Beyond the Breakthrough: Who Shapes the Future of AI? | DVC's Marina Davidova

I am so excited to welcome all of you and I'm blown away by the talent in this room. I see amazing community builders like Tonnie and I see people building the future, connecting AI, using AI to connect people. I see Cedric building that I see so many talented people. We have an amazing selection here on stage with us today and we're going to jump right in. as we were talking earlier, my name is Jenny and I'm the co-founder of Women XAI, where we elevate the voices of women in the AI era. Last year, we handed the mic 247 times to women. We're going to do more than that this year. Let's do intros down the road. Meg. Hi, I'm Meg. I'm COO and co-founder of Mixies and we create digital infrastructure for IRL communities like this one here today or like large conferences to find what they need and coordinate at scale.

my background is in VR and gaming and tech for the last decade and also very specifically I've thought a lot about just how do we use technology for leveraging better opportunities for human connection. Hello everyone, my name is Shub Shvastava. I am a cloud architect at Google focusing more on agentic AI and AI infra. I take care of most of the best enterprise customer for Google cloud and I'm really really excited to be here. So thanks to Jenny for making me a part of this amazing panel and I'm really excited to talk to you and discuss more about breakthroughs we're building in AI. Thank you.

Thank you Anastasia.

Morning everyone. My name is Anastasia Zamskova. I am a VP of strategy at Jet Brains. We build developer tooling and we've been building developer tools for 25 years and counting. And my job is actually to make sure that in this new AI reality, we as not a newcomer, but as someone who is having institutional trust from developers are actually paving the way into how AI could be adopted at organizational level at scale. previously I was one of the founding team members at NBS. This is one of the largest AI infrastructure companies and a CMO. I'm super glad to be here today. Thrilled to have this discussion.

And my name is Marina David. I'm managing partner at DVC. we've been investing in AI for the last 12 years. U 12 years ago it was still a niche. Now everything is AI. I'm glad that we are all here today and I'm glad that we start the day with women in AI because it's the best way to start the day.

Absolutely. I couldn't think of a better kickoff. So, one of the things I really want to talk about today is what happens after the AI breakthrough? How do we operationalize that AI? You know, they're in the research lab. They're looking at things and discovering new things, but how do we apply that to business? And so I'm really curious who determines what AI becomes when it comes out of the lab. And I'd love to hear from each of you on that. Who wants to start?

Probably I can start. So the model can give us an opportunity but I I believe only only real world can determines what what survives and yeah we can the capability curve is moving very fast right now. later today during the conference our friend Vanessa Parley will introduce her Stanford index AI and she will cover that last year only 12% of all task has been completed successfully by AI agents. Now the rate is 66 which is amazing

but don't don't get me wrong AI agents are still make mistakes still hallucinate and doing loops so and sometimes behave like a very confident intern that makes mistakes but at the same time we already the question is not about like whether AI can do job can do tasks it's whether how quickly companies can absorb it and how quickly can we scale and answering who determines how what AI becomes. We could we could ask for probably engineers who build AI into systems or founders that build it into products or maybe even capital because I'm venture capitalists that back these products maybe even I don't know regulators who answer what is acceptable what is not but this is all these answers are not correct and the only one correct answer users determine because customers consumers ers and enterprises we all work with our dollars and without it there's going to be no capital no companies no products no systems so breakthrough I would say gives us like raw electricity but only users with their dollars determines whether it could be light or fire I don't know just noise

so all of us in this room get to decide we decide with our dollars with our tokens and where we spend our time. So, that's really cool. And you mentioned a couple different layers of looking at it. And I'm really thrilled because we have all of those layers on the panel all the way from venture to people who are building at their company to people who are scaling large corporations. So, I'm really excited to hear from everyone. Anastasia, do you have anything to add?

Sure. It's a great topic to talk about. Yes. So overall I feel like infrastructure builders around AI adoption are heavily underestimated in a way in the current narrative altogether because of course modal capabilities are great. modals can do absolutely astonishing stuff. But for enterprises, for organizations to adopt and to start doing that astonishing stuff with the modals, so many things have to be in place actually to make sure that we can govern to make sure we can calculate to make sure we can scale overall. and there is an important question in the philosophy of language too which is about whether language is actually just capturing the world describing the world or constituting the world and I believe that it's the same about AI narrative too it's not just describing what AI can do but it's defining it in a way and right now the definition is happening right on the side of the ones who are actually adopting and not on individual level but on the level of teams on the level of organizations.

So I would believe that the people who are owning the narrative.

Absolutely. And I've been hearing a lot about that. I've seen a lot of folks who have stepped in individually and now they're trying to bring that into their team or their company and it's been really excited to see it grow. I know you do a lot of work with that at Google. Can you speak to how you're seeing teams grow or your teams?

100% and plus one to whatever we have discussed so far. I I feel like we are at that juncture where we are not just you know focusing on the models and the capability because we have seen they can do breakthrough work. We have multimodal capabilities. We have omni models whatnot. But to your point the discussion is about how we are like extracting value out of these models. How we are adding some value out of these models to our day-to-day lives. So Jenny, to your point, I want to kind of like give analogy of a car. Like the model is the engine, but we do need like the steering wheel, the tires, the road, the infrastructure,

the brakes.

The Exactly. the brakes. That's the most important thing for it to work and function right. So for me, the decision maker or the people who decide how it becomes reality are the builders. how we are thinking about governance, security, how we are doing the unsexy work of like having a hardware, infrastructure, GPUs, I think everything kind of like contributes to that conversation. So, plus one to what Anastasia said. Yes,

thanks for shining light on that. The analogy is perfect. Meg, I'd love to hear from the startup founder side, what does that look like for you? Yeah. I mean, when you asked the question, I automatically was like, I think it's us because that's literally like the kind of the whole reason of why I do what I do and why I wake up is because I want to have a role in how these technologies are shaping society. And the way that we shape society is through the tools people actually use and are actually being in front of people. It's not all the other stuff that's happening. And I think very much here in San Francisco, we live in this bubble where we think everyone knows all of these different like Claude and Chat and Gemini and like outside of this bubble, most people have only maybe heard of chat at this point like they don't even understand all of the other models and things that are happening.

They just understand like the tools that they're using in their day-to-day lives. So I think from a founder perspective that's what's so exciting about being able to build at the forefront is that we get to actually decide what this becomes in the world.

That's amazing. And we do a lot of work with women xai elevating voices but also educating. And so we teach executive women how to use AI and I'm always surprised when we have members who are not from the Bay Area. there's just less exposure to the AI that we're all using today. so it's really interesting and I love that you're building stuff to bring that to people. I'd love to do a show of hands in the audience. If you're a founder, raise your hand. If you're a builder, raise your hand. Keep your hands up, founders. You're builders, too. All right. Yeah. Thank you. so this question is for you guys to get a great answer for Marina. Marina, I'm curious, you know, from the investment side, what signals tell you a company is building something that can become a real thing, the infrastructure, really digging in instead of just riding the current wave.

I mean, you've been doing this for a really long time. You've been doing AI before it was cool. How do you tell what's hype when startups come to you and how do you tell what's something tangible?

well, first of all, we are all riding this AI wave. otherwise we risk drowning into it. So we're just enjoying it like surfers until it crashes. So, but for me as as an investor, the first signal is how close the company to the workflows that are really painful because if it's just like a beautiful demo on top of the model, I can get nervous. But if it sits in between the processes that are painful and repeated and budgeted, that's when I'm pay attention to it. The second one is is whether it's enough context. and I'm not talking about just like abstract data. so if the company have enough operational context about how users behave or what the intents like how where memory stores where permissions all of this like ugly plumbing inside layer but if if the company have this context they they in a good position and probably the last signal that we are looking for is whether the customer is is ready to change its behavior to u adopt AI into the processes into the workflows and redesign the processes because it's it's very different from like we tried it and we liked it then like okay we are ready to redesign the structure and architecture of what we are doing like and u we are ready to build products around it that's an infrastructure signal for me

that's really interesting thing I hear that you're talking about where the customer is and when the customer is ready to go on the journey with you and I think that's something that's super important. Thank you for sharing that with us. I think it'll help all of our builders in the audience today. My next question is for you Anastasia. So my understanding is that Jet Brain sits very close to the builder layer of AI and I want to know how are you seeing AI change the way developers actually work dayto-day? What's that look like for you? For me, I think it's not even about how code generation got changed because this is not a secret for anyone, right? That people are not writing code with their hands as they used to before. Right? Now, it is agents that is doing a lot of job, right? I think that the main shift is in the cognitive load overall because right now it's not about writing code but it's about reviewing code and about getting responsibility for the code that wasn't written by you and so it feels overwhelming that you are constantly having kind of a junior developer just as Marina mentioned right that is working together with you and acting smart but sometimes just confidently wrong.

And you need to still make sure that the code that you're shipping is actually the code that is not going to break things and you're still accountable for the code because accountability is not delegatable. And there is a lot of pressure on developers because of that and it all is happening so fast. And there is also another thing that everyone are right now their own engineering sorry integration engineers in a way right because they are trying to put together everything before they've been working only in integrated development environment and ideides precisely the tools that Jet Brains has been specializing in for 25 years but right now there is also lots of platforms and also CI tools and it costs a lot of cognitive heavy weight lifting that people are doing which is also tricky. So I would say that the cognitive shift and how the job itself is is changing is probably the main thing that is getting that is getting shifted right now.

Thanks for sharing that. And I've heard from a few founders and Meg, I'm not sure if you're willing to share your story that you told me a couple weeks ago about when you spun up a couple different agents and was managing them and how how that went for you.

You mean the Wait, which story? Remember?

There's a few. the one I wanted to I thought was relevant was when you were saying you spun up a bunch of agents and they were doing many things for you, but then you had to review it all and it was Yeah. Yeah. Yeah.

Yeah. I mean, I think who if anyone can find out a way to do like QA at scale very fast and that's going to be like the next super unicorn for sure because yeah, our team is just building way faster as a whole than we ever have before. and then me myself who I do not have a technical background also just building way faster which is maybe scary to the rest of my team. and so yeah, just and beyond also just the the QA, it is like this a lot of cognitive overload as you're saying that's very overwhelming and draining, right? So how do we kind of like parse that out and and be able to control that better? Yeah.

Yeah, absolutely. I'm I'm seeing that as well. Shu, I wanted to talk to you about what changes technically and operationally when an AI system or AI product goes from demo into deployment.

Yes, definitely. And I want to do plus one to what Meg was saying. We need some really good eval because that's where you want to decide those breaks for your enterprise. and to answer your question, Jenny, I feel like we started not even long ago, like two, three years ago, where we just we were just prompting these LLMs, then we added context, then we were doing rags, then we gave them data sources, but now the agents are autonomously making those action on your behalf. So, it's like they're acting on my behalf and they are kind of having that agency, right? So it's very different from like from moving from the demo to the deployment especially when they're making those decisions and those decisions could be updating a database or updating like rolling out a transaction. So from that demo part when we when we when you're demoing something or we are building the prototype it's all about what is the art of possible what I can do but when we are going into deployment to your point you mentioned we have to think about tokconomics we do have to think about cost governance if I'm giving all the access to my agent I don't want them to go rogue and like break bunch of processes which are already there right how do I integrate everything into my system als Also I feel like tooling for building these agent is so fragmented.

So having that cognitive load as well like what do do I learn? How do I create the framework? How do I deploy it? If I have a longunning agent, how do I make sure that it's sustaining for like days and I have the enterprise ready that's deployable and it's ready to serve these agents. Have really good eval because agents are non-deterministic. So to perform a particular action, they can take different routes. So having all of these tools in place I think changed that shift from like demoing something, prototyping something into actually becoming more, you know, aware of like performance, latency, cost, and all of those factors. So yeah,

thanks for highlighting that. I think there are a lot of pieces to getting your AI ready for prime time. one of the things I wanted to talk to you about, Meg, is the relationship between humans and AI. I think that that is so important. We heard Marina mention it, you know, it's the consumers that also need to be ready and u people really making the future of AI together. So your f my understanding is your work focuses on AI to help humans connect. can you tell us more about the role you see AI playing in human connection and what you're building at Mixies? Yeah. So, I think a lot of the companies we're hearing about more so at least on social media and like on Twitter is how are humans being replaced by AI? And I think that's causing a lot of like anxiety and fear.

And it's like replace your you don't need a co-founder anymore because you have Claude or you don't need a lawyer because you have Harvey, right? It's like you don't need an Uber driver cuz you have Whimo, right? It's like constantly we're being shown, I think, how little we need each other and how kind of unimportant we are to one another because now we don't need to rely on people in the same way that we used to. And I think like when you look at that one by one, it's not so bad. But when you look at it like holistically and if we get to a point where like 90% of the technology that we're using is all about like you don't need anybody.

I think that's kind of a scary place to be as a society. Like what does that do to us? so very much I've always been really focused on like know how do we build the other 10% that is instead about how do we use technology to better help us have opportunities for connection. And so like for example with what we're building with Mixies, right? We want to help each other find and connect the dots at scale that you wouldn't usually see. And so like at a conference there could be like 20,000 people in the room, right? you are most likely gonna talk to maybe five over the whole entire conference. Maybe you remember two people's names. Like we don't have the capacity to understand the context of each human in that room and to even have the memory to put together the dots of how we are relevant to one another.

So when we're thinking about how we're using the Mixie platform, that's really what we're doing. like Mixie is able to chat with all 20,000 people, have that context, understand the nuance of what each person is looking for, and then connect the dots on the platform to find you that one person in the room, that needle in the haystack, who can move your life forward in the way that you're trying to move it forward.

That is so cool. And just being able to use technology to bring people together, I think is really beautiful. And I'm excited that we have founders thinking about it from that lens. I also wanted to talk about you know right now there's a lot of conversations around new models and oper operationalizing AI and I want to talk about what happens you know the market is maybe underestimating some of the complexity that comes with that I know we touched on it earlier but I'd be curious to hear from any of you on the panel what is the most underestimated piece when we go to bring AI into teams into companies

I think that the most underestimated piece here is that the capability ceiling is definitely right now far beyond the ceiling at least temporary of institutional operationalization right

and that's for sure and the narrative overall feels like still being very concentrated on the individual level consumption

well it is only starting to shift right now when people are starting to talk okay now AI has proven its superpowers on individual level, what next? How could we make sure that the teams are actually getting the best out of it? How do we making sure that the investments that we're putting into it on the org level is actually paying back? And so I would believe that if we think about organizational kind of problems that it's not always about AI and technology limitations. It's about how people are actually working together and how organizationally institutions are actually ready to adopt AI and not to only automate the mundane tasks but actually to reinvent how stuff is working inside of the organization. I think that's underestimated.

Very cool. It's the people and the structure. I absolutely because I think that the only limitation technology is evolving very fast but the only limitation is that our human interface is not perfect. we are adapting very slowly and our human behavior changing it takes time to change and the biggest I think mistake is to think that like okay if we plug in AI we'll just get the productivity and everything will be faster but it's it's a mistake we underestimate how much attention it requires and sometimes it requires a lot of attention and AI itself is so impactful that we can reassess like the whole business from the first principles. but it requires like rebuilding a lot of processes from from these first principles. We all understand like okay like there's a model on top there's like we understand there's a magical user interface but there's something in between in the middle that is harder to get like it's that we all talk about evaluations and like I don't know always like middle layer orchestration data permissions governance and all that stuff and we underestimate how much work and how lots of like it requires new full stack of things that are has not been existing like eight 18 months ago.

yeah

yeah yeah the speed of change is insane

and like deploying copilot is actually a very easy part but rebuilding hot like old processes around it is actually leadership challenge in even like in our small team in our fun venture capital firm we I used to have like five people in the team now we have the same amount of people in the team. but we we are doing much more. We like I can say that we six times more productive. We can like u make more deals. We are looking at more pipeline. but it's wrong to look at this that I'm managing the same five people in the team. I'm managing five people and tens of agents. And sometimes agents require more attention. they sometimes have more hallucinations and tantrums than people in the team. Oh, absolutely. And one of the things I really appreciate about what you said, Marina, is the fact that your team size is the same, but you're doing more.

And I wish more companies were looking at it like that because think of how far you could go. Six times more. That's amazing.

Yeah. But at the same time, it's just not like five people. It's now we are managing people and AI agents and like

Yeah. They're teammates. They work together.

Yeah. Absolutely. And it it also requires like changing not it's not only technical questions how you're like building processes but it's also governance questions, cultural questions and sometimes even leadership questions.

Absolutely. That digital transformation that has to take place in your organization as you're including more AI. I think it's often overlooked but maybe one of the important most important pieces of actually having a successful AI integration that's delivering on the ROI that you expected. So, I'd like to close out the panel with a question and then hopefully we'll have some time for audience Q&A. And my question to all of you is really we are looking to the future. We have a ton of builders in the room. What do you hope they keep in mind as they go forward building? Is there anything that you want them to intentionally include as they build the future of AI for us? Meg? yeah, I think being aware of the incentives that you're setting up for the user with whatever you're building and what does that look like in the long term.

I think there's a lot of incentives for us as builders actually to build for the short short term and so I would say think about the long-term effects that you are having on the world and on the people that you're building for.

Absolutely. Thank you, Meg. Sure. I wanted to add one point to what Marina said before about you know what we're under underestimating and advice for the builder going forward is like in any process or any agent tech workload any agent any AI workload it is as powerful as the context you give it to. So my point being having that really curated data you know structured way of consuming data for your agentic processes that is really important and the second thing is reducing the blast radius right and me had a great point before like if there is a company that can do like testing these agents or evaluating and like how they're impacting your overall business I think having that observability telemetry is paramount before you're actually deploying ing that for your actual business users. So reducing the blast radius and having really good context and data for your agentic workloads or AI workloads to do get maximum out of out of it would be my advice for the builders.

I really like the lens of reducing the blast radius. I haven't heard that before in this context.

Anastasia,

that's so cool. Well, yeah. I think that mine is going to be a little bit less concrete, but what I would want the builders to think about is recoverability

generally.

Yes.

Because all of the AI ethics talks is about not making mistakes, you know, and trying to

be sure that everything is in place control and stuff like that. But mistakes are inevitable. They're going to be in there because even the smartest and the greatest is still making mistakes. So what we hear from developers for example is that they want to have environments where they can actually do mistakes and then they can understand what the mistake was about and get back

in a safe environment in a place that they can roll it back or you know not reduce their blast radius.

Exactly. And I think it's applicable for all of the spheres and all of the industries basically.

That's really helpful. I love that Marina. yeah, I think my what I would ask leaders and us as well to build intentionally very similar but like I would call it a trust. So I really want us to build system where people can actually understand that AI has been used that we can trace the decision making process where we can where people humans can still understand what they responsible for and I don't know automation is not an excuse to like reduce judgment to like go rid of judgment at all. So because I the next area now we are always talking about the speed and how technology is evolving super fast but speed without trust is just chaos and automation without accountability. We do not put human context is just very efficient but not very effective nonsense. So I I really want to have trust on every layer.

Trust on data, trust in governments, trust in workflows. Well, in maybe trust that we understand that humans are the reasons why we are all building it and yeah we can automate a lot of things but not like human interactions and keep it in mind.

Absolutely that trust at every layer at every piece of you know the AI journey there's a lot to it and I think that your point of having trust on each layer is so important. we have the amazing Tani Rios Castro who will be helping us with Q&A. If you've got a question, raise your hand and she might bring you the mic to make sure that everyone heard like the question is that the younger generation hate technology in the eye and we are not translating enough what is going on on the top layer to everyone else to share the same incentives and sentiments. Right. It sounds like there's a lot of, you know, trying to focus on what does the newer generation think about this because maybe it was sort of pushed on them or they weren't involved in building it. And I've seen it firsthand.

I've spoke, you know, at Stanford, at Deansza Community College and the students there are worried. I talked to recent graduates who are worried. some of you in this room might have a kid who's worried and I'd be really curious to hear what the panel thinks about that.

we have a teenager. She hates everything about AI and she hates she she doesn't want to be an engineer. She wants to be a doctor. Real doctor who can cut and do surgeries something like that. Not without not with robots like real real surgeries. It's a lot of uncertainty and we are all scary if we don't understand something and there's a a lot of uncertainty in the future especially for teenagers because like the their whole life is so uncertain and to be honest the system it's it's not only about how we PR AI it's how the whole system is not prepared you're absolutely right they're going to colleges and colleges provide like giving you outdated information for like very expensive outdated information and after the colleges they cannot find a job. So we have to it's not about like how we PR AI but how we give them opportunities to learn and how we prepare them for the future.

there is a nice initiative from Khan Academy and TAD right now. So they're trying to bring together corporations and colleges and they're making like new higher education for students where they corporations are the ones who providing like their requirements what they need from from the future workforce. So it's something like that we really need we really need like all participants all stakeholders not only like I know engineers who are building AI not only government everyone should think about that the whole system Anastasia do you have something

I just love this question I just love thank you for that this is about almost about storytelling right and it's very hard to tell a good story to the listeners and to predict how they will react if you don't see the whole arc if you're only seeing the bits and bits of it, right? And we are working a lot of with with developers and they are lucky and unlucky enough to be on the forefront of this AI revolution, right? To be among the first ones to be reinvented from the scratch and people tend to be nervous when they do not know what is the end target state and they want to know what is going to be in the end. So are we going to suffer now? Are we going to be rebuilt? And what's next? What's in the end of the day? What's after?

Happy ever after. We struggle a lot right now because of the lack of the human centric narratives overall. And AI labs is doing just great job in progressing with the modal capabilities and stuff like that. But I feel like this is only the beginning of putting people you know in the very center of it so that they know what exactly that all means for them in the very end. So this is why I think that the narrative builders are heavily underestimated because they are not only describing what AI can and cannot do but they are actually defining what AI will be tomorrow.

Thanks for sharing that. I I believe that's really insightful and just looking at, you know, where are we going and being able to draw out that future or envision that future because if you can't envision it, it's pretty hard to get there. Most of the concerns about the future and but to be honest, they use an they live with AI, they use an AI on a daily basis. like I don't know they use AI to to do the homework they communicate with each other they talking to AI friends at snap and understand that this is like the AI friend with whom they can share dramas and all that stuff they hate AI and and at the same time they love AI

well and it's interesting when it's embedded in something they're already using like Snapchat and so I think we're going to see a lot more of that a lot more of AI being embedded in things and maybe you don't even really know you're using it. So, that will be interesting. I hope there's transparency though, so people know when they're using it, but it it'll be more accessible. Maybe not its own app. One last question. Thank you for these wonderful insights about the humanity in the center of AI. I but I want to take you back for a minute for the implementation. How do you make sure that the entire organization sort of moves in the same pace because engineers you know deliver features in a in a super quick way but maybe other departments aren't catching up. So how do you how do you build around like the organizational sort of alignment and pace

when I'm working with enterprises and developers and infrastructure team that's one theme or one question that we see all across is like how do different teams who are building these agents or workflows in silos how do they progress together so that is standardized across their company so I would like to think about this by dividing into four categories, right? One, number one is build. When you have developers, when they're building different tools and I think we touched a little bit about like these can be very fragmented and it's very important to pick and choose the tools and framework that you can standardize across the organization. So building and finding the right tools and right velocity across your system which can be interoperable. There should be flexibility for models which models you want to pick and choose. and having that narrative across the bill build bill state is important.

The second one is scale scaling out. Once we have built the model, I think it's important to have that organizational strategy on how do you deploy them, scale them, manage them like session management, context management. and third thing that Marina talked a lot about and that's one of the most important foundational aspect is governing the technology. So governance is very important and last but not the least is optimizing because these agents are so smart that they try to like optimize themselves by looking at their past behavior. So having a strategy about that also helps kind of ties down all the pillars together to give you this robust strategy across your organization to create your agentic infrastructure and harness around it.

Thank you for that show and being able to connect all of the teams that way on those four different layers she mentioned. That is our time for today. So, please join me in a round of applause for our amazing panelists.