Venture Europe: Entrepreneurship | Technology | Venture Capital | Eu/Acc
00:22:25 · Casper Henningsen, Calin Fabri
[00:22:25] S0: also communities as LLMS has in certain areas of the internet. The problem with it is there's no context, there's no real human depth. In the best examples I've been looking into, approximately 3% of the stack had what I consider qualitative understanding and context. What you end up with now is very interesting. Back in the days, it was like a research gap or a research liability. That turns into, when running the agenda enterprise, a commercial existential risk. Decisions are going to speed up, it's going to accelerate. And if your stack of data are not representative and contextual relevant for the markets you're building towards, you're going to end up in a very hard situation. So exactly what you're saying. I'm imagining a world of a human layer. I have that on my LinkedIn. It's between two to ten years in front of its time. But trust me, it will come now. Every model that is born and built to serve a commercial application needs a layer of human understanding as a competitive vehicle for that application's performance. That is my belief. [00:23:47] S1: Absolutely. And where do you see the synthetic data? Synthetic
00:23:47 · Calin Fabri, Casper Henningsen
[00:23:47] S1: Absolutely. And where do you see the synthetic data? Synthetic [00:23:53] S0: is a huge opportunity. I'm still very cautious of how to apply but I think synthetic is to be front end in a lot of the agents that we're going to feed into the enterprise from understanding. The difference though that the front end and the synthetic layer, because you need that to actually interact with the models, needs to be built on grounded interviews and real truth data. So imagine this idea of we run AI moderated as I said before real human beings being interviewed for a long time really on layering a lot of different topics to build replicates on that as a digital twin or a synthetic simulation will be the future as the front end of the application but you still need verified human to sort of like be vibrant and constantly updated in the back end of that model. If that makes sense. [00:24:49] S1: Absolutely. Now, I've watched yesterday like the latest All In podcast and like they were talking about composable models. I almost see that also here you will have composable models where, let's say at the forefront, you have the synthetic data which can take and answer some questions, but in the back there always need to be humans. Maybe you do have some data with a certain degree of certainty on some specific questions, but then in the back you would always have, you know, the humans that kind of back everything up.
00:24:49 · Calin Fabri, Casper Henningsen
[00:24:49] S1: Absolutely. Now, I've watched yesterday like the latest All In podcast and like they were talking about composable models. I almost see that also here you will have composable models where, let's say at the forefront, you have the synthetic data which can take and answer some questions, but in the back there always need to be humans. Maybe you do have some data with a certain degree of certainty on some specific questions, but then in the back you would always have, you know, the humans that kind of back everything up. [00:25:18] S0: A 100%. When we talk about synthetic as a front end and as a modality that starts to engage with the agentic enterprise, it needs to do two things. Today, have a very reactive way of looking at models, which I'm prompting and it brings back something. So like all modalities across is kind of like still the majority is like that. The modality that I think we will be leaning ourselves into are autonomous loops between models. The true agentic modality and here you would find that writing code would interact with customer needs autonomously. Creative development will run autonomously with understanding of consumers. And when we run simulations in pilots now in our models, you can easily run thousands of loops over durations of hours. And you can follow kind of like the code or you can follow the creative just completely being reshaped by the amount of loops you're running.
00:25:18 · Casper Henningsen, Calin Fabri
[00:25:18] S0: A 100%. When we talk about synthetic as a front end and as a modality that starts to engage with the agentic enterprise, it needs to do two things. Today, have a very reactive way of looking at models, which I'm prompting and it brings back something. So like all modalities across is kind of like still the majority is like that. The modality that I think we will be leaning ourselves into are autonomous loops between models. The true agentic modality and here you would find that writing code would interact with customer needs autonomously. Creative development will run autonomously with understanding of consumers. And when we run simulations in pilots now in our models, you can easily run thousands of loops over durations of hours. And you can follow kind of like the code or you can follow the creative just completely being reshaped by the amount of loops you're running. [00:26:25] S1: Just to just to make sure, sure that I understand you have several autonomous loops that gets the data of what your client would want to and then can go back and already start to build the product. So for example, when developer or the engineer going to be
00:26:25 · Calin Fabri, Casper Henningsen
[00:26:25] S1: Just to just to make sure, sure that I understand you have several autonomous loops that gets the data of what your client would want to and then can go back and already start to build the product. So for example, when developer or the engineer going to be [00:26:43] S0: in Agendaq. Yes. Fully. That is set out to build essentially code. Yes. The job now going to be how good code can I write to capture my audiences or my customers needs? Exactly. That world going to be autonomous for a long period of that loop. The agent of coding will be guided by the needs of the consumer. The creative agent within the enterprise needs to run loops to meet a certain threshold of quality before human judgments come in and said, you know what, let's go or let's make it, know, edit it. That is the world where I think we lift the quality of all our application to a new level, where I see it very, very difficult for enterprises that are not on that train to even be near competitive modes when this starts to happen. [00:27:37] S1: Of course, because it's automatic. It's automatic on the consumer insights of what your client wants and just goes automatically. And if you can also then do multiple type of simulations, then you kind of bridge the variant, let's say of getting it wrong.
00:27:37 · Calin Fabri, Casper Henningsen
[00:27:37] S1: Of course, because it's automatic. It's automatic on the consumer insights of what your client wants and just goes automatically. And if you can also then do multiple type of simulations, then you kind of bridge the variant, let's say of getting it wrong. [00:27:55] S0: That's one and the other thing is we're putting humans in the moment where they're best on judgment, creativity, strategic alignment. And that is a world where I think value creation going to have its next leap. So I'm very optimistic about elevation, human judgment, human involvement. But I'm so bullish on the agenda enterprise and I'm obsessed with how that can be guided and fueled [00:28:24] S1: by true understanding. Absolutely. And I think it's also beautiful that, you know, you're obsessed about the world of agentic enterprises, but at the same time, one of the core missions is to understand empathy. [00:28:40] S0: That's actually totally right. And if you extrapolate that thinking, what else is there? Not much. There's not like empathy. Then, yes. Why is Why company that is famous for saying you know build great stuff and talk to users. Yeah. If build stuff is going to be easier you are sitting with the one only source of true differentiation which is how good are you actually to understand your target markets, segments, your customers and convert that into experiences that drive the likeness and uptake of your margins and your revenue. We are there. How do [00:29:17] S1: you ensure then the quality of that part? That is of course where we use the majority of time. What are