data set was 3 ,600 different value streams of various organizations. The next set was 8 ,000 value streams. And we realized these problems are improving very slowly. And all of a sudden, what's happening is that the ability to amplify knowledge work outputs, whether that's code or campaigns or any kinds of digital outputs, was not really going to work.

4:58because all these organizations were structured for maximizing outputs and not actual outcomes. And so that's really where two years ago, and by the way, I spent a lot of time building out an agentic solution at the company that acquired my company at PlanView, learned just how powerful that was going to be and realized that these 10x productivity benefits will be real, but none of them will be realized if our organizational structures remain the way they are today. Yeah, because you can do 10 times the work, but that doesn't matter if your organization can't absorb it or, you know, put it in the flip, right?

5:35That's right. Exactly. Like you said, we tend to measure outputs versus outcomes, right? So yeah, here we did a bunch of stuff, but was it worthwhile? yeah exactly and i think what's what's happened is that we've been in this and i kind of in the book out but outcome this new book i look at it through the lens of these technological revolutions because every time we've had these major technological revolutions uh we've they've come with new management systems and new organizing principles right organizational charts came way back from the age of railways steam and railways right because all of a sudden in the u.s you now didn't have a bunch of people in a factory you had this very distributed organization that had to make trains run in time across North America.

6:16And we still have those managerial tools of organizational charts. This age of software and digital that really was the last, really since the 70s, since the microprocessor, took over the peaches near where you are, the peach orchards where you are, Pete. We've had these new organizational principles around agility, around product management, around technology and innovation that have really created a new set of management systems that some companies wielded really well and became these new tech giants and unicorns and that other companies just never quite adapted to, which was what was showing up through our data set.

6:49And I can talk about that a little bit more. And now I think with this, every one of these revolutions means some kind of constraint on production just becomes easy and scalable. And now all of a sudden, the constraint we had on, let's say, developer productivity, knowledge worker productivity is gone or it's quickly disappearing. So we're moving from a world where the thing that organizations optimize for so much, right? Which is the constrained productivity of development teams now is going to be unconstrained. And what happens then? Because all of a sudden you can build, where you're going to build one app, you can now build 10 in parallel and in a day, not a month or not a year.

7:28And how can the organization, the market absorb them? Do you have the right organizational structure and feedback loops? And what the data showed us is most organizations just don't. They don't have those structures in place. No. And they're likely not moving on it very quickly right now. Yeah. Yeah, is my guess. Yeah. Talk more about the data and the companies that were unable to make the shift to become the tech giant or the unicorn. What were the patterns and the themes that you were seeing with those organizations? Right. Right. So yeah, so what we were seeing and the actual data and sort of the most key finding that we found is when we studied the end-to-end flow of value all the way from ideas and initiatives being planned to when they were delivered to customers, only 8 % of that end-to-end time was teams actually creating value.

8:23was a value creation done by teams and the rest was upstream processes and planning and approvals or downstream reviews and security and so on and so you've got this world basically these these value streams and these organizations where if you multiply the productivity of that team by two so all of a sudden they can do twice as much you're actually not getting any more value at any faster to the customer, to the market, because they were not the constraint. The constraint is upstream and downstream of the teams. Now try to multiply that by 10, and then you still get nothing more. And that's why we're now in this world where there's some amazing work that's being done by teams adopting AI and token maxing and doing all those things, but it's not translating into customer outcomes because the organization is not structured to translate to those outcomes.

9:13And we realized this runs very deep because if you're going to actually get that kind of fast feedback loops from teams directly to customers, you need completely different kinds of ownership and empowerment structures, which I think we saw a lot of those created through this era of software and cloud. But again, a lot of organizations have not adopted those. And I think the ones who move faster to be just really empowered, more autonomous teams directly connected to customers are now able to leverage AI much more quickly than organizations who still have these legacy ways of working. 8%, so crazy.

9:47point. So yeah, why do you think? I mean, the data's out there, right? The visibility of successful organizations, the things that they've done to be successful, and yet here we are, 8%. We still seem to struggle with sort of embracing and adopting and adapting to change, right? Like, what do you think is the biggest sort of blocker for organizations actually making a significant change? Right. And I should qualify this, the data is through to 2024, right? So it's really before that kind of this new wave of ai and agentic engineering hit first of all right but my prediction is the number goes down with that but yeah exactly so what we that's right i think it's and that's really the challenge is that i think we've seen what good looks like in organizations we've seen it in again startups and tech giants who who are it's where that eight is more like 80 right we've seen it scale up to organizations the size of say say amazon who actually have that entire very hierarchical type of autonomy and empowerment with really basically 10 levels of cascade and so i think the and those are the kinds of structures and this is really some of the points i'm making up with outcome where you can amp you can apply ai and really amplify the the pace of delivery of outcomes to your customers but i think what's happened for a lot of especially larger scale organizations enterprises but also organizations who just have not grown up entirely cloud native.

11:17I think the cloud native thing really did help. That said, I can also point to a bunch of cloud native organizations who are in that like 8 % to 20 % range for the portion of time that gets spent creating value rather than in meetings and approvals and planning and replanning and budgets and all those other fun things. I think what's happened is that we've created this organizational wiring that's creating overly high coordination costs and not providing enough autonomy low enough in the organization. And I think a lot of that happened with sort of, especially organizations that went through digital or agile transformations, they basically added all of these additional ceremonies and matrix lines and all this additional process on top of their org chart.

11:58And they actually never changed the org chart. And so you've got these competing organizational structures. So you've got that, you know, the structure, which is your org chart. You've got the structure, which is the way that you deliver value, which is your kind of these horizontal value streams or ads always have working. And it's got its own set of meetings and ceremonies and reviews and so on. And then you've got the incentive structure of leadership, which is based on some kind of objectives and NBOs or OKRs and so on. And then you've got the technological structure, right? Which is the actual, the architecture of the organization in terms of what good looks like as those things are more closely aligned.

12:32And I actually do bring up some stories of Netflix, Jessica, because Adrian Cockroft was one of my inspirations through this. and he helped us. Yeah, I hired Adrian into the company. Yes, I did. I found him. Uh-huh. I sure did. My gosh, that's amazing. Yeah. Yeah, he definitely inspired me and how I scaled my own company and provided me a lot of mentorship, which I then reflect and output the outcome. I tell some of the stories of my experiences with him teaching me. Yeah. But I think what Netflix got right is that alignment of the technological architecture, the team structure, and the leadership structure, and hopefully the org chart.

13:09But a lot of companies, those are misaligned. And as soon as those are misaligned, you've got these like three or four competing structures which have to be coordinated all the time, which means there's all these meetings and reviews and documents and presentations and so on. And that's what creates all of those basically weight states upstream and downstream of teams. And it just does not scale. And with AI, of course, the scaling pressure is now 2x, 10x, whatever it's going to be as frontier models continue to improve. but I think we just have to completely revisit the organizational chart. Yeah, no, I mean, it's true.

13:44And I mean, I wouldn't say like we evolved, you know, our structure over time for sure. But I think one of the things that even as we were getting big, we still stayed fairly flat. Like we never implemented, everybody was that you were a senior engineer or a senior IC. Then we had manager, director, VP, and C-level. And that was it. We didn't have like senior director, senior vice president. We didn't, even in engineering, we resisted adding like architect or principal or it's like, no, everybody's an architect. Everybody's a principal. Like, you know, we didn't want those, like the complexity and all the layers to, you know, not have people take accountability and responsibility.

14:26Right. And, and, and so every year it was like, oh, can we, you know, add a layer? It's like, no, we can't add any layers. You know, because, and mostly people wanted to add, and I think companies do this and they do it for well-intentioned, but I see companies as they're scaling, they're like, okay, well, we want a ladder for people to grow, right? And so then they add, oh, senior, senior engineer and senior director. And then it ends up doing the opposite of what you want it to do. It just makes everybody slower. And you can pay people at a senior, senior level. They don't need the title, right?

15:06And when people are like, oh, that's my job. And then that's not my job. So I'm not doing that. That's when the complexity starts to kind of enter the system. And we just never wait. resisted yeah actually and i've been guilty about myself as over engineering those title levels right it's for with the right intentions right and a lot of engineers wanting to dissect them into yeah all sorts of bands and all sorts of titles and all sorts of functions and so on but yeah i think what we're seeing is when that becomes when you have to go up a level or two to ask permission to do something that slows everything down so if you've got a culture or organizational structure that's that's really all around empowerment of teams and lower down the org chart you can move fast but when you've got a culture where you have to go ask permission up or kind of beside you in the in this in this structure then things go slow and then there's this concept actually it really comes from uh gene kim and steve steve's book buying the winning organization that comes out as one of the models in Al Patelcom of independence of action.

16:12If teams don't have independence of action because they don't have that kind of empowerment, because you created the wrong kind of hierarchy or two hierarchy in the organization, or you just don't have the right mechanisms of ownership and empowerment in it, things just go slow. What do you think is going to happen to the companies that don't get on top of this? I think they're going to spend a bunch on tokens and AI. budgets i think their cfos are going to notice and i think they'll they'll basically i think their cost base will change because they're going to continue investing to some level because there's so much pressure to do so so they're going to reallocate and this obviously this is i think terrible outcome reallocate a significant portion of their budget to tokens and away from humans um and they're going to end up with lower business performance not higher and then they're going to try to adjust by rehiring humans but then they'll be further behind because they they won't have fixed the fundamental problem which is again the organizational structure so i think in the same way that we saw companies that were too slow to move to kind of the digital native organizational really it's about operating models right it's organizational structure plus all the mechanisms which which is really the the operating model um the ones that were did not shift quickly enough to those operating models were you know unless they were in kind of captive markets they were they were disrupted so i think we're just going to see a much faster rate of disruption of companies that are not in these captive markets with the companies who do figure this out and end up and have ai native operating models being able to capture more and more of their markets and the book by the way the whole goal of the book is to say well we actually don't want an economy where you know 10 companies are worth 10 trillion each and have the entire thing we're better off having a more diverse a company where a lot of businesses, whether they're startups or incumbents, do transform, leverage these new ways of working and get these productivity benefits of AI.

18:15Yeah. I mean, it's so interesting too, because you see these AI native companies who don't have that scale and that complexity, and they're just doing it. And they're doing it with like 10 people. Just like, okay. But if I'm the CEO or the CHRO, COO of this bigger company, what should I do? Because it's really hard to undo all of that. It's a big system. And you can't just scrap it and say, we're starting over. Or can you? I don't know. or can you yeah i think that's the i think that's the key question so i so the guidance and the the output tech and book has these mechanisms which are just these these different models mental models that can be whiteboarded and the ideas and comes with these actually they call them the seven shifts where you can kind of pick and choose each one to get started on this because i think it is the challenge is it's a really fundamental and issue because it has to do with actual operating model of the organization so yeah changing them overnight is not usually realistic i think just got to get back to your point that it's a fully ai native companies like that one of the case you know i've got a case studies there of these big enterprises in the book like you know vanguard or come of bank of australia but also like cognition ai right who grew extremely quickly with an entirely ai native operating model and a cadence basic planning cadence that's a week not a year and a kind of unconstrained output environment where they're able to throw away more of what they build that week than they keep and pass on to their customers.

19:54So they're able to hypothesis test in parallel and do all these sorts of things. So I think the good thing is we now have or are starting to see enough examples of these fully AI native ways of working. We're starting to see larger companies who did structure themselves well through the age of software and digital, like Netflix and AWS and parts of Microsoft and all these like a whole lot of the startups who thrived who are leveraging ai effectively and then so i think it's it really is a question of how quickly you can move a larger enterprise to those models and i think the at the core of that is this harder problem which is these organizations are they have again they have these separate structures it's really how quickly you can deploy what i think has to be a kind of a single unifying structure which where the basically the org chart the ownership structure the the value stream structure where you're delivering to customers in terms of products and platforms they're just one it's just it's just one tree structure that that i call the outcome tree in the book where you've got this cascade of basically inputs like strategy and budgets going down all the way to teams now of course some of those teams are fully autonomous they're just they're just agents but it's a human leadership structure because in the end the book actually proposes humans should continue reporting to humans which i know i don't know if that's I think generally it's a good idea.

21:18Yeah. I outlined quite clearly why this is actually a control structure for leveraging agents to do what we want them to do, rather than us thinking agents will make better decisions than we do. There's actually a whole theory behind it that I think is really important, which is that there's some things that AI is just not going to get that much better at than humans. And those are highly complex systems with emergent behaviors, whether that's stock markets or weather or decision-making in large enterprises or any size of company, really. But where you basically have this unified structure. So to me, it's a question of how quickly can you move this organization away from the crazy matrix stuff you put in place and all those processes to this single structure where all of these things are aligned around this cascade of, and hierarchical cascade, interestingly, because that's the only thing, as Adrian Cockroft pointed out, to me that really scales in nature and organizations is these self-similar hierarchies where you've got kind of encapsulation and ownership at each level, right?

22:12that the CEO should not be telling teams exactly what to do. Now, how many branches you have at each level of the outcome tree? Obviously, you've got examples like NVIDIA where you might have 60 directs from the top to the next level, but you still have ownership as you go down. You still have encapsulation. You have direct connectivity to customer outcomes as you go down this tree. So it can be a very flat organization with very broad branches. I think Netflix is another example of that, as you mentioned, Jessica. but you i think organizations need to move that structure and away from these these again parallel structures where let's say again the the product value streams and platform value streams are completely misaligned with the incentives of the leaders behind them and that by the way is back to that issue with the eight percent a lot of these transformations failed because organizations put had the technology teams put in place the right structures nothing really changed that significantly on the business side, the compensation structure of leaders did not change significantly.

23:12So no one was investing properly in platforms, right? What good looked like in terms of technology organizations the last 10 years or so, 15, is around half your investments, broadly speaking, were into platforms. What it looked like in enterprises, it was like 15, 20%. And that's because of course, you didn't change the incentive structure of the business leaders who were never properly incentivized to invest in platforms because there wasn't a direct customer outcome that next month from from investing in the platform so interestingly and bizarrely i actually did spend like half a chapter just on incentive structure in this book yeah really yeah which i didn't think i'd be writing about but i thought okay someone's got to write about this say more say more well it's it's again it's making sure that with this outcome tree where this is really the structure let's say use okrs use some kind of objective tracking system there's this cascade of ownership at each level of the tree you've got a leader you've got a team of humans and agents and their objectives and their incentives have to be the same thing and their objectives have to be 100 on outcomes 0 on outputs and the team has to have full autonomy on what outputs they build to deliver those outcomes they have to have a budget like they can't blow through the whole company's token budget in the next month.

24:29So they have an input of a budget and that's be lying to the strategy, but they have full ownership of outcomes and the leaders incentivize purely on those outcomes and it cascades down. Which by the way, that's exactly how I had set up my organizational structure. And of course evolved it many times as task top scaled. And I think a lot of others, these structures have been proven out. But again, if you look at cross organizations, especially enterprises it's the it's the exception not the norm i say one topic that comes up all the time for us is performance management i think we were just talking about it in the last one but yeah i mean you get the behavior you incentivize and yet we seem to be surprised by that so often right like well you know why are they doing that it's like well it's because of how you created your incentive structure right and honestly it seems like we reward outputs over outcomes most of the time, right?

25:22It's just like activity over results is what gets you your bonus or, you know, how you're incentivized. I think just because that part feels, and maybe I don't know if that's the case, feels easier to measure or, you know, outputs activity, we can say, look, look, all that, I did all this stuff. Whereas outcomes in terms of business impact seems to be harder to translate. Yeah. Yeah. And I think that's part, that's a huge part of the crux of it, right? Is that outputs are easier to measure we've been i think we've got management systems that are very good at measuring outputs right and a lot of those systems come back from just really management systems that came from let's say manufacturing right like if you're if you're yeah widgets exactly it's about widgets so if you're manufacturing cars you've got like the number of cars your your factory produces and the quality of those cars and can is going to be a measure of your success because you're output constrained, right?

26:20But now all of a sudden, if you could, and this is kind of around the corner, I don't know if it's 10 or 30 years or three years, but if you have a dark factory producing unlimited cars and the cost of the cars is really just the cost of electricity, then you're no longer output constrained. And measuring how many cars you're delivering becomes meaningless because really all that matters is how quickly you're able to deliver cars that delight customers and that grow your market share and then increase your profit margins and all those business metrics. So I think it made more sense. And also we were fundamentally output constrained.

26:54The reason so many of the management systems around agile and everything else, there's so much of it was around prioritization of the scarcest resource, which was your developers and that productivity and all your other knowledge workers. And again, that's changing right now too. So that constraint gets removed in the age of AI. And so yeah, the management system has to shift from managing outputs to outcomes, but the challenge is it's been easier to, and I think I just want to unpack this a bit more because I think it's such a fundamental issue. It's been easier to measure outputs, right? We're very good at measuring pull requests and features and vulnerabilities fixed and tokens consumed, like all of, well, tokens consumed are actually just a cost, but a cost because it's easiest to measure cost metrics.

27:35Then the next easiest thing is output metrics. And then the slightly harder thing is outcome metrics. but i think that's where you know that's given how much more quickly right now you're able to iterate one of the reasons output metrics were sorry outcome metrics were were harder to measure is the lag they were lagging indicators right so you would build some cool mobile experience you'd add chat to it ai chat and so on and you know previously it would take time before that translate into better net promoter scores from customers or or better uh or just market share or arr growth or those sorts of things but we're now because of kind of removing the constraint on outputs that's able to happen so much more quickly that i think outcomes are actually becoming easier to measure because the loop moves faster you're going faster through customer iterations you can do parallel iterations if you kind of farm out agents to do these sorts of things and again they become the only thing that's that's really meaningful in the age wage age where outputs just trend towards the cost of electricity.

28:38Yeah. And so if we're talking about, you know, sort of alignment and outcomes and empowerment, you know, kind of in the system, how important do you think things like OKRs are anymore? And I've never been a fan of bonuses to begin with. I think they're silly because, again, and I would say in the world of AI, things are changing even faster to what the goal is, right? Because, you know, especially if you're in a startup, you don't know what the next six months is going to be. You're making it up as you go along. So why are we going to bonus people on X when actually we want them doing Y? And we figure that out in the first three weeks of setting the bonus, right?

29:22So, I mean, how important do you think those types of goal setting and like just even having a bonus is anymore? Do you think that those are relevant yeah i think those are really interesting i know i certainly experimented with zero bonuses the zero bonuses for sales people is fascinating that was a very short-lived experiment yeah i think for sales it's still it still works but for engineers yeah i don't know actually where i left like as we right before we sold desktop we were still zero bonuses for engineers or engineering leaders yeah and of course commissions for sales and and then and And then other bonuses, the rest of the bonus is purely, purely like 100 % based on OKR attainment.

30:06So I think there are kind of two things here, or maybe there are probably more than two things. But one is kind of with AI changing the pace of things and both for established companies who may need to evolve their business model or where their business model is impacted as well, or for new companies who are just going to pivot their way through finding just product market fit while frontier models change all the time and their harness becomes obsolete the next time Anthropic releases something and then those sorts of things. So I think the key thing is we're in a world where the pace of change is obviously insanely fast right now.

30:43And so these long horizon things are problematic. At the same rate, if you're you know whether you're deploy there needs to be some some some kind of longer term planning so i think it just becomes a question of what's your pace of planning yeah because you still see the aina of companies who need to plan on a quarterly basis like quarters are not going away right i just don't and so and actually rather than saying everything's going to turn into completely dynamic planning and so on the approach i took with output to outcome is to actually say well just just have your core mechanisms again at that team level just be much faster where their outcome loop is a very fast loop but it does cascade up into a weekly monthly and quarterly structure like you just and maybe you can do your budgeting like you know if you're moving fast enough you can do your your bud you know like replan on including budgets at a monthly and quarterly basis rather than annual but there still has to be this cadence where you're bringing in the various stakeholders, sometimes board members and others, if you're making fundamental shifts in what you're doing.

31:50So I actually think that, and I think it'd be interesting to debate this, that OKRs or a structure that's similar to that are still very important because they have to... The other tricky thing with outcomes is that people, it's very easy to define activities. Like we launched X, we hit our 1.0 goals and so on. We built the next model. Whereas I think the the best practices on OKRs, and certainly what they put in an output outcome, is they actually have to be quantifiable. They have to be numbers. And those numbers have to be based on outcomes, not activities. So all of a sudden, you know, if you've got, you're able to plan those on shorter timeframes, it's more work to determine those outcomes.

32:27But then you can actually, you know, put agents on, you know, a team of people and agents, or in some cases, it's an autonomous value stream with agents fully running the loops. They're the ones delivering on outcomes and reporting that to some kind of human. It'll be the first line leader in the organization at that point. But I think in the end, there has to be a planning mechanism that's based on quantifiable outcomes. Because at this pace, and especially with significant amounts of agents in the loop, it's the only way you can iterate quickly enough. And I think that pace of iteration shortens obviously the ability for you to plan out a year in terms of people's bonuses and so on i think is is uh is very tricky but i do think having that at least weekly monthly quarterly cadence of planning through okrs or something similar is is key yeah i think planning is very important and and you know and that's where you really you get alignment you figure out what worked and what didn't and what you're going to do next and why.