The Odyssey of Innovation: Classics, Code, and Cognitive Load | Two Sigma's Matt Greenwood

25:51Are we getting towards this omega? Maybe should we change the omega? Because we've learned things during the epsilon that make us rethink how the future is going to look. And so those are the things that we kind of constantly talk about. That's the language that we've built here. I think that's one other thing that I encourage people to take on. And that is to be a little forgiving. We have phenomenal people at Two Sigma. I am privileged to hire people who are brighter than me every day of the week. And that genuinely is one of the joys that I get here is hiring people and thinking, wow, I don't know if I would have made it if I came now.

26:29These people are just stunning in their approach to everything. But one of the things that happens is you have a company that's around 20 years. You have millions, maybe tens of millions of lines of code, hundreds of pieces of platform with some maybe questionable architectural decision now. And what I constantly remind people is we should forgive our past selves. Let's assume that when we made those decisions 15 years ago, we made the best decision that we had at the time. Now, with the benefit of hindsight, they look kind of maybe dumb. But back then, when you only had two computers or you didn't have enough storage, they might have been the best that we could do.

27:11And so we have to look at the past, look at what we've built and take that, give it a little bit of forgiveness and understand that in the future, our future selves might be looking at us in the same way. So we want to pay it forward with a little bit of forgiveness as well. So it's really important to kind of bring that humility to the process because we're really trying to get to a better future for the company. rather than prove a point to each other. I like this idea of bringing people in to move along this platform journey or like the S-curve as we continue to talk about. And your past mistakes, they might look silly or the decisions that you made, they might not make perfect sense now.

27:49But I think that's a great lesson for everybody to be more forgiving for your past self. You're operating in an environment where you have the best knowledge available to you at the time and you're making the best decision that you can at the time. And like you mentioned, when you innovate for a long arc of time, you accumulate some baggage or you accumulate architectural decisions that you have to go back and revisit. But that's the benefit, I think, of that platforming approach is you get that vantage. You now have all the viewpoint you need to go back to those older decisions and refine them.

28:18And along the way, there's a lot of things that can slow you down in innovation. One of them is process. And we talked a bit about process initially. And you really called out the danger of being beholden to process and how it slows you on this journey what does good process look like to you yeah so so uh you know at two sigma um you know we deal in forecasting right that's fundamentally what we do here we try and forecast the markets and so we we we work a lot with our data as what we call time series right so you know the date changes through time and and what we've learned is that developing software for time series for kind of time series of wet software is absolutely critical.

28:59And one of the things here that you have to be aware of is that there is another time series going on, and that is the time series of the company. And so all of these things that we're talking about, processes and things like that, they exist both along the arc of time, but also at every point in time. And so one of the things that's important to understand is that you need to continue to evolve and grow and change these processes as fits the company at the time that it's in. And one of the most amazing things about building a company like this with thousands of people in it is that you can actually have multiple kinds of company in the same company at once.

29:41You can have one portion of the company that is kind of building a good, steady platform and looks like a thousand-person company. And then you might have teams here and there that are trying to innovate very quickly and look more like startups. And you have to understand that even a single point in time process doesn't work across the whole company. So you have these multiple dimensions. And so the key, you know, what I try and land with folks is let's try first to kind of think deeply and understand why we have a process in the first place. And we have a process in the first place fundamentally to what I call lower the cognitive load of the organization.

30:20You know, sometimes people think that's a personal thing. And they come to me and they go, Greenwood, you know, this isn't easier for me. And I'm like, no, no, no, no. It wasn't for you. This is for the company. It might be more complicated for you, but it's less complicated for the company or less cognitive overhead for the company. And so what we need to do is we need to kind of investigate just the process that I'm using here that I would immediately apply to this. So I have a process for, let's say, code release across the main platform. Do I want to use that same process for this kind of fast team?

30:55Now, ordinarily, you might think, yeah, sure, I have a process. Let's just apply the process. But actually, these two parts of the organization exist at different points in their life cycle, right? I don't want to take a mature code release process and apply that to four people trying to get great ideas quickly off the ground. And so it's really incumbent upon everyone in the organization to say, okay, what processes do we have? What are good to apply? And what should we change for the circumstances and the time and the part of the organization? And I think that that really is key. And if you don't do that, then you really can become beholden to your process.

31:34It's very, sometimes very easy to give up in the face of a process. I'd love to help you, but the process says fill out these 105. No, we, you know, we, we need to kind of constantly, and everyone in the organization, I think, has a responsibility to that process, right? I told folks I was at lunch with the other day, I'm relying on the fact that, you know, I'm not coding every day. I wish I was. Maybe the company doesn't wish that, but I wish I was coding every day. But it's important that I don't see what goes on on the front lines, right? I'm not there. I'm not at the leaves. I'm not coding.

32:09If you see something wrong, you have to say it because I won't see it unless you put your hand up and you say, you know, this emperor may not be wearing any clothes. It's really important to encourage that kind of communication up and down the tree. You know, going back to something you said earlier in our talk just now about even in the early days of big data and data culture and kind of building that within orgs, you were, I use the word cajoling, you know, getting on the phone and talking with folks about their data practice, the stuff they had on hard drives. And this is before the cloud where it was just easy to just like, they give you a URL.

32:42You had to go and get some hard drives and then you had to actually like yank the stuff out. So this is really a cool thing to dive into, I think, because you're talking about innovating on a new idea before there was tooling, before there was common language, before there was a real understanding. That sounds very familiar to the environment that we're in right now with things like AI and how people are trying to build business units within their orgs to go really fast. And you see these large traditional companies kind of just like giving free reign to smaller teams to just build, build, build.

33:13Don't be slowed down by our process. We need you to find the new process. And I know there's lessons packed in there from your journey in early data. And I'm wondering, how did you build kind of that understanding of a sustainable data practice before big data was even a thing? How did you socialize it? Yeah, I think that really the journey at Two Sigma in many ways begins and ends with the people that we look for. Look, you know, we're not a big company, even though we have more than 1 ,000 people. That's not big by comparison with other companies. And so we do have a luxury. We are privileged to really be able to hire phenomenal people.

33:57And to Sigma, almost everything begins and ends with the people that we have here, right? I just believe fundamentally that if you hire great people and give them interesting ideas and then stand back a little bit, they'll do great things. But you have to also kind of explain to them a little bit about not what great looks like. I think everyone understands what great looks like. But you have to explain a little bit about how you continue to make sure that you're doing great, right? So one of the problems that you have in any company that's successful is you tend to believe in yourself a little bit too much sometimes.

34:35And so it's really critical. And we are a systematic company. And so we have tied ourselves to this scientific process. But really, it's one of query and curiosity and asking those questions and then being really intentional about it. I have this phrase that I use at Two Sigma. I call it turtles all the way down. right the idea being that you can come and give me a number but i'll really only buy that number from you if i if i have a belief that all the way down to the bottom you understand where that number's coming from and what will happen is if if there aren't turtles all the way down you know if you're if you're drifting on nothing everything falls apart and so i really ask my people you know there are many different management techniques one of them was five whys and so and two so what's So the idea being that you ask, why, why, why, why?

35:30And sometimes I sound like a little child, but the idea is how you show me that you've really thought this problem through and you're intentional about it. And then pedal to the metal, let's go and get it done. But if you're just doing this because it's fun or big companies sometimes will just say, just let them loose and see what happens. the chances of you getting something good are like maybe monkeys typing shakespeare right so you have to kind of really think hard about what is that process that you use to encourage innovation and it being intentional and really being able to explain how you get from a to z it really what it boils down to is being aligned on impact sounds like ultimately you need to understand what you're doing turtles all the way down is a great way to describe it um i you know the why why why really making sure that you can fully understand their train of thought and why they are justifying their decisions and then like you said then once you will have that alignment again on the impact then you can go uh really fast and i'm wondering too about how you're applying this nowadays how does two sigma think about keeping up with the latest and greatest in this impact driven way like specifically around like ai and llm technology yeah so so again happily we've been doing this for a long time yeah we've been doing machine learning we've been doing ai for literally decades.

36:52We have built a whole practice around how to understand data, how to understand data specifically as it changes over time. And so all of that really has helped by providing a great foundation when we come to this new and incredibly exciting kind of gen AI. You know, maybe flowering is the wrong word. I kind of, it's more like mushrooms after summer rain than flowers. It's a little bit, you know, havoc ridden. And so, you know, that has definitely, you know, given us, let's say, comfort. We think we understand the questions that we should be asking. But really, you know, deploying AI effectively is first and foremost, understanding how you think you can use it.

37:39And that is non-trivial, right? You know, we say, oh, we have to use it, right? That's true. But understanding how we have to use it is really key. And it's not about how we have to use it now. It's how, you know, where do we want to be in three to five years, right? There's no question about it. Gen AI is going to, you know, be like a sea change across engineering. And so you want to think about how do you benefit from that inflection? You know, what do you do with your people? How do you figure out? And I have, again, like I keep saying this, I have great people. I have great people. I have people who graduated from the most amazing places in the world.

38:23And they say to me, I'm nervous about the future. And my role as leadership and management and innovation is to try and explain to them what a future looks like, where their role and their value will be even more than it is today. And, you know, I understand that it can be concerning, but the future is incredibly bright if you figure out where you're going. And so first and foremost, on top of that foundation that we've built is thinking about how do we use Gen.AI to automate? And we've been doing automation again for 20 years. And I love automation, right? Because if I can automate something away, that means I can spend more of my time that I used to do that thing that may have been fun, but it was repetitive to actually think about things more interesting.

39:13And I'm still a humanist, right? I still just believe that we have this little spark that no machine is going to come and take away for any time soon. And so I just fundamentally believe that freeing up time for humans to be creative and innovative has got to be better for all of us than worrying about whether they're going to type 100 lines of code tomorrow or some gen AI will do it for them. So it sounds like a lot of the ways y 'all are harnessing AI now is around things like automation to free up cognitive load, human process to work on a deeper process. There are many ways that we use AI today and we kind of tend to clump them together.

39:53And so the first one of the first steps about trying to understand how we're going to use this gen AI in the future is to try and understand all the ways that we're even using it today. Right. So, for example, I'm sure like most of your listeners, I have some subscription to to some kind of Gen. I something or other. And I use that in what I would call an advisory role. Right. I ask questions. It gives me answers. And that's it. That's an advisory role. And that's a great role for Gen. I. And it lies, you know, half the time people like to say hallucinates. I don't pull punches. It lies. It just plain out, plain out.

40:28long. Then there are, you know, what I would call a more Oracle role. Well, I guess I'm getting classical again, right? Which is to say something that will accept or reject the outcome of an advisory role, right? So, you know, hey, can you write this piece of code for me? Now I've got a piece of code. I want to know is this true? Like, does this do what I want? So I can pass that off to an Oracle and say, here's the code. Here's the thing I wanted it to do. Can you just check that it does the thing it wants and just give me an answer and i'll accept the answer that's kind of more of an you know an oracle type role yeah then we have another role that we use ai for around which i would call an operational role can you create a code base check it out again make a branch these kind of things where it would be a bunch of you know pretty standard uh um uh commands that I would do that together make a small piece of workflow, tiny piece of workflow.

41:28And so when I was programming once upon a time, we used to have people who looked after the build, right? They call them build masters. And so that's a classic example of where you might use AI for an operational role. And then there is the kind of maybe the top or the bottom, depending on which way we're drawing this ladder, the agentic role. And that is someone who kind of coordinates all of these. So, so, so you might have, and then, you know, we've seen that, you know, there was a, uh, an article I read the other day, I think it was my, I, some large company, I won't name who tried to create a company of agents.

42:02Right. So they had business analysts and so read this. Yeah. That was very interesting. You could think about, you know, an agentic role where you would have a bunch of advisory roles. So these would be writing code and a bunch of oracles that were saying that the road through the code would be correct and maybe operational role. who's checking things in and out of gear. And then an agent role, whose role it is to kind of give tasks out to these advisors, right? Maybe features, right? And now what does the human do in this case? Well, in this case, the human is left to do what I call the most interesting thing, which is actually describe what they want.

42:39At the end of the day, very few people I know who write software, write software to write software. they write software to express an idea to create something and we have better ways as human beings than writing in you know c or python or even pearl you know when you talk with like the smart people who work at two sigma and they come to you and express like oh you know i'm i'm nervous about the future or like things are changing so quickly is this part of like the mentality that uh within two sigma how people are preparing for the future is that like we need to understand how to interact with this technology and that might mean having very clear roles like how are you helping them be future uh ready in that regard it's a great question i i think that um you know this notion of roles is like my formal training is a mathematician so i like crisp definitions that leave no room for doubt and what we found over the years is that some of the ambiguities that we have amongst human beings is because people don't understand their role.

43:45And there's a classic tool, a racy diagram where you write who's responsible, who's accountable, who's consulted, who's informed. Most problems that go wrong are because people are in a conversation and they don't understand which role they're taking in the conversation. And so they believe they're being consulted or informed when in fact they're responsible. That's a bad one, right? Or they want to be more responsible, but actually you're just informing them. Then they get upset. And so, you know, if we see this in humans, there's no reason to believe that we wouldn't see this in computers as well.

44:15That is to say, if we can help define what roles are and kind of keep them tight, then we might be able to get better juice. Now, that doesn't mean to say that we've covered that there are turtles all the way down, right? We may be missing a few. That's where we live in AI right now is terribly exciting. There's always new things going on. And so there may be more learnings to have, and we might look back and say, wow, those were dumb decisions we made. But at each point, you want to try and be as intentional and as crisp as you can. And when it comes to people, I don't know. I think it's complicated.

44:54On the one hand, I think it's incumbent upon every leader and every manager to understand that today is different than even just a few years ago. Right. There is, there's a lot of concern and that's rightfully put. Even if we believe because we have the pattern match to 3000 years of history, the things will get better. Nevertheless, you know, it's important on the one hand, on the other hand, my job here, your job, our job as leaders and managers is not to kind of, you know, wrap cotton wool around every developer. We have to give them the tools that they can then work through those concerns themselves.

45:33My job as a manager is to be able to give you enough information that you can work through that concern that you have. And if I don't do that, then you're going to have more concerns. That's reasonable. And if I say don't be concerned, then I'm actually depriving you of the ability to work through that. And so it's really important, and it's a little bit of a balancing job here. It's really important that we find the right path to be able to give people the tools they need to work through the problems rather than keep them from thinking about those problems. And has that problem itself ever changed through all of these hype cycles?

46:08Is that always the core problem to solve? Or is there maybe different nuances that you've noticed having been in a company across so many hype cycles in tech? I think this does because I, you know, I, I think unlike other hype cycles, it's not obvious to people on the ground that the outcome of this is rapid growth in, in, in, in their opportunities. In every other hype cycle, if you kind of, I'm just kind of going through them in my head right now. Um, you could see massive opportunities ahead. Yeah. Internet. Wow. Like websites. Unbelievable. Whoa. Like huge opportunities. Here, it's not at all obvious that there are huge opportunities.

46:53And so that is fundamentally different about this cycle. I'll give you an example. A couple of years ago, my daughter was graduating high school, and we were talking about what she wants to do. And like me, she's kind of split brain, history and math, something like that. and i said you should just do computer science because there is no like this is your time this is your time and that was literally six months five or six months before chat gpt and after that so i like it i think it came in november right something like that yeah the cycle hadn't finished yet she's still applying for colleges and i'm like you know what philosophy is a good major because actually asking questions might be the right thing that you need now and so people still don't understand, but there is this concern that opportunities might be narrowed, and that is really different than previous hype cycles.

47:50Yeah, that's definitely a difficult one to navigate. And even within your own organization, over those hype cycles, you know, you've grown this engineering organization. You've moved through S-curve to S-curve, platform to platform, and you've found those new heights, those new advantages that give you new opportunities and new challenges. And along the way, you know, you've had to kind of take the temperature on your own engineering org, what they're doing, how they're adapting to your process, the one you've been explaining to us so clearly in our conversation. And it's clear that you lead with a lot of empathy with how you approach problems with your people.

48:23And I'm wondering, how do you, what are the underrated signals, you think, of like a healthy, adaptable engineering organization? That is a phenomenal question. perhaps if someone had asked me that 15 years ago we would be in a different place um you know i think this kind of goes back to my comments right at the beginning about um have being a full-rounded organization yeah uh you know one of the one of the key lessons that i i talked to my initial manager you know my first-time managers about is understanding that everyone brings their whole self to work and and you know there are many many dimensions about what that means right I mean, they have issues at home, they're going to come to work.

49:07Whether you see them or not, they're going to be there. And so, yes, you have to lead with empathy. You have to understand that you're seeing this person for a small amount of their life and they have a whole life out there on the one hand. On the other hand, the more, and this is maybe going to sound slightly Machiavellian, but the more of them you can bring into the company, the longer you're going to keep them, the more excited they're going to be, the better it's going to be for the company. And so think about how you kind of engage, you know, more of the people who work with and for you. One of the drawbacks of having phenomenal people here, most not drawback, that's the wrong word.

49:46One of the potential concerns is that they don't have to come to work here. These people can literally get jobs anywhere. And so it's very different. So you have to make sure that it's exciting enough that they want to come back to work. every single day um and we we've done out of two sigma by doing kind of notoriously geeky things right we are we we've talked about being the best place for nice geeks and one of the things we did very early on was have a a hacker lab so we have a hacker lab which is exactly what it sounds like it's full of old equipment um where we do things from the obvious soldering to kind of knitting and crocheting and baking.

50:29And we call it a gym for the mind, right? The idea being, you know, it's clear, I think, to every manager in every company that if you have a gym, it's going to be better for your employees. Why? Because they get to exercise and physical exercise is good and, you know, pumps the blood. The same thing is true for your mind. And the more you can encourage people to kind of use that whole brain that they have, the the more chances that more of that brain will be used to further the aims of the company, which is ultimately what we all want. What a great note to end that on. I think that's such a good takeaway for everybody.

51:05And for me, this was like one of those episodes that makes you want to study both the old and the new. We covered so much great territory here that just is like evergreen, that you can really is relevant forever, which is why Matt and I have such an obsession with things like classics. And, you know, Matt, thank you for showing us that lasting innovation doesn't come from chasing things, but it comes from cultivating depth and challenge and curiosity and both yourself and your team. And before we wrap up here, you know, where can our audience go to learn more about Two Sigma and the work that you're doing there?

51:37Certainly begin at the website, www.twosigma.com. And we have on the website, we try and put blog posts that we make and kind of a window into what Two Sigma looks like inside. I think people are going to want to peer in that window because we covered some really cool stuff and people are going to want to go check it out. So we're going to be sure to put that in our show notes. And listeners, you know, if you're curious about what Matt's building, please go check it out. I've checked out extensively their website. They have some really cool videos, actually, that do a good job of kind of like visualizing how this data is impacting our world and the kind of things that Two Sigma is doing with it.

52:14Very cool. And so we're going to drop that in the show notes. And thank you, listeners, for joining us this far. If you made it all the way to the end, then you clearly liked this conversation, and you should probably go pick up a copy of The Odyssey. Be sure to subscribe and share the episode, and check out our sub stack for Leadership Insights, because this is only half the story. We're going to be covering more of it there. And that's it for this week's Dev Interrupted. See you next time. Thanks very much.

52:50you