Egor Kraev

Post-Peak Agent Hype and Domain-Centric Intelligence

At the Panoramai AI Summit, Egor Kraev, co-founder of Motley AI and former Wise machine learning leader, delivered a contrarian analysis of agent framework evolution. Drawing on his extensive experience building ML teams from zero to 30+ people and his self-described career "since the last millennium," he advocated for domain-specific solutions over generic agent capabilities while challenging industry hype cycles.

Peak Agent Hype Assessment Kraev's provocative opening challenged prevailing industry enthusiasm: "what I think is happening is that we are very near peak agent hype and quite possibly past it." His analysis deconstructed agent marketing inflation: "Really anything that's any kind of piece of software or automated workflow that happens to call a large language model somewhere inside it can be marketed as an agent. It often is."

Technical Agent Reality He provided precise technical definitions often obscured by marketing: "If you get to be slightly more technical, then maybe it's something that calls an LLM and then has a loop inside it. So the LLM does ask the outside world to do something, the results back into the LLM. And so you have this kind of a feedback loop, and that's called an agent. And really, that's all there is to it."

Value Source Analysis Kraev's framework identified where real competitive advantage originates: "So if you look at what value the different agentic wrappers, because agentic frameworks are a dime a dozen these days, I'm sure you could rattle down a dozen names. You know, anything from line graph, crew AI. Everybody has the agent framework these days. But when you think about the value they add, really it's not that big. So the real power comes firstly from the large language model that you use. And that's a very competitive space, thankfully. And secondly, from the tools that you use."

Domain-Centric Evolution His strategic thesis positioned domain expertise as the new differentiator: "And for me the change that the differentiator now is going from being tech-centric to being tech-centric and domain-centric." He elaborated: "I think where we're going to is more these domain-centric agents, where you start with cognitive biases. So for example, when you hire a human to do marketing, You don't have to explain to them from scratch, if they have prior experience, what marketing works, or what marketing is, how it works."

Storytelling vs. Question-Answering Philosophy Kraev's most compelling insight drew from science fiction to illustrate AI limitations: "Because there is this wonderful story that I encourage you to read. It's actually open source by now as well, in public domain, by Robert Sheckley, called Ask a Foolish Question. And the story is about this all-knowing being sitting somewhere on an asteroid that literally knows everything, but it has a constraint. It can only give exact answers to precise questions." The lesson: "Because to ask the right question, you must already know most of the answer. And so that's what we're trying to do at Motley AI, is to go from question answering to storytelling."

Pragmatic Technology Philosophy His implementation guidance emphasized ruthless simplicity: "In fact, what I keep saying is you should always use the dumbest thing that works. If you can get things done without LLM, don't use it. If you can do without multi-agent, don't use multi-agent. But sometimes you really need them, and then you use them."

MCP Enterprise vs. Consumer Analysis Kraev's assessment of Model Control Protocol revealed important enterprise limitations: "At the same time, I think MCP will matter a lot more for the consumer space than for the enterprise space. Because for the consumer space, there's much more demand for things that can do anything, or the way you have to discover your tools as you go. Whereas in the enterprise, generally, you want your agent or your workflow to do, you want to allow it to do certain things, and you want to constrain exactly, constrain as much as you want what it can do." He identified the core benefit: "And for me, the biggest benefit of MCP is tool discovery, and that's just less relevant in an enterprise context."

Dynamic Registration Critical Barrier His technical insight revealed a crucial MCP limitation: "Just in case you think of providing a tool via MCP, please, please make sure that dynamic registration is supported. Because right now, the one thing crippling MCP adoption is that to do dynamic authentication, like I just need to click a button to allow my agent to access this tool. Each tool needs to pre-register with authentication provided by OAuth. And the MCP standard doesn't require that. It's just a decision."

Organizational Orchestration Vision Kraev's perspective on enterprise automation leveraged existing infrastructure: "Yes, so I actually think that the answer to that is pretty simple, which is that one of the biggest use cases for large language models in enterprise is just as glue. So converting things, inputs from one shape into another, or extracting fixed shape data from squiggly and wishy-washy data. And what this has done, this ability, is it's just done composing things much easier. And that's just supercharged existing automation tools, like Zapier and so on."

Key Takeaway: Kraev positioned domain-centric agents with embedded cognitive biases as the next evolution beyond generic frameworks, emphasizing that sustainable competitive advantage comes from specialized intelligence rather than technical sophistication.


With a foundation in the Russian mathematical tradition, refined at at ETH Zurich and the University of Maryland, my journey in machine learning spans decades, crossing from economic and human development analysis to advanced AI applications for business. As the Head of AI at Wise, I've had the privilege of molding its data science landscape, in particular in fraud detection, trading algorithms, causal inference for marketing, and transformative LLM applications throughout the company. My leadership has seen the Wise Data Science team grow from an idea to a powerhouse of over 30 skilled professionals. My passion for AI is not confined to work context; it extends for example into experiments in reinforcement learning for molecular optimization, and generative AI for educational purposes. Currently, my focus is on multi-agent LLM approaches, pushing the envelope for practical applications of GenAI. I am eager to connect with VCs seeking expert guidance, AI startup founders in search of a seasoned collaborator, conference organizers looking for transformative insights, and brilliant minds keen to join me in forging the future of AI. If you're at the forefront of innovation and looking for a partner in navigating the ever-changing landscape of artificial intelligence, let's explore how we can make cool stuff happen together.