A (4:06)

I like that.

B (4:07)

Let's do that through their hats about it, but actually are not that close to it. And so your work at Google Labs, developing tools that are helping people to use AI in much more powerful ways, you're closer to it than almost anybody. And besides, in addition to that, you have a ton of interesting ideas about organizational structures and business structures. So I want to get some of the basics out of the way, which is what is Google Labs and what is Breadboard and what is Opal?

A (4:41)

Oh, these are good questions. To begin with, Google Labs is a fairly small but very fun org. This is probably the most fun I had working at Google, working at Labs whose challenge and job to be done is to find new interesting opportunities for the company to pursue. And hopefully some of these opportunities will be really big. Some of them might be very small, but looking for new things and looking around the corner is what we're about. I think this fits very well with the project that I've been working on, which is Breadboard. So we started Breadboard two and a half years ago now. And the basic idea was that, hey, people are playing with LLMs a lot. Can we build something for them that would allow them to combine and somehow translate ideas they have about using LLMs into something tangible, something they can play with and rapidly iterate on? And so we started this project. It took many pivots and turns. People who follow this project, they'll be like, I remember when it was that. It's like, it's not that anymore. I remember when it was this. It's like, it's not this anymore. And it's really interesting because by building a product that was meant to help people experiment, we ended up also, in this really weird meta way, experimenting ourselves because the field is so new. Even though the idea of Automation and flows and composition are all pretty existing. What we're dealing with is qualitative step of where do we want to play was very different because most automation tools. When I say automation tool, first thing that comes to mind. What is it for you?

B (6:32)

Oh, I think of process automation.

A (6:34)

Honestly.

B (6:35)

Yeah.

A (6:35)

Boring. That's the first word.

B (6:36)

Boring and difficult. Not just boring, also difficult.

A (6:41)

Right. I actually tried to like resist the word automation because there's so many of them tools out there. I don't want to be doing any of that. I want a person to come in and have fun and go like, whoa, I can do this with AI and I can do that. And ooh, gives me another idea. That's the kind of thing that we wanted to do. It's a different area to play. And eventually we zeroed in on this product. So if you think of Breadboard as a project, which is where stuff happens, where technology develops, and then Opal as a product, this is where we ended up. I think this is going to work. And effectively, this summer we launched Opal, which is a no code prompt composition tool. And everybody calls it different things. And primarily because we didn't say exactly what it was, we launched with little or no announcement. There's like one dinky blog post on Google Developers because ultimately what we want to do is we want to be out there with the users, hear what they're saying and shift and pivot. By the way, we're getting into strategy Aperture, and here, right away, one of the big things about products that become hits and grow very fast is that you can't change them anymore.

B (7:59)

Wow. Right.

A (8:01)

And so one of the big fears that I had, people who work with me will tell you I would actually like, oh, my God, we have growth. Oh, no. Oh, no. Because one thing that I'm trying to avoid is I'm trying to avoid getting locked into a very particular shape where there's a ton of users who are like, I want my thing, I would like to have an automation tool. And it's like, no, no, no, that's not what we want to do. But that's one of the interesting things that we wanted to do with Opal is that we wanted to launch quietly, attract a small number of people, and then see where you can go from there and play with and work with them, and every week do something new and that type of thing.

B (8:39)

I have a couple things to say, actually. One is the name Breadboard, I think, is just beautiful for people who are listeners, who you might think of a board with bread on it, where you cut the bread. But actually, if you've ever played around with electronics, and you have to be old enough that electronics were the kinds of things that were discrete components that you could assemble, there was this thing called a breadboard, and it was this incredibly flexible place where you could plug in resistors and capacitors and diodes and you could make electronics in a very rapid, prototypey way. And to give you a picture of what breadboard is and in particular what Opal is, Opal sits on top of breadboard. So we'll explain that later. The components that you're plugging in, instead of being resistors and capacitors, are essentially units of cognition. They are ways of assembling series of cognitive tasks in such a way that you can get them to do things that they couldn't do individually.

A (9:38)

By the way, this is beautiful. I'm going to totally use this because with a breadboard, you assemble an electronic device. With Opal, you assemble a tiny brain.

B (9:53)

On work for humans, we've been exploring the principles of multi sided management, which is the belief that work is a product that every company designs, builds and delivers to employees. Along the way, people started asking how they could put these ideas into practice. So I founded the work design firm Elevenfold to help your company create the kind of work that makes teams feel alive and engaged instead of dead and dull. So you can reduce turnover and build commitment. We're doing something revolutionary here. Learn more@elevenfold.com that's 11fold.com. Yes. And do you remember the moment when you said, I have an idea? And how did that come about?

A (10:43)

Actually came about very rapidly. I spent a bunch of time playing and building with AI myself because I was really excited about the potential. I could see the limits, but I was really excited. And the first thing I ran into is I want to run this prompt and then I want to run this prompt and I want to then combine them in a certain way, take results and do something with that. And I was like, wait, everybody's going to have this problem. And I think there was a moment where I was on a chat thread with my friends. By the way, Alex Komorowski, he was present in that chat thread where I was like, we need something like a breadboard for AI. And then it's like the light went off and suddenly all of the things fell into place. And it was very sudden. It was like a collapse. Almost like, you know, how function collapse. There's really interesting things. Everything unstable. And suddenly like, this is what I'm Building. And so it was somewhere in May of 2023. And then I went on vacation and over vacation, and my family will attest to that. I just built the first iteration.

B (11:50)

I'll tell you one reason why I love the name Breadboard is because breadboards to me imply fun. They imply hobby. This is the heath kit thing.

A (12:01)

This is.

B (12:02)

The thing is I want to build something I don't on instructions. I want a palette, and I'm going to play on that palette because it's made it easy for me to play. So there's a couple terms I want to pin down just so that we can use them going forward. When we say AI, we're mostly talking LLMs, I assume.

A (12:20)

Correct.

B (12:21)

The second thing is when we talk about collections of LLMs arranged on a breadboard or using opal, that's an application in the language that I've seen is people refer to that as an application. I don't know where agent lands in this as a word.

A (12:38)

I think we need to kick agent out of this conversation entirely because it's such a bad word. Everybody's like, agentic. Are you agentic? I'm very agentic. It's like, what does that mean? I don't know, but I would like five more pounds of agentic, please.

B (12:55)

Okay. We will throw this out. It's very funny. By the way, in part of this same series, I just spoke to a playwright and the person he's working with from Microsoft to be creative and write plays and reflect on his own plays. And one of them referred to their life as training data. What's happening is that we take metaphors from life and we apply them to AI and make all these assumptions. But now we're starting to take the metaphors from AI and apply them to our lives. I love it.

A (13:23)

I love it.

B (13:24)

So did you come across what I'm going to call a pattern zoo of cognitive components that people would want to assemble onto a breadboard? So we had Anthea Roberts on the show, and it was partially because of that conversation about metacognition that I said, oh, my God, why haven't we had Demetri on the show? And we talked a lot there about thinking about thinking, and you've been thinking about what are the components that go onto a breadboard. And I'm wondering if you found that there are repeating patterns.

A (13:58)

It's really interesting. At the core, it's all about composition. In fact, I think metacognition generally is a process of composition. If you just tear away all of the actual thinking at the end of it, it's an ability for a person, and some people are much better at this than others, to decompose or recompose or recombine various bits of thinking into patterns or into complete frameworks. And by the way, Anthea is freakishly awesome at this. Some people are just really, really good at it. And I don't know how that works. And so when I looked at Breadboard, and one of the big challenges we've had is to try to translate what does it mean in terms of LLMs? Because metacognition for humans and metacognition for LLMs are eerily similar, but different, because humans think and reason, and no matter how much we want to believe, LLMs don't think or reason. They perform some process that we can call thinking or reasoning, and because they're so eerily familiar and similar to each other. In fact, when you use any model that uses thinking today and you open up that twisty that shows thinking, you're like, oh, my God, it's doing something. It's clearly thinking. And then you start reading it. It's like, this is not thinking. This is weird garbage that is happening. I don't know what's going on here, because it's not really that process. So one of the interesting things we had to go deep in is what is the distinction? What is that unit of composition for the LLM that we can say all the patterns are based on that. And that unit of composition is. We ended up with something very simple and not profound at all. It's a prompt. The unit of composition is one call to an LLM, and then stuff grows from there. And so you can say, this is what Breadboard does, and this is where Opal wraps around. This idea is that you give one prompt to an LLM, get a result, then give another prompt to an LLM, and then by varying these prompts and building different prompts and composing them into a flow or a graph, you effectively simulate metacognition. And so Anthea, for example, built these really, really cool breadboards where it's like, think about the problem this way. Now think about this problem this way, and now simultaneously think about this problem. Are there three different ways? Then let's reconcile the differences. And so every one of them is a single instruction. So humans would go crazy if you tried to ask them to think that way. But for lms, it's completely normal. They're like, of course I'm going to first think about the problem this way. Then I'm going to do it this way. But the interesting thing, what Anthea has found is in each of these pivots, to think about it this way, to think about it that way, we break the LLMs predictive thinking process out of its normal way of thinking. Does that make sense? So it's like out of its repetitive way of just reasoning about the problem. So if you compare Breadboard runs with just chatting with an existing chat like interface, it's going to be different because it actually tries to like, let's go this way and let's go that way. And the lamb goes like, whoa, okay, I'm here for the ride. I'll take your instruction and do that. But the actual wisdom and the pattern that it exhibits is encoded by a human who understands metacognition, who understands, ah, this is a problem that requires me to think about it this way and this way and this way and this way. So what I'm going to do is I'm going to instruct an LLM to follow this road and not try to overload and say, do it all in one go. Because what's going to happen is you're going to get averaged out and instead you act with precision. And this is where the most interesting patterns emerge.

B (17:58)

I'm going to say there's different categories of things you might ask an application to do once assembled. In preparation for this, I used Opal to generate your biography and a background and a set of questions that I might ask, which I'm not going to ask because I didn't have time to make the prompt, really create the questions I would want to ask. But first of all, let me say two things. I tried to do that earlier on Breadboard, before Opal. I just didn't have the patience to do it. And Opal's interface made it something I could do in first time out of the door, 10 minutes and I had it. And the reason is that you interact with Opal using LLMs. So the interface to Opal is I'm just going to say what I want and it's great going to assemble it for me. But my purpose was really kind of basic. It's nothing I couldn't have done if I'd read everything you'd written, which I keep track of anyway, so I don't really have to read it. But all I was doing was digesting more content than I had time to digest and having it do that for me. So it was a transaction. I think of it as a very simple transaction and, and the way I did it and this is why I didn't love the outcome is that I asked it to do it in a very generic way. And so it sort of went down the middle of the road, the middle of the highway, along with everybody else. And that is not what I want. So that's one problem. But then the other thing is that Anthea is trying to create new knowledge, and she's trying to do that by taking a cubist approach to a problem space and applying different lenses to it and then coming out the other end with something new. And I just think it's more advanced. What other kinds of creative, surprising applications have you seen out there of how people are assembling? Because I'm sure 99% of everybody looks like what I just tried to do.

A (20:02)

There's, like, a really interesting shift in mindset that happens when a person gets into Breadboard or Opal and they're like, oh, I can take what I know and transpose it into a composed workflow. And now that thing incorporated what I know, and it can basically pretend to encapsulate this and act on that wisdom or with that wisdom. And I think we've seen the majority of people do exactly as you said you type in. And our little planner, which is very smart, but the little planner is going to, like you said, go the average road. The user said, I want this, I want that. Okay, this is what I'm gonna do for you. The interesting thing is when the user goes and says, oh, actually, this is not quite right. I do want to build a narrative here, or I want to build a story arc here, but I want to have characters that have this and that. And suddenly you can no longer use the average. You need to go, this person knows something. This user knows something about the process that either nobody else knows or is rare knowledge that's not average. And I think that's the interesting part, that shift. And I've seen a few people, a few users shift. And Anthea was the pioneer of this, where for her, it was like, that's my job. My job is not to ask Breadboard to build me something simple. My job is I know all these frameworks, and I am starting to actually get overwhelmed by them. And how can LLM help me incorporate the wisdom that I have into the process? So right now, we haven't seen that much of people actually going into this mindset shift. But I really want it, because I think this is where we all have unique, amazing knowledge that when we die, will die with us. And at some point, we want to hold on to this knowledge and keep it inside of us because we believe it gives us advantages that other people don't have. But there will be a moment where you go, you know what? I actually want to share this with the world because I think the world will be a better place if this wisdom is shared out there. And I think this is the very idealist me is why I built printport. Does that make sense?

B (22:21)

Yeah, absolutely. And it immediately started occurring to me that I need as a component in my Opal models a little avatar of me, and that that little avatar is probably wearing a particular hat. So it's not Dart, the person who likes the fish hat, it's Dart, the podcaster hat. And I will ask it as an input or a context to figuring out what to ask as interview questions. I would take all of my previous episodes. So that dart avatar that I should build, I could give that to my kids as an inheritance and they'd say, don't want it, dad, don't want that. We're just going to put it on this thumb drive. I mean, that's an interesting exploration. There's your avatar and then there's the mode that you're in.

A (23:16)

Well, can I take this discussion just a little bit further? In fact, I believe that one of the most profound changes that we can make, and I don't know if we'll get to do that, is to apply LLM to extract tacit knowledge in a way that makes it scalable and useful to more people. If you think of what happened with the Internet and search, search allowed indexing of the Internet and finding things that you wouldn't have been able to find before. What if the big thing that we get out of LLMs is the ability to actually scale tacit knowledge that we might not have, might not be able to share otherwise, and then scale it and do something with it, something good with it. Of course there's opportunities to do something bad that always comes with it. But I think that's the most interesting thing for me is can there be something like that? First of all, wouldn't that be cool? And second, would that not have profound impact on how we learn, how we educate ourselves, how we build organizations? And if tacit knowledge is transferable, what does that mean?

B (24:29)

Oh, right, that's right. There's so much in what you've said over the last few bits. One of them is that the value I bring to tasks with AI is my variation from the average.

A (24:44)

Yes, yes, yes, yes, yes, yes.

B (24:47)

Awesome. In other words, it's the unique me that I'm bringing that because the LLM is average.

A (24:57)

In fact, LLM is average. U if you kind of do subtraction, that's what it is.

B (25:02)

Yeah. So that's one thing. The second thing is the idea that we gather tacit knowledge over time. Some of it I can't tell you. I can't tell you the tacit knowledge, some of it lives in my muscles. It's going to be hard to get that out. Some of it if you're a guitar player, it lives in your fingers.

A (25:22)

Yeah, that's exactly right.

B (25:23)

But I'm not at all convinced. I've been thinking a lot lately about what it's going to be like to go through university now. And what I think is that I'm going to have a recorder that follows me everywhere I go and looks at everything I do. And that what I'm doing as I go through college is I am essentially gathering training data for the LLM I'm going to take into my post career for the model. And so there's a way in which I'm not even going to be independent from the AI at the end of college if that's the way I go through it. There's some things I need to know that I need to carry in my head, but there's a lot of other things that I want to carry them with me. And I've been thinking a lot lately that there's going to be this question if I'm going to form a contract with a company and I'm going to have experiences in that company, how are we going to manage ownership of the stuff that goes into my model, the one that I carry with me everywhere I go. I'm not going to sign a contract that says my LLM can't learn, my model can't learn while I'm working for that company. So I'm not sure that the distance between us as individuals and AIs is going to be quite as distinct.

A (26:37)

It's really interesting though, right, because you said something about what makes me unique. And I firmly believe that as we as a world carefully climb up the Maslow hierarchy of ladders, things that are in scarcity reflect the next step Today, what's the most scarce thing? It's not the ability to have information, it's the clarity. We're in a world where clarity is scarce and whoever has clarity or purports to have clarity is the most desired resource, has value and I think uniqueness and non averageness. I don't know how to frame it yet might be that next thing. And you're exactly right to look at education. What does education look like in a world where uniqueness is the actual value, where if you just want an average software engineer or an average writer, you actually don't have to pay anything, you can just get it? Does that make sense? There's something really phenomenal because if you have 8 billion people in the world, all of them are unique in one way or the other. But finding that uniqueness, identifying it is hard, and it's really interesting, right?

B (27:55)

Yeah. Yeah. Well, it's interesting because one of the things, and by the way, this is something that Opal told me that I would not have necessarily used as a phrase, is that you apply lensical thinking and that you talk about lensical thinking. And first of all, I'm just going to ask, how do you describe lensical thinking so that we have that out on the table?

A (28:16)

Lensical thinking is something that many people do automatically, and for some people, it's very hard. It's just a way of thinking where instead of seeing or firmly associating with a particular view of the world, you realize as a person, as an individual, that there is many other ways to look at the same problem. And these different ways are called lenses, and that you might be better off in many situation to shift those lenses and use different ones to better see this situation that you're facing. And so lensical thinking is simply a way or way of stating that, hey, when you're facing a challenge, you might be better off to gain clarity on this challenge, to understand what the problem is about, to have a bag of lenses or different ways of looking at the problem. There may be frameworks, there may be very simple idioms or rules of thumb, whatever those things are, that allow you to just switch them around and say, if I think about it this way, is this that or is this this? And that just gives you an immediate advantage, because the previous you, who was glued to a particular point of view, could not see things that the new you, who is now able to switch those lenses to see.

B (29:37)

And this is very much what Anthea is doing, which is that her work in international law led her to establish very clear understandings. It's interesting that we talked about clarity and now we're talking about lenses, which is that she has a set of lenses from her work in international law, that she can now create clarity along one dimension, because a lot of times, the way you get clarity is by flattening the dimensionality of the problem. So it's not the whole problem, it's just a useful flattening. And by combining those together, she's creating new value in the world because each one of those lenses contains so much expertise that probably only one human could hold it. That's right, in many cases. So I want to talk about some things that I think are related, which is your ideas around an elephant versus a dandelion strategy. And then let's talk about the ways in which one of those strategies is manifesting in your current work.

A (30:37)

Dandelions and elephants. I really had gone on the whole kick and I built slide decks and written posts about it because it was such a really good metaphor. It's actually borrowed from biology, where there is a lens of looking at species as either R selected or K selected. So if you think of the r selected species, these species are focused on rapid replication and growth through mutation. So evolution through mutation. So the first generation of locusts comes over diesel, the next generation is a little bit smarter because the locust that didn't die clearly had some genetic markers that were worth retaining. And so our selected species are numerous and they just making as many copies of self as possible and mutate through generation to generation. And this is how knowledge of our selected species, or wisdom how is acquired. How is a tiny little spider able to have this really sophisticated strategies of capturing the prey, our selection and the then k selector species were getting into biology. And by the way, I know nothing about biology, so it's like this is me reading a Wikipedia article. But K selector species are about carrying capacity. K is a capacitat, a German word for capacity. And basically these species take a completely different approach. They're like, hey, we're going to grow big. We're going to try to stay alive for as long as possible because we're going to accumulate the wisdom, we're going to produce one or two offsprings and then we're going to spend and invest time into teaching these offsprings how to be with the world. And so the passing of wisdom does not happen by mutation. It happens through careful process of transfer. There is an explicit knowledge transfer. So obviously humans rk selected species, we're not r selected. And elephants are another good example because they're big, they grow. That distinction was so much fun to play with because if you try to apply it to realm of ideas, you can clearly see that some ideas are selected. And so like memes or various ways of X or Twitter, whatever you call it, it's all R selected, right? Because the lifespan of a tweet is very small. But the next thing and all of the learning transitions, all of the idea mutation happens to the next one, to the next one, to the next one, to the next one, to the next one. But then there are some ideas that are very much case selected. Organizational strategy, organizational culture, things that are hard to build, manufacturing processes are all ideas. Also ideas that are very clearly carrying capacity based. You have to transfer them, you have to have trainings, you have to carry the ideas very carefully from one body to another and pass them along.

B (33:42)

So chip fabs, multibillion dollar investment, you expect them to last a while.

A (33:48)

Yes, yes.

B (33:50)

Your competitive advantage there is I've got $10 billion and I can build this and I can place that kind of bet. But then, for instance, something that's more K selected in business would be all the people spinning up little apps and throwing them out.

A (34:04)

R Selected. That's correct.

B (34:06)

Sorry, R selected or the Daniel approach.

A (34:09)

Correct.

B (34:10)

Now I want to bring it back now, which is AI is at this place where we don't know what's possible and it's incredibly cheap to start experimenting.

A (34:23)

Yes.

B (34:24)

And so I'm not sure if AI is just really good for dandelion strategies or if it's just where we are evolutionarily in where AI is that I'm confusing where AI is in its evolution as a technology versus its native ability to support dandelion strategies.

A (34:44)

Yes, yes, yes, yes, yes, yes. This is so good, by the way. This is really, really good. I love where you're going with this. This is really interesting because I don't know if you remember, but we used to have an explosion of cell phones of various shapes a while back. And now if your phone is not a slate with a glass screen, that is touch screen, that's not a very popular phone. So I think there is definitely a progression. Right. There's definitely a progression that every successful idea, as it becomes to gain value, as it gains value, becomes less dandelion and more elephant. It sort of descends through the pace layers. It moves very quickly. Id 8. Id 8. 8. And then it's like, whoa, whoa, whoa. This is good. Let's slow down just a tiny bit. Let's build around that. And then it's like, let's slow down a little more because now we have a million users, we probably should be very careful because now we have a value attached to the idea. The idea has grown and now, ooh, this dandelion is looking more like an elephant suddenly. And I think what we're experiencing with AI is this really Fun thing where the medium, the AI itself is incredibly cheap. And dandelion y it was not cheap to manufacture all those phones. It was not cheap. There was a lot of investment going on that ultimately didn't pan off. And we have many corpses of the companies around us that you can look at this and say, well that didn't pan out. But with AI, because the medium is so cheap, will we ever get to the point where there's elephants?

B (36:22)

Well, one of the things that's going to happen is that AI is cheap because large corporations are carrying the front loaded cost.

A (36:30)

That's right, that's right, that's right. This is one interesting thing about our selected species is I like to say that you will not have ants if you have no food in your house. You have to have sustenance for our selected species to exist. And so if there's ever a winter, the bumble bursts or whatever that thing happens, we will probably see elephantification of these things. And it might be really interesting because it's almost like we're doing the musical chairs and whoever sits down, that's going to be it. And I think that's going to be really interesting.

B (37:06)

My analogy for that because for instance, I was working for a large corporation when the dot com bubble burst and I called it sledgehammers in six feet of water. And what that meant was famous story about a gunfighter who challenged the blacksmith in town to a duel. And the thing is, if you're the challenger, the person who's challenged gets to pick location and weapon and he chose sledgehammers and six feet of water. I missed the part where the gunfighter was short. So that's the sort of thing is that elephants are better at getting through winter. Ah, this analogy just blew up in my face.

A (37:42)

I don't know.

B (37:43)

But that's the idea.

A (37:44)

You're on it, right? Yeah. We can take it weird places, but we're not going to.

B (37:49)

Yeah. So I want to talk about, about the experience of work.

A (37:53)

Well, can I pitch an idea to you?

B (37:55)

Yeah, pitch, pitch, pitch.

A (37:57)

It's very interesting because when you're operating in different environments, in dandelion environments or elephant environment, you will need different organizational structures. You will need an entirely different way of thinking about how to solve problems depending on which environment you're operating in. If you're in the dandelion idea world, you need to have very rapid experimentation because again, mutates learning through mutation. So you have to have a lot of copies and a lot of mutations and things. And so startup World is famous or infamous for this idea. It's like you just throw the seeds out and just try to see what happens. And it is very hard for large companies to be productive in this environment because if you have a lot of value accumulated around yourself as an organization, betting on new ideas is not just hard because all of your employees and all the culture tells you to be careful, educate and learn instead of just go into the wild and throw things at the world. The thing is that throwing things at the world actually may impact negatively the value you already have. And so in dandelion environments, put larger companies on the back foot and they're always in this place where it's like, what do we do? And it's very similar. How elephant might weather a locust infestation. Does that make sense? Just like endure, just do something and maybe be angry about it, but that's about all you can do. And that's very interesting because again, if you look at the elephant organization, if you're in elephant environment, if you're trying to build a new phone or a new chip, for example, you can't be a tiny startup because you need a supply chain. You need the ability to navigate the supply chain. You need to be able to have some command authority to make choices that would be long term good, but short term expensive. And this is not something that a dandelion can't do. And Dandelion doesn't have attention span to go through any one of those things because again, learning through mutation. And so this is why, for example, we don't see many startups that try to build new interesting phones or new interesting things that already have a well defined shape. It's because that shape is set.

B (40:27)

What's so interesting about it is you would think that making more money would make it possible for you to take more risk. But a part of the reason why you wanted to keep Breadboard small at first is because if it made enough money, it would get pinned down.

A (40:44)

It would get pinned down.

B (40:46)

There's a way in which profitability slows down. Mutation.

A (40:50)

Yes, it's a very interesting thing. I call it strategy aperture and this is an audio podcast so you won't be able to see my hands. But if you imagine in dandelion environment or a dandelion attuned company has a very wide strategy aperture, it has so many choices it can make that it can take and just make this happen. Oh, we were a chat app. We're not going to be a chat app anymore. As of tomorrow, we're now an AI assisted, blah, blah. What does it take to do this? Nothing. Because you're a dandelion, you can just basically simulate death and birth in a week inside of your company. And so your strategy aperture, the ability to make choices over time is very wide. Whereas in an elephant environment, while you're very precisely focused on a thing that is hard to build and know exactly how to build it, your ability to go outside of that focus is severely limited because all of your people know only how to build that thing because that's how you hired, right, you hire experts to build that thing. All of your supply chain, all of your ability to make big moves is severely limited to that thing. And so in some ways this notion of the strategy aperture is that depending on what kind of company, what kind of organization you are, what type of environment you're attuned to, your aperture will be wide or narrow. And this is not a mutable thing. It's like a fact of life. It's like a thing that you have to recognize. And kudos to Google for recognizing that it needs Google Labs, because Google Labs is exactly the dandelion style organization that sits on the side of Google, the big Google, which is clearly an elephant and allows you to experiment and build interesting new things.

B (42:49)

I'm going to ask this question different ways. There's actually two ideas I want to tie together. One of them is what makes a good dandelion farm, I guess for the people who are in it. And it's super exciting to be in a dandelion farm, although some people probably hate it. But let's talk about that first. If you were designing an AI lab from scratch, no blueprint on the frontier as a dandelion, what would you do?

A (43:18)

First of all, what a tempting offer. Second, I'm going to introduce a new lens. And the lens I'm going to borrow from Steve Jobs who said I better be a pirate than join the Navy. And it's the lens of pirates and sailors. And the interesting thing about the pirates is that they dream of the possibilities. And their biggest fears is that they're going to miss some opportunity or miss some interesting thing that otherwise just would pass them by. It would be like one of these big chance they will not be able to take advantage of. Whereas sailors is a mindset where you primarily worry about things getting out of control. And primarily your biggest angst is that something is not aligned or not exactly right or not built correctly. And I think if I was to build a new organization that is a dandelion farm, I would look for experienced pirates, these are the pirates who have been always a pirate, who have learned to be in the Navy, have learned all the rules and then were like, ah, I now know which rules I can break. You know how there's like somebody saying that there's three stages of a man development. Man believes in Santa Claus, man does not believe in Santa Claus, and the third stage is man is Santa Claus. It's the same thing. It's like, first, you don't know what the rules are. Second, you know what the rules are. Third, you know which rules to break.

B (44:50)

There was a two by two that was created around your discussion of pirates and sailors. And the one axis was pirate or sailor and the sailor was in the Navy. The other axis was think and do. And so we had the idea of do pirates and do pirates are. They're tough to have in a large corporation because they have a tendency to build stuff that is counter architectural. Like you're never going to be able to incorporate it with the architecture of the whole product because they just built it. And now it's popular and it's built on spaghetti or something.

A (45:25)

What do we do with this now? Is the question that people ask.

B (45:27)

And the DEW pirate is like, don't know. I'm off, I'm going to go do it again. And they're like, we hate that guy. But dang, the thing that's on top of the spaghetti everybody loves. And so, you know, if you're a dandelion farm, you wanted a lot of DEW pirates.

A (45:41)

Yes, I think everybody is a pirate at heart a little bit. You have to have this very particular yearning. I think it's important to go through the Navy first. I've seen many Googlers who fail in startups because they leave Google thinking they're going to stay in the Navy. And I think the Googlers that do succeed are the actual pirates who've been in the Navy who did their tour. They're like, great, now I know which rules I can break. Now I know what I don't have to do because all of those things that I learned, some of them are optional in my particular environment. And that flexibility, that mental shift, that ability to be flexible with which rules to break, it's very, very important. And let me give you an example. When you're at Google and when you're landing code, you have to have this thing called code review, which means that another person has to review your code, which is a really good idea in an elephant organization. Is it a good idea in a dandelion Organization? No, because what you want is you want people to just go crazy and write code like nuts and just explode. And yes, they will create technical debt and things, but technical debt is zero value or zero cost if this thing does not survive. So there's something very interesting about this. But at the same time, if you just say a junior engineer who doesn't know how to write code correctly yet, who hasn't gone through the effectively the crucible of being an experienced software engineer and say don't do code reviews, is that a good idea? No, because they will just make mistakes where they work themselves into a corner.

B (47:23)

Some mutations are lethal.

A (47:25)

Exactly. And so it's like a very weird thing where if you are to start a new organization that is meant to be a dandelion farm, you have to find this fairly rare breed of people who have the mindset of being a pirate but have the wisdom of having been in the Navy. I think that's super, super important. I don't think this is like hard and fast rule, but this one is super important.

B (47:53)

Now I'm building a boat for your pirates. What does that boat have to do? I mean, I just have a couple of ideas and then we can go from there, which is it can't make them walk the plank for not doing a code review. That's an example. What else is the environment going to contain to support your pirates?

A (48:12)

There's something very interesting about balance of agency and belonging. Again, another lens. If you view agency in belonging, at the very extreme of agency, there's this maverick who always works by themselves and does their thing because they want ultimate agency. And then the other extreme is the person who only does things when they feel like everybody else agrees and everybody else says, do this, this is a good idea. One of the biggest reasons why an organization or a team of pirates, a band of pirates, will fail is if this balancing of this belonging and agency is skewed in one way or the other. There's a very tiny sweet spot of finding non mavericks who still desire a lot of agency but are willing to go into their own places, but at the same time willing to listen to other people and listen to their suggestions. So I think a culture inside of this organization requires mental flexibility. And in fact, dare I say, lensical thinking is one of the key elements of the organization. And in fact, one of the greatest things that I like, the people I love working with are people who have this holding their ideas lightly but not being shy about expressing them. And I think this is one of those things, especially when I was younger in terms of working in organizations. It would be very uncomfortable when you would meet with somebody and they're like, no, this is a terrible idea. And you're like, oh, my God, my idea is terrible. And it took me a while to go and say, no, it's not terrible. Here is why. And the person goes like, whoa, I didn't think of this. This is a good idea, meeting these people and working with these people. This is, by the way, one of the reasons I work with labs is because we have this culture, we have this ability. Our leads have very strong product and engineering nose, but when confronted with evidence, they will go and say, oh, that was a good idea. We should do that. The ability to have that is incredibly important. It is almost not important a Navy at all. Because navies are better off being structured as hierarchical organizations. In fact, one of the big problems that I think Google is facing internally in culture is that Google thinks it's a pirate organization, where it's been a navy organization for years now. For decades. Not decades. For a decade, at least. I joined Google in 2008 and I remember the pirate times.

B (50:50)

You and I actually both contributed to an article in the Flux Review that was called the Unvarying Infrastructure of variation.

A (50:58)

Yes.

B (50:59)

And it was all about how biological organisms are incredibly obsessed with faithful replication. And a part of what they faithfully replicate is the ability to create tons of variation where they touch the immune system and the generation of antibodies is one place where there's just 10 to the ninth variations in each human. And then sexual reproduction, which is. If you're just all about faithful replication of an organism, you wouldn't do that.

A (51:29)

You wouldn't do that.

B (51:30)

Mix everything together. So I think a company can be both. A company can have both the same way that organisms do. It's just important to know when you are which of those things, you don't want your payroll department to be expressing a lot of variation.

A (51:47)

Yes.

B (51:48)

You don't want it.

A (51:49)

This is a really good insight, by the way, in terms of. If you define the variation as the requirement, you can build out almost a neat little framework. Right. Because then the lions and elephants, variations are meant to do different things. For dandelions, variation is, oh, a new idea. Different one. Completely different. So very drastic pivots are fine. You know how many times we pivoted with breadboard? It's crazy. But like with the elephants, variation is still very important, but you have to constrain it in a way that only hits places that don't create impact in Negative ways. Does that make sense?

B (52:28)

Yeah, it absolutely does. And it's funny because I've come to this realization recently, and this has to do with the design and delivery of the experience of work in companies, is that there are some parts of it that you want to not have a lot of variation. Everybody wants it the same. They want their payroll to be accurate. At the center of what everybody wants the same. You can scale it and you want to use things like Six Sigma and process methodologies to reduce variation. But at the edge is where every single individual in the company wants something different in terms of their personal experience of the work that they're doing today. And so over there, you need to be a lot more dandelion, like very low cost but high information in terms of understanding what the variation of customers wants. And so it's very much the same. And it's a different methodology because out there on the edge it's a design methodology, but in the center it's a process methodology.

A (53:28)

Yep, it's a paste layer just all over again.

B (53:31)

Yeah, no, actually, I hadn't thought about that.

A (53:34)

So another lens, by the way, paste layering, if you'd like to look it up. It's a really cool little lens, but it does give you this ability to vary your dandelion as to elephants race. And this is how you structure organizational dynamics. Right. You basically try to establish where do we want to preserve the value, where do we want to be the most conservative about change.

B (53:58)

I think one of the differences between what I said and pace layers is that pace layers are over time where you want the variation in the faster pace layers happens with a high tempo of change. But the variation I was talking about is a variation in customer demand.

A (54:14)

Ah, I see.

B (54:15)

It's still. Right. Which is that the pace to respond to changing customer demand from person to person or something is just not quite as time based.

A (54:25)

Yeah, it's very interesting. Very interesting.

B (54:27)

You've spoken in the past about embodied strategy. I think, about that concept as you introduced it to me all the time as it relates to organizational change. What do you mean by it? And then I have some questions.

A (54:40)

Embodied strategy at a very, very high level is that when you're an organization, a strategy picks you in some ways. The different bits of culture that you have embedded in the organization, the different ways in which you work, habits of the organization, all the tacit knowledge in some ways defines what the strategy will be. And it's not necessarily that you might not want that, but if you want something different from who you are as an organization, you might have difficulty implementing these differences because the embodied strategy, the things that you are meant to do as an organization, are very hard to change because a lot of them are tacit, a lot of them are underground, and they're not something that is visible easily. And so this is why organizational change is so hard, because sometimes you realize, how hard could it be? We're just going to do this thing now. Here's our new way. Here's a strategy. We're going to think about this problem differently. We're going to approach it from this particular perspective. And then you try to implement it and nothing happens. People go like, yay, we're doing this. And then at the end, three months later, it's exactly the same thing as before. And embodied strategy just is a word to describe this fact that organization structure dictates outcomes. And the structure of the organization, it's embodied. It's not something that is easy to change. You can't change the structure.

B (56:12)

It's like the first time I got on a snowboard, I thought, looks like a skateboard. And so I rode it, I think, exactly three feet and broke my collarbone. And the reason is my embodied skater brain stem had it wrong, had the physics wrong, didn't understand it. It was way off. But the thing I'm curious about, and it's a mystery to me, is what is embodied strategy in a company made of? And the reason I ask is that if I want to swing the hammer to change the shape, what am I hitting with the hammer? Where does it live?

A (56:53)

It's really interesting because I think it's actually a misnomer to say that this is a strategy, because in some ways it's such almost like a biological bone structure type thing that a good way to describe this body strategy is that I say, dart, we're going to fly today. And you're like, no. I mean, I want to, but I don't have anything that allows me to fly by myself, at least without some aid. And I think this is exactly what it is. And I think rebounding strategy hides in small things that seem not important. Like one of them, like, let's go back to code reviews. I'm not going to say this is bad, but the fact that you have to do code reviews when you submit code dictates the cadence with which you can submit code. It also dictates the quality of the code that you can submit. So it is definitely some component of embodied strategy. Is it a bad idea? No. Is it an idea that confines you to a Particular way of working? Yes. And this thing, like, for example, let's have an organization where everybody works in the same office. Is this a bad idea? No. Does it confine you to a particular way of working? Yes. And so anytime you ask a question, does it confine you to a particular way of working? And the answer is yes. Finding what that confinement is, finding what that constraint is going to be, will reflect on what the fragment or facet of the embodied strategy it represents. And the reason why I like the word embodied, though, is because just like embodied cognition or something like this, or more biological concepts, you can't point at it.

B (58:43)

Right, right. You can't point at it.

A (58:45)

You can't point at it and say, this is what it is. And one of the biggest difficulties and challenges I had, so I spent about two years, a year and a half maybe, working as a strategist, actually. So I took off my engineering hat and worked in strategy. And one of the hardest parts was when trying to come up with a strategically important direction or something interesting on the back of my head, I was like, yeah, yeah, this is a good idea. But is this something that's feasible given this organization's embodied strategy?

B (59:18)

Yeah, yeah, I have a good idea. I'm going to ride a snowboard. And you're right. I like the idea of embodied because it gives you the idea that there are things about your body that are not available to your consciousness too, which is your muscle memory is an example, or I would argue your aesthetic is something that you're not entirely in control of, which is, I think order is beautiful, or I think order is horrifying.

A (59:49)

Creepy, right?

B (59:50)

Yes, creepy. Yes, creepy.

A (59:53)

This is one of the interesting things in personal life. I've met people who are like, order is creepy. And I was like, what? And that's how my head got cracked and opened into the world of pirates.

B (60:09)

I ask a few questions at the end of every show. This is going to be interesting, which is, I ask, what job do you, Dimitri, hire your job to do for you?

A (60:18)

Oh, that's a good question. I think it's purpose. I'd like to wake up every morning and have a sense that there is something that I'm doing that directly contributes to. To me feeling good about what I'm doing and indirectly contributes to some very large goal that I call the arc of humanity. And if I can triangulate those, and I usually do, I actually have a. Every morning I wake up, I have a tiny routine that I call set intention for the day. Where I look, I have A list for myself of these things, and go and say, what am I going to do today that will advance me down those two things? And if my job allows me to do this, then it's awesome. If it does not, then it feels like I need to find another job.

B (61:07)

What is your favorite minute of work? And I'll tell you, it's very interesting. I talked to somebody yesterday. I asked them what they hire their job, like, what's their favorite thing? And they said, staff meetings. And I was like, are you crazy? This is what a researcher should never do. A researcher should never say, no, actually, that's not what I said. I said, that's sick. That's what I said. So your favorite moment or minute of work, what are you doing?

A (61:38)

It changes quite a bit today. I can tell you very concretely that my favorite moment of work is when I sit down to work on a coding problem, that I don't know how it's going to work yet. I have this really weird anticipation that it's like, this is going to be so hard. This is going to be weird. I don't know how to go. And it's like something with dopamine goes. And I was like, oh, my God, you have just supercharged yourself, Dimitri. That's my favorite mini to work.

B (62:10)

That is. I hire my job to solve puzzles, and it's great if those puzzles align to the arc of the world.

A (62:19)

Yeah, yeah.

B (62:19)

But it's also pretty fun, even if they're just sitting there in space. What does your job cost you?

A (62:25)

That's a really good question. I think I cannot be 50% or 70% on the job. I have to be 100%. You can tell that I stopped writing. I don't know if you noticed, I don't write anymore. That's because Breadboard ate everything. I don't show up in flux meetings and things like this. It's because this is the cost. It's a lot of fun, but it's also like, it just fills all the space. And I feel like this expands to professional, it expands to personal, it expands to everything.

B (62:58)

It's interesting because somebody just called me out on LinkedIn. I was arguing for creating work like that for people, the kind of work that they just dig. Right? And they said, is there a dark pattern there where you're asking companies to create crack for their organizations, for the people who work there? And I haven't really solved it yet. Because if you had the choice of working on Breadboard or not and having a more diverse use of your time which would you choose?

A (63:33)

I think it depends on where you are in life. When I was writing a lot, I went through the phase where I just needed to slow down and just look around. And it was very important. It was one of those things where I remember going like, I'm going to go home and I'm going to get excited. I'm going to write a post because this is the only thing I can think of. But now it's like it's in the space where there's such a close, tight feedback loop between what I do and what happens that it's just. Oh, it's just addictive. Is there a dark pattern to it? Yes. Do I feel like I might be overdoing it? Yes. Am I still gonna do it? Yes.

B (64:16)

Yeah, that's right. You know what it was? It was in the William Gibson novel, I think it was Neuromancer, where the main character, the first thing that happens is somebody takes all of his drug receptors so he can't be a drug addict anymore. And at the end of it, he makes a ton of money and he goes to the place that did that and he has all his drug receptors put back in so that he can be an addict again. And, you know, I don't think I can feel bad about creating that experience. This is an odd question which I ask a lot of people. What's your favorite tool of the trade? Of your trade?

A (64:50)

A whiteboard.

B (64:51)

A whiteboard.

A (64:53)

It's so bizarre because I sometimes just need to pick up a marker and go to the whiteboard and my ideas start get clarifying. It's like so odd, a good whiteboard clean that I can just go at, oh, my God, it's just so good. Maybe breadboard really was born out of whiteboard. I don't know because it's all boxes and diagrams. It's all the same thing. It's just I'm a visual thinker and it really helps. And I really enjoy drawing those boxes and arrows and trying to understand relationship of things and lay it out spatially. This is how I think if I.

B (65:31)

Could do one thing to design your work environment or your job, whatever you want to call it, to better achieve the job you want to get done, what would it be?

A (65:43)

That's a hard one.

B (65:44)

More whiteboards.

A (65:46)

More whiteboards. The interesting thing here is that I find that I tend to adapt to environments until they don't work for me. I usually place very low requirements on what I want from my job environment until it doesn't fit. And so clearly I'm blind to something here. It's one of those things where you ask me, what else can I do to help you? I was like, I don't know, you.

B (66:10)

May have the right situation. And it's one of the things that I see is that when people can work on designing their own environment, oftentimes they've got it what they're looking for. But if your environment starts saying show us profitability numbers for next quarter and you're a dandelion where you need some room to spread, that would be hard.

A (66:31)

Things like that, that would be hard. Yeah, I can definitely point you at the fences where like don't go here.

B (66:38)

Yeah, yeah, yeah. Exactly. Okay. Where can you people learn more about Breadboard and Opal? And before you answer that question, I gotta tell you folks, Opal, it's really amazing. I am not technical and in 10 minutes I was getting to the place where I could start to iterate and make it better. So how can people learn more about those and how can people learn more about you?

A (67:00)

Well, it's Breadboard and is an open source project and it's on GitHub. Probably the best way to go to opal is just opal.with google.com it's unfortunately us only at the moment we're trying to. Remember I was talking about limiting the audience. I'm trying to make sure that it's a fairly small cohort. But Also my blog, Glasgow.com, i'm going to try to write more, but I do have links to both.

B (67:28)

That's a very deep resource and I just want to spell it for people. It's G L A Z, K O V And you don't have to be posting there on a routine basis for people to be able to spend a months there learning stuff. So that's great. Thank you for coming on the show today.

A (67:44)

Oh my God, this was so much fun. Thank you.

B (67:48)

Thanks for joining me for another episode of Work for Humans. If you enjoyed this episode, please give us a five star rating. Wherever you listen to podcasts and share the show with one person you think would get value from it, Believe it or not, this really helps us grow the show and reach more people who want to build a kind of work, work that people really want. As always, thank you to my producer Jason Ames at 9th Path Audio for his insights into content and his high standard for quality. Final note, the opinions shared here are my own and not the views of Google or Cisco Systems. Thanks again for listening. See you next time.