Pod Insights a16z Podcast Marco Argenti (Goldman Sachs): Turning Developers into Clients About this episode In a new episode of In the Vault, a16z General Partner David Haber talks with Marco Argenti, the chief information officer at Goldman Sachs, about bringing fintech processes into financial services, turning developers into clients, and how AI is a major inflection point in the history of technology.
“In the Vault” is a new audio podcast series by the a16z Fintech team, where we sit down with the most influential figures in financial services to explore key trends impacting the industry and the pressing innovations that will shape our future.
0:00 - Intro
1:30 - From tech to financial services
12:05 - Turning developers into clients
17:30 - GenAI and productivity
23:50 - The future of financial services
28:50 - Outro
Resources:
Find "In The Vault" on a16z.com: https://a16z.com/podcasts/in-the-vault/ Find more about a16z's vision for the future of fintech: https://a16z.com/banking-on-AI/
Find David Haber on Twitter: https://twitter.com/dhaber
Find Marco Argenti on Twitter: https://twitter.com/marcoargenti?lang=en
Find a16z on Twitter: https://twitter.com/a16z
Find a16z on LinkedIn: https://linkedin.com/company/a16z/
Subscribe to the a16z Fintech newsletter: https://info.a16z.com/newsletters-fintech.html
Transcript 0:08 Speaker 1 Hello and welcome to In the Vault, A16Z's FinTech Podcast series where we sit down with the most influential leaders in financial services. In these conversations, we offer behind the scenes view of how these leaders guide and manage some of the country's most consequential companies. We also dive into the key trends impacting the industry, and of course, discuss how AI will shape the future. Today we're joined by Goldman Sachs Chief Information Officer Marco Argenti. Marco leads the firm's global technology organization and is a member of its management committee. Prior to Goldman Sachs, Marco was a vice president of technology at AWS, where he ran several business units, including mobile, serverless compute, and augmented and virtual reality. He also spent several years at Nokia, where he was responsible for the company's developer ecosystem and app stores.
0:59 Speaker 1 In this conversation, Marco and I discussed automating workflows in financial services, turning developers into clients, and how AI is a major inflection point in the history of technology. Let's get started.
1:12 Speaker 2 As a reminder, the content here is for informational purposes only. Should not be taken as legal business tax or investment advice or be used to evaluate any investment or security. And is not directed at any investors or potential investors in any ACCZ fund. For more details, please see A6CZ.com/slash disclosures.
1:31 Speaker 1 Yeah, Marco, thank you so much for joining. Really excited to have you here.
1:39 Speaker 2 Yeah. Yeah,
1:39 Speaker 1 I was really excited to chat with you. You've had a fascinating career, really at the forefront of a bunch of kind of technology platform shifts.
1:52 Speaker 1 At AWS, obviously during kind of the rise of the cloud, and now at Goldman, you know, during AI. So maybe that's a good place to start. You know, maybe you could walk us through a bit of your career arc and you know, what were the kind of key learnings from each of those different periods and perhaps how does that you know inform what we're going through today?
2:08 Speaker 0 I've been in technology pretty much all my life. Um I've tried different fields in the within technology, but it was always been uh uh you know technology centered. And uh since I was a kid, I was uh you know, literally like you know, we're talking about 40 plus years ago. I was uh I was fascinated by by computers, I was fascinated by
2:28 Speaker 2 uh the
2:29 Speaker 0 fact that you could uh have uh this in impossibly malleable uh state machine that potentially you can get it to do whatever you wanted. Uh and so the limit was not the machine itself, it was more like yourself and then and your ability to come up with stuff. And so this removal of the physical uh constraints. So before computers, of course, I was kind of you know doing some electronics, and the frustration was uh you know burn your fingers or you you use you know you kind of burn down the house in the worst case scenario, and so all that within a computer in a computer in a completely virtual world was something that uh you know kind of made me dream of uh of something bigger. I guess uh the internet uh um was uh you know, I started just with regular software development doing uh um you know software of different kind, and then I started selling uh I had my own startup, but then really was the internet that kind of uh changed everything, and uh and uh as a result of that I moved to Canada uh in the in the in the 90s where uh I was fascinated by this idea of uh uh the kind of limitless uh um distribution of knowledge, uh, which is something that by the way, it's kind of something that uh we'll we will uh uh kind of reconnect when we talk about AI.
3:49 Speaker 1 But
3:49 Speaker 0 this idea of actually taking away the physical barrier of uh of access to knowledge, and so I started an internet company in the 90s. We started the one of the very first uh uh online stores. Uh in fact, we launched uh a little bit before Amazon, and uh obviously we weren't as successful, so timing was great, although something else probably wasn't and uh that was relatively successful. Then we started to do to turn the store into an e-commerce platform, so going into B2B, and we built uh uh you know, also in collaboration with Microsoft, we built some of the largest online stores back then, online stores back then in the US. But then uh mobile was kind of the next obvious thing. And it was really early days. We're talking about uh 1999 or so. You know, I actually moved to Europe where a friend of mine was doing an IPO of a company and he acquired the mobile company almost like as an afterthought.
4:44 Speaker 0 And I was like, no, no, no, that's actually the most interesting part of the whole thing.
4:49 Speaker 1 Yep.
4:50 Speaker 0 And so uh became the CEO of that company was called Wireless Solutions, and then we started to really kind of go down the path of uh mobilizing the internet, creating uh um like infrastructure for uh for uh uh for uh others to create mobile websites. We did uh you know something called an SMS gateway, which was like a messaging system and so on and so forth. Okay, but so then I went into apps. I went into kind of this idea that uh the mobile web uh was probably not rich enough for a small screen and uh and and for the experiences. So we started uh actually the first app store that I know of, like the OVS store, which got pretty successful because of the install base. So we were doing something like 17 million downloads a day or something like that.
5:36 Speaker 0 So it was actually quite a big thing. Um I guess that was one of the biggest learnings of my career, which is uh A, again, you need to be kind of the biggest user of your own product, and and the second one is that sometimes you know, like little subtleties in user experience have a disproportionate effect. Obviously, I at that time I was uh dealing with a lot of app developers, apps uh uh require a back end, and and those developers were really great at doing the application on the phone, but then uh it was actually hard for them to do a backend to stand up with the server side uh of things. So sometimes apps were you know had literally like a server in their under their people's desk as uh as a back end, and uh and so this whole idea of a kind of compute uh on tap was so fascinating for me.
6:25 Speaker 0 And uh, and so when you know Amazon called me uh and they're like, hey, you know, want to come with us and kind of do you want to actually start kind of the mobile uh services, a back end uh uh uh business within within AWS, I kind of jumped to the opportunity. Yep. We started to go down the path of like really that the internet and compute and uh and then serverless should really be everywhere, should really be in every in every uh in every object. Well, then I mean that led me to um start to talk to a lot of companies that were like, for example, in manufacturing because we were getting into IoT and uh and and supply chain uh and then started to talk to CEOs about uh uh the whole idea of digital transformation, really like something that was pervasive throughout the organization.
7:14 Speaker 0 And so from that on uh I started to feel this uh like okay, you know, like it's pretty clear that uh technology is really moving from kind of the back office into the center of the strategic agenda for CEOs. So I was talking, for example, to the CEO of Volkswagen or the CEOs of uh you know Philips and other companies, and all of them the common theme there was we're starting to actually be limited by our technology capabilities with regards to our competitiveness, and we need to really form and train a muscle and technology that is just as good as any other company, including technology company. So then I met David Solomon uh, you know, at uh at reInvent, which is like the AWS conference where he was speaking, and then we started to talk, and then you know, it was pretty clear that uh Goldman had uh the same uh uh you know vision that uh especially in the world of finance, uh technology will play a disproportionately important, uh important role.
8:13 Speaker 0 And that you know, that's what led me uh ultimately to uh to join Goldman in uh in 2019. So that that's kind of I know that this was a little bit long, but uh that's kind of the arc uh of uh about 40 years in about 10 minutes.
8:27 Speaker 1 40 years and four minutes, that's perfect. Um, it's awesome. And and that's obviously I mean brought brought us to the time that we met, you know, back in 2019. Yeah, um you know, maybe um yeah, maybe you can talk about kind of that transition, you know, uh, you know, coming from uh you know business like AWS and and Amazon, you know, two financial services, which I you know, I think was new for you. You know, what were some of your expectations versus reality? You know, I know you brought a very kind of you know Amazon like approach, you know, to the firm at the time, and maybe just talk about that transition.
8:58 Speaker 0 One of the big things when you do a transition from a tech company to uh you know a company in a different sector is whether that sector is regulated or not. And of course, regulation uh is something that you know can be in in a way a barrier or or slowing you down, and you can think about that as a slowing innovation. But then kind of I learned uh, you know, maybe from my control theory background, where you know I when I studied in university, I was studying like I studied like my major was on uh on control systems, by the way, based on uh an early form of AI, that you can actually use controls at your own advantage to kind of gain uh velocity in the long term. So that was a big learning, kind of an adaptation where regulation and other controls are actually very important uh for sustained speed of innovation.
9:48 Speaker 0 So you might be slow at the beginning, but then eventually you'll give uh velocity. And so that was uh um uh something that I think anybody that comes into a regulated industry has to realize. The other thing is uh really like introducing uh in the engineering. Uh team, the concept of kind of working backwards from the customer and really like turning it into a principle that everybody can understand. So we the we developed this leadership tenet for engineers, the first of which is like build with purpose. And it kind of sounds fuzzy if you talk about it and kind of you know, sort of obvious. But really, this idea that engineers, in order to be good engineers, needed to really kind of sit down with the businesses and really deeply understand uh the why.
10:34 Speaker 0 Yeah, because engineers they tend to start from the how.
10:38 Speaker 1 Yeah.
10:38 Speaker 0 So if you if you get a bunch of engineers uh you know in a room, uh, like you know, for example, if they do if they need to do a a web project, uh, you can count the seconds before the discussion on frameworks come up,
10:51 Speaker 2 right?
10:52 Speaker 0 And maybe not that that's a good discussion, but not the discussion I would start from. And so this idea of kind of saying, okay, you know, we need to ask ourselves fundamental questions who's the client, uh, and really, you know, what is the problem or the opportunity that we're trying to solve, and then uh uh how do we measure that? How do we know that that's actually something that uh the you the client will uh will uh will like those questions, and then having them written into narratives, it kind of fundamentally started to change the culture because engineers tended to kind of get really fuzzy um uh requirements from uh uh from the businesses, and then kind of you know go into sort of a seriousness of approximation. Whereas this idea of shifting left and actually writing down a plan that is very solid that any business person can understand before you write the first line of code, and so this idea of introducing the concept of of uh platforms was uh was extremely important, uh, and really to kind of you know speed us up.
11:50 Speaker 0 And now, again, fast forwarding, we have about 10 of them uh firm-wide, like uh that are heavily adopted, like cloud platform, the data platform or uh the identity and access management platforms, uh uh the API platform, mobile platform, and so forth. Yeah, that developers can just build on top of that. And
12:07 Speaker 1 I think that I recall, I mean, that being actually a really important kind of message, I feel like that you brought to the firm, which was you know, we're gonna make the developer our client. You know, and I think it was kind of one of the first times that Goldman thinking about the developer as a client. It was always a corporate, an individual, an institution. And I think that it was developers internally within the organization within other divisions, you know, kind of leveraging as you're describing kind of platforms built in core engineering, but it's also developers at other companies who can build, you know, kind of embed Goldman's capabilities into their distribution or into their workloads. Maybe talk a little bit about that kind of shift. Yeah, and
12:44 Speaker 0 that was definitely a new it's the same thing. Remember when I said that you kind of need to understand as a leader what uh uh what your uh clients are thinking by kind of trying your own uh product, and I think also today, given the importance of developers, uh I think it's also a responsibility of a leader, not only a technologist to try to really understand what developers are thinking and what they want and uh and and how they think and meet them where they are externally and internally, because sometimes uh even externally in your clients uh the decision makers are not only the business people, but are also like the choice of technology and and and which leads them to the choice of a service is done by developers, totally so you really need to think of them as uh decision makers, uh, even when you're dealing with uh with an external client, and we're seeing that more and more uh every day.
13:32 Speaker 0 And of course, you know, some some of our clients are kind of more like that, like for example, some of the hedge funds or some of the quant funds uh, as you know, I mean, you know it really well that you know, we're talking to the technologists is just as important as talking to the to the business people, yeah. And then you kind of there is a little bit of a spectrum there, but I think uh uh really back to the idea that uh um really the role of technologists uh have really changed dramatically, not only in internal dynamics, but also in the way you actually do business. Totally.
14:04 Speaker 1 I guess to that point, you know, another I feel like shift that has happened, um, and I feel like it was happening while I was there, and I'm sure it's even only accelerated was kind of this willingness to adopt third-party technology. I think Goldman, you know, famously had built a lot of internal systems. I think I remember you know, folks talking about the kind of word processing system the firm had built in the 90s or an email client. But I feel like you know, I mean, you you very much understood what it meant to be a platform, obviously AWS was a platform that other people built on. I guess how do you today you know think about leveraging internal engineering resources in terms of
14:40 Speaker 2 building software?
14:42 Speaker 1 How do you decide what you need to build internally versus what you're willing to partner with third-party companies or you know, buy software? Like how do you how do you sort of reconcile that decision for what needs to be owned and built in-house versus leverage from the outside?
14:57 Speaker 0 Yeah, so uh at the very beginning, we kind of introduced uh uh our digital strategy tenets uh if you remember, and one of them was uh uh look outside first before you build. Just because you know you have a bunch of smart people, you don't necessarily need to build everything. And I think to me, the kind of the asset test is uh where do I actually want to employ like the you know the uh you know my best people and my best resources into which kind of tasks and and is that uh something that will create some form of competitive differentiation
15:28 Speaker 2 or
15:28 Speaker 0 you know, contribute to the growth of the business, etc. If it is commodity, if the problem is already being solved by someone else, uh, there isn't really good argument to actually build it yourself, and also like uh it doesn't just apply to vendor software, it applies to open source in general. Obviously, like open source uh is uh is an incredible wealth of uh of intellectual property where you know there is tens of thousands or millions of people that have actually spent uh you know hundreds of millions of hours in kind of figuring out some of the most uh difficult problems, and so kind of really like uh asking that question first, uh, which is uh okay, what is out there that could actually give us this functionality, and then of course you go into a very rigorous selection process because you need to make sure that the solution is widely adopted, that is supported, that is safe, that is secure, etc.
16:20 Speaker 0 etc. And so many in many cases, you know, uh like we decide to use a vendor or we decide to use open source, and then we build uh kind of you know, like let's say a scaffolding around it to make sure that uh it integrates with our systems and with our data, uh, B, you know, it operates in a compliant and uh and and you know and and and and safe and secure way, but then at the core, the functionality, it's something that you can talk, you know, you kind of kind of leverage to uh uh to make sure that uh you're kind of removing this undifferentiated uh heavy lifting in a way and uh and and really focus on things that are uh that are creative to your business. Open source is uh you know something that of course you need to be careful about, you need to be extremely selective, but it can it can it can actually um help a lot uh you know in in kind of uh I would say uh put your priorities or your or your capital uh uh uh you know human capital in the things that really matter, totally instead of reinventing the wheel, but it's also something that actually motivates developers, especially when you uh when you when you create an open source project for others.
17:26 Speaker 1 Um I want to shift gears a little bit because you know it feels like we are entering a new platform shift, you know, with AI. Um and I'm curious, you know, how are you thinking about you know generative AI? What are you guys already doing with it? Um, and maybe we can kind of imagine the future forward and think about how it may impact you know, kind of financial services or Goldman specifically.
17:48 Speaker 0 I will, you know, like I would I would start by saying that despite all the skepticism that I try to apply, and uh, you know, and uh you know trying to suspend like suspension of belief in this case, but uh I'm I'm pretty profoundly convinced that at this point uh that we really kind of hit uh uh a major inflection point in technology that has really profound uh impact. There are societal impact and and and business impact, uh probably like uh, you know, and again, in my history of uh this 40 year is for me, that's uh probably bigger than any of the revolutions that I've seen. Also, because of how quickly that almost became almost like an essential tool for everyday's life for a lot of people uh that I know, including myself and my daughter. There was a time in which, in order to study a subject, I needed to physically go to a library, and then it was like you know, it took me hours before I could actually read something.
18:43 Speaker 0 If I didn't understand the subject deeply enough, I kind of had to match myself to the right book, which was like you know, not easy, and then climb this ladder of knowledge until I was finally equipped to understand a certain subject. Physical barrier removed, but knowledge barrier was still there from you know the age of Gutenberg to now hasn't really changed. It has it has changed now. That's why I kind of sometimes compare the invention of AI to the invention of the printing press in a way that it removed uh a major barrier. The first one was a physical barrier, this one is uh a knowledge barrier. Your prompt most often determines uh the level your level of knowledge and the AI actually explains things to you that are uh um attuned to that level of knowledge, and then you can kind of climb that ladder by asking more and more questions and interactively, you know, the AI will kind of get you deeper and deeper into a subject.
19:38 Speaker 0 It's almost like having a personal tutor that can be available now to billions of people, which is kind of unprecedented. And I think that will have profound uh uh repercussions on uh the way we kind of shape the education system and the way we shape uh you know just the access to knowledge, but also how quickly you can ramp up. And become proficient in your own job. When it comes to uh impact on the business, I kind of tend to think about that and also kind of the way we think about it in Goldman, like as a sort of a spectrum where uh sort of uh you know on the left side, which is kind of the more immediate uh um uh opportunity is kind of the impact on productivity. And I think that could be pretty uh you know, pretty uh uh significant, possibly one of the potentially more significant significant productivity uh enhancements that we've seen.
20:27 Speaker 0 And then you kind of walk the spectrum all the way on the right side where you have more like uh uh you know, sort of alpha generating or or growth generating use cases, which might not return on investment right away because you actually might need to train your AI, you might need to uh you know make it really like understand your business uh and will require investment. And so as we navigate that spectrum, first of all, one of the things that I was really kind of and I am really kind of keen on is hey, this is something really new that we profoundly do not understand. Nobody really understands AI profoundly yet. The things that we don't know are probably more than the things that we know. So we need to kind of sandbox that uh into an environment that uh will allow us for safe uh innovation and safe experimentation.
21:16 Speaker 0 We mandate that everything that we do is start starts from the understanding the why and starts to understand. So we have set up uh uh you know a working group or a committee that kind of reviews all the uh the ideas, and we selected out of hundreds idea of ideas we've selected about 15 or 19, actually at this point that were kind of you know at different phases of piloting. I think the very first thing that we're actually starting to be uh using AI at scale is is developer productivity back to developers, right? Yep. And uh and we're seeing the developer uh productivity go up, you know, can be anywhere from 10 to 40%, depending on the developer, and depending on the use case, which is massive, even if it stays at 10%. But then also it has a direct impact on developer happiness, which by the way, you and I know that uh it is one of the most uh important factors for productivity because a happy developer that is in the wave that is in the in the zone actually is you know, like 10x, you know, like the famous 10x uh productivity gain uh is pretty well understood, and then of course it's a journey because at the beginning you implement tools that are kind of generic, but then you start you know adapting the whatever co-pilot type of AI you're using for your developers with uh institutional knowledge.
22:33 Speaker 1 I think that's like an it's an interesting point because you know, I mean, first of all, people people you know may not know or or forget that you know Goldman has I don't know a 10,000 plus you know engineering organization, so it's you know a pretty material percentage of the overall you know head count of the company. So even just you know the 10 to 40% impact on productivity, you know, can have a significant impact on bottom line. Um
22:55 Speaker 0 but then you can extend that to how many of our 45,000 are actually people that are in the knowledge business. I mean, and so most of them, right? And so then when you started expanding to that to bankers, to investors, uh you know, to uh uh personal uh uh you know, like to private wealth managers, uh you know, and then even eventually to uh to registered investment advisors. Uh so that that knowledge uh it's so powerful because it really changes at the end of the day, it changes uh not only the uh you know your own productivity but also your dynamics with your clients, and then eventually we'll actually become the new normal in the way actually clients expect to deal with you and actually can create the competitive differentiation uh on uh on the way we actually create products and make decisions and create strategies and so on and so forth.
23:46 Speaker 1 If if we kind of you know imagine kind of into the future, I'm I'm curious what you think AI's impact might be on you know financial services as an industry, you know, maybe 10 to 15 years from now.
23:58 Speaker 0 You know, it's funny because uh this is one of those things in which uh it's even hard to predict the next six months, right? With the speed uh with the speed of uh things that are evolving, and there are like really still very big open questions right now. Uh first of all, it will be a sort of a table shifting type of uh of phenomenon where uh uh you know it will create new leaders. Um it will reinforce uh uh position of some of the leaders that are actually able to kind of be at the forefront of this and be able to uh leverage that in the proper way. Um so I think it will create differentiation for sure. I think uh uh you know, one trend uh in financial services has been that uh speed is becoming more and more of a factor. Price is important, but uh how quickly you can price is also very important, uh, especially with some uh classes of clients, and the switch from batch to real time and more like having a real-time uh a real-time business where uh uh you need to make decisions really fast, that you need to settle really quickly, that's uh you know, you need to go through front to back extremely quickly.
25:06 Speaker 0 That's uh and so all that I think uh this shift towards more of a real-time business is something that AI definitely will accelerate. Uh the ability to have very complex uh models of the world and of the economy. I think the relationship with clients will change. And as I said before, the expectation of clients of you being uh extremely uh good at providing personalized advice, personalized portfolios, uh understanding deeply what are the repercussions of certain events in your uh in your investment strategy and being able to kind of provide that kind of information at uh at a speed and at a quality level of quality that is probably unprecedented.
25:46 Speaker 1 Anything we didn't touch on that that you wanted to you know kind of discuss uh before we end.
25:51 Speaker 0 I think uh back to uh what we quickly talked about uh you know, like some of the big open questions. I think one of the most interesting and fascinating questions right now is uh how to kind of uh navigate uh I have open source models uh that might be maybe less capable in a way, but could be could have advantages such as uh you know they can run on premise or they can be fine-tuned uh um etc. Or you know, do I rely on uh uh you know benchmark performance or uh reasoning capabilities or emerging abilities are kind of the most important things, and that do I rely on uh on large proprietary models. And I think that's one of one of the big strategic questions that every CIO needs to face today when uh when kind of adopt an AI strategy.
26:34 Speaker 1 I yeah, I'm very curious, and also if you're fine-tuning on your own kind of corpus of data, but yes, I'm curious your kind of your your perspective between some of these large kind of closed models versus maybe open source.
26:44 Speaker 0 Probably nobody beats those large models with regards to actually reasoning capabilities, okay. But then uh there is a lot of uh uh specialized knowledge where uh you need to somehow tune the model
26:56 Speaker 2 in
26:56 Speaker 0 you know, and and it doesn't matter if it is fine to pre-training or if it is fine-tuning or if it is rag, but you kind of have to bias the model somehow. And and and then you know, you can think of a future where uh you almost gonna have uh, you know, I I kind of call it hybrid AI. It's kind of the same way as the world of the cloud evolved, where you know you you know, every company at the end of the day has uh a part of their workloads in the cloud where you leverage elasticity, agility and scalability, but then you also have on-premise low workloads, which might be necessary for latency purposes, sometimes for you know data sovereignty purposes or for whatever other purpose uh you might you might find relevant. And I think the same is gonna be with AI, where you're gonna have uh a reliance on uh you know large proprietary models mostly for uh the reasoning capabilities, almost like you know, I really want to understand what you're asking.
27:51 Speaker 0 And then kind of they will generate and they will orchestrate a set of smaller models uh which have been trained and highly specialized on specific tasks, your own data on uh um, you know, like if you have uh like we do an internal language, which is slang, uh, you know, train it on your own that or your own knowledge base or your own institutional knowledge, your own environment, and then uh or performing specialized calculation, like I don't know, calculating an efficient frontier or whatever that might be, and then uh they return uh you know highly accurate uh results back to more like the general reasoning uh uh AI, which then does a fantastic job of summarizing it for whatever purpose uh you want, almost like a constellation of planets and satellites.
28:35 Speaker 1 Marco, thank you so much for for the time. Always a
28:39 Speaker 0 thank you so much.
28:39 Speaker 1 Uh I gotta go to your next concert here in New York City. Uh for those that don't know, Marco has a great grunge band. So uh not only is he uh an amazing CIO, he's also an amazing guitarist, and uh yeah, I just really appreciate the time and uh always a pleasure.
28:52 Speaker 0 Thank you so much, David. Thank you for having me.
28:55 Speaker 1 I'd like to thank our guests for joining in the vault. You can hear all of our episodes by going to A16z.com backslash podcasts. To learn more about the latest in FinTech news, be sure to visit A16Z.com backslash fintech and subscribe to our monthly FinTech newsletter. Thanks for tuning in.
Previous Next-Gen Gaming: AI Souls, Real-time Culture, Personalized Avatars Next Can AI Advance Science? DeepMind's VP of Science Weighs In Page updated August 25, 2026 · transcript and summary are machine-generated