In the fast-paced world of artificial intelligence, new figures emerge who don’t just iterate but fundamentally rethink the building blocks of the technology. Boris Kriuk is one such figure. As the Chief Technology Officer at Strevio, a company pushing the boundaries of agentic AI, and the head of his own research group, Boris Kriuk Labs, he is at the forefront of innovation. We sat down with him to discuss his work on a new language designed exclusively for AI.
Michelle Kellet: Turning 21 this year, you are already considered one of the key AI figures in Hong Kong. Do you agree with this statement?
Boris Kriuk: Honestly, I don’t really think about titles like that. My focus is entirely on my responsibilities to my teams at Strevio and the lab. My primary goal is to ensure we are building technology that is not only groundbreaking but also delivers tangible value. There are complex challenges to solve and ambitious goals to meet, and that’s where my energy goes. The external labels are a distraction from the actual work that needs to get done.
Michelle Kellet: Speaking of that work, your lab has been developing something called Compressed Vector Language, or CVL. In simple terms, what problem are you trying to solve?
Kriuk: The problem is a crisis of verbosity. Today’s AI agents are incredibly sophisticated in how they “think.” They can form rich, complex, nuanced concepts internally. But when they need to communicate with another agent or even remember a thought for later, they are forced to do something incredibly inefficient: describe that rich concept using human language.
Imagine having a beautiful, detailed painting in your mind, but the only way to share it is to write a ten-page essay describing it. The person who reads it then has to reconstruct the painting in their own mind from your words. It’s slow, expensive, and a huge amount of detail is lost in translation. That’s how AI agents talk today. CVL is our solution to let them send the painting itself.
Michelle Kellet: That’s a powerful analogy. So what makes CVL a “language” rather than just a more efficient data transfer method?
Kriuk: This is the core of its novelty. A language is more than just data; it contains intent and context. CVL has this built-in through a concept we call “pragmatics-through-compression.” It’s not just about what an agent says, but how it says it.
The real breakthrough is that we’ve created a direct link between an agent’s internal state and its external communication. Current systems are stateless in their communication; every message is a standalone piece of text. With CVL, the message itself carries information about the sender’s cognitive state. It’s the difference between sending a text message that says “I’m stressed” and having your voice crack with genuine stress over the phone. One is a description; the other is a direct transmission of feeling. CVL allows agents to transmit their “feeling” or “certainty” about a concept directly.
Michelle Kellet: Could you elaborate on how that works? How does an AI signal intent?
Kriuk: It comes down to a trade-off between detail and size. Think about human communication. The phrase “I need your help” can mean very different things. Said calmly, it’s a standard request. Shouted urgently, it’s a critical emergency. The tone of voice carries vital meta-information.
CVL gives AI agents that “tone of voice.” By choosing how much to simplify the core concept before sending it, an agent can signal its intent. A highly detailed, precise message signals “this is critical, analyze every detail.” A highly simplified, compact message signals “this is a low-priority broadcast, just get the gist.” This allows agents to communicate importance and priority implicitly, making their interactions far more sophisticated and efficient. It’s a language where the very structure of a message conveys as much meaning as its content.
Michelle Kellet: What are the long-term implications of a technology like this? Where do you see it being used?
Kriuk: The implications are foundational. In the short term, it will make AI assistants and agentic systems much faster and cheaper to run. But in the long term, CVL is a blueprint for a new operating system for AI cognition. It could become the standard protocol for massive swarms of autonomous drones or IoT devices to coordinate in real-time. It could enable a personal AI on your phone to perform complex reasoning without constantly connecting to the cloud. We’re not just building a better way for AIs to talk; we’re building the infrastructure for a future where they can think and collaborate at a scale we can’t yet imagine.
