Hello, everybody.My name is, Subhashini Wasan, and I am the global head of delivery at Rakuten Symphony, where I take care of the post sales motion after a product gets built in house. So what am I gonna talk about today? Right? We live in a super hyped AI cycle where model building, model development, agent take, and all the jargons that we keep hearing inside closed rooms of product development have got and are riding the hype cycle. Right? But given my experience of staying outside of the product development and focusing on what really happens on the field, I'm here to tell and I'm here to share my experience and observation from the industry that the hype cycle of product is overrated, and the hype cycle of actual skill, the execution skill that is required for the AI era, which is out there in the field deploying and delivering customer outcomes, is the need of the art.

So we are moving from capability to real outcomes, and this is where the future of the real AI talent is. And I'm here to speak from my experience. Today, if you see, you know, everybody around has asked and is asking, do humans need AI? And rightfully, the technology industry has answered it by investing in billions and billions of dollars in enabling both the infrastructure as well as the products in building AI first or AI led. But we haven't asked one very important question, which is the real question for today's leadership is you have AIized everything. AI is everywhere. But does AI need humans? Or we are going to do away with humans. This is the new technical leadership imperative because as we roll out AI at scale, we just simply cannot miss the stakes that are extremely high.

But before that, let me give a background on the context which was with which I'm talking. Right? We at Rakuten and my background has always been large scale rollout. Today, I mean, I am at the forefront rolling out very large scale network and connectivity telecommunications globally. It is not just one country, one state. It is a global rollout. And what I'm seeing here is that more and more, we are going and navigating to big things, big data centers, I know, which are running massive training and inferencing workloads along with usual workloads. The clusters are changing. GPU is super powering them. The networking is changing. We are no longer Ethernet. We are moving more towards, RDMA, which is remote direct memory access sorry, over Ethernet as well as InfiniBand fabrics.

We are no longer raising user request or, you know, the traffic in terms of tens and thousands. We are moving into millions of queries. And complexity at that level of scale in a real world deployment beats any context or the visibility a single agent that was trained inside the closed doors of a of a cluster can actually have. What we are seeing over and over and over is that the agent acknowledge is siloed. And when it comes to multiple forces like what we are seeing at scale coming together, there is an absolute context collapse, and AI gets it wrong more than what we know. And the bad thing about the way the outcomes or the results that AI produces when the the when the environment is not actually ideal and close to how it was trained, The problem here is AI gives you results with confidence. Results or, you know, findings show that AI can bluff with 35% confidence factor. It speaks with confidence. It hallucinates with confidence. It's not a bug. It is structural.

That is how AI is being designed. And we human beings have a huge challenge in the sense that we always mistake confidence for credibility, which is fatal, which is wrong because at scale, in critical mission workloads, which is impacting real user lives, errors are business crisis. A Facebook going down, a telecommunications, AT and T network going down, or any such order of collapse is a business downward spiral. All the GTM conversations, the strategy, the product development that is going on in a c suite will come to a halt if the business reputation goes for a toss by AI hallucinating on real world wrong decisions when things go wrong as they will. So given the context that I have established that at scale, AI is not dependable and it doesn't have a single AI agents trained in silos, do not have the knowledge, let's look at what skills will actually stick. I'm calling out based on my nearly two plus and more than years of experience for a category creation of skill set, which is called the AI first executionist.

You can think of this skill set or the person as the orchestrator of orchestrators. A new archetype of technical leadership in the era of stochastic systems, which is extremely probabilistic. There is no determinism anymore. Everything is a probability. So we need a new kind of human in the loop. What is this human? Who is this? First, the person thinks in systems. The person understands that in real world, multiple complex systems come together, excuse me, in delivering real world output. So it is not task led anymore. So we need people with systemic thinking. Right? People understand how different agents and humans work together in bringing products to scale. The person hunts for problems, not solutions. Charlie Munger is very famous for saying that what leaders need is inversion thinking. What will go wrong, not what will fly high.

So we need people that can actually go with the assumption that the system, the product is built by AI. The workflow is agentic. The decisions are by AI, and so they are looking for problems, not conviction. The third one speaks AI and business very fluently, understands the technology part of AI and understands the business or the real world, impact of large systems and how they can go wrong in real life. And fourth, absolutely knows where AI hallucinates. In spite of all the data points leading to taking a particular action, this person is able to find out that this is actually a hallucination call. It's not a real call.

So you could think of this person as somebody that can orchestrate, AI agents and humans and understand the complexities involved and still can make realistic judgments. This is the future talent that will matter in the future, not prompt engineers, not data scientists, because this is today a redundant skill. The more and more ins I go and sit with product teams, nearly 80% of their work is done by cloud or similar. Right? Prompt prompting writing good prompts, processing data, they are good, but they are not inevitable. They will be replaced. What will not be replaced is a person with multidisciplinary systems against AI agents which are trained in silos for specific problems, people that can confidently own real world operations, taking products to the field and, you know, steering a new kind of observability for the AI era and owning business outcomes in terms of confidently telling that, yes, I will use AI first principle, but still I will use my judgment and my knowledge and systemic thinking before I press the go button.

And finally, the absolute straw soft skill of strong intuitive judgment, which cannot be just earned in one day, but it requires years of technical knowledge and also soft skills in terms of being in the business or general management. And the era of the generalist who understands multidisciplinary systems combined with soft skills and understands AI first principles is here. So AI does not need better models. It needs, better humans around it. So that is the case that I'm trying to make here, that the skill that we need to really foster in our, companies, in our culture, and in ourselves or people that can actually own real world outcomes at scale.