.png){kind=logo}
The founding engineer sees the next frontier in systems that let people shape matter with the same flexibility coding agents bring to software
The first time a program works, it can feel almost unreal. A few lines of text become something that moves, calculates, responds, or changes what a person can do. Konstantin Marunchenko remembers that feeling from childhood, and it still shapes how he thinks about technology. At 21, the founding engineer has already built enterprise AI pipelines at Echelon, but his longer-term interest reaches beyond software alone. He wants to explore what happens when AI begins to act more directly on the physical world.
“Programming always felt like turning thought into something that could act,” Konstantin says. “That was fascinating to me as a child, and AI makes the idea much bigger.”
Konstantin started programming at 8, completed a course with adults at 12, got his first job at 14, and founded a mobile gaming company at 15. That early path gave him a direct relationship with software. He did not treat code as an academic subject first. He experienced it as a way to make ideas operational.
That fascination has changed with AI. Traditional software could automate tasks, move information, and create tools. Newer AI systems can work through more ambiguous instructions. Coding agents can help people write, refactor, explore, and build software in a way that feels less like operating a tool and more like collaborating with a system that understands intent.
Konstantin is interested in what that shift may mean outside the screen.
“Coding agents changed how software feels,” he says. “I am interested in what happens when that kind of interaction reaches the physical world.”
That is the future he wants to build toward: a company at the intersection of frontier AI and the physical world. His view is that the current approach of training highly specific models for physical-world tasks may not be enough. A model trained narrowly for one task can be useful, but the physical world is not narrow. It is messy, varied, and full of conditions that change. Konstantin believes large models and coding agents may eventually offer a more flexible path.
His phrase for the idea is simple but ambitious: people should be able to “edit” matter the way they can now use coding agents to edit software.
That does not mean physical reality becomes easy. It means the interaction model changes. Instead of needing specialized systems for every task, people could work with AI to plan, adjust, test, and control physical processes with more flexibility. In that future, building could become less dependent on rigid tooling and more responsive to human intent.
“The future I care about is not only software becoming faster,” Konstantin says. “It is AI helping people change real systems with more precision.”
His work at Echelon gives him a practical view of what that kind of future would require. As engineer number four, Konstantin helped build core AI pipelines and led the onboarding of several enterprises. The company grew from him and the founders to a 12-person team with multiple enterprise contracts. His deployments produced strong customer feedback and helped lead to larger contracts.
The systems he built were not simple automations. They involved messy inputs from real businesses: documents, ERP discovery, employee calendar and email signals, and pipelines that connected those sources to identify savings opportunities worth millions of dollars for large enterprises. That work taught him how difficult it is to make AI useful when the environment is not clean.
“Real systems do not arrive in perfect format,” Konstantin says. “You have to discover the structure inside them before you can make anything useful happen.”
That lesson transfers to his physical-world ambitions. The physical world will not behave like a controlled dataset. A machine, building site, logistics system, factory floor, or resource network contains constraints that are partly technical and partly human. People use systems in unexpected ways. Information is incomplete. Conditions shift. The AI has to handle ambiguity without pretending the ambiguity is gone.
Konstantin’s earlier work in mobile gaming adds another layer to that thinking. He founded and operated a gaming company for four years, with employees and revenue. In consumer products, he learned that building something does not prove people will use it. Teams can test dozens of ideas before one works. He saw how difficult it is to predict what users actually want, even when the product is technically possible.
That makes him cautious about overconfident predictions. He does not believe AI progress removes the need for discovery. In fact, the more powerful the tools become, the more important it is to understand what problem deserves to be solved.
“AI can increase what is possible,” he says. “That makes judgment more important, not less.”
His path has made him cautious about big technical promises. Konstantin left Ukraine when the war began, skipped college because he was already working, and learned in startups that systems are only valuable when they survive real use. That lesson matters even more when AI begins moving from software into the physical world.
Software mistakes can be serious. Physical-world mistakes can be more direct. Systems that touch materials, resources, machines, or environments need to be built with care. The promise of AI moving into the physical world cannot only be convenience. It has to include reliability.
“Working under pressure taught me that an idea is not enough,” Konstantin says. “The system has to survive contact with real use.”
For now, his day-to-day work remains in enterprise AI. He is building pipelines that help companies understand their own operations and uncover value inside data they already have. But the larger direction is visible. He sees AI moving from assistance to agency, from software generation to systems that can take on more complex work, and eventually from digital environments into the material world.
That vision sounds far ahead, but Konstantin does not approach it as science fiction. He approaches it like an engineer who has been watching one boundary after another move. First, text made programs. Then AI made programs more flexible. The next question is whether similar flexibility can reach systems made of machines, materials, and resources.
For Konstantin Marunchenko, the dream is not technology for spectacle. It is a future where people can shape physical systems with more clarity and less waste. Code taught him that text could become action. AI taught him that action could become more adaptive. The next step, he believes, is finding out how far that idea can go beyond the screen.
