Konstantin Marunchenko Is Building Enterprise AI Where Discovery Matters More Than Code

The 21-year-old founding engineer argues that the next advantage in software will come from understanding customers deeply, not simply generating more code

The easy prediction is that AI labs will eventually build all the software anyone needs. More models, more coding agents, more output. Konstantin Marunchenko does not think the future will be that simple. At 21, the founding engineer has been building enterprise-scale AI pipelines at Echelon, where the hard part is rarely producing code for its own sake. The harder part is discovering what a company actually needs, finding the useful signals inside messy business data, and turning that knowledge into a system people can trust.

“People think the bottleneck is writing code faster,” Konstantin says. “In many products, the harder part is discovering the right problem.”

That view runs against a popular story in technology right now. If AI can generate code, the argument goes, then software creation becomes a matter of scaling production. A lab can build models, connect them to coding systems, and produce what the market wants. Konstantin sees the appeal of that idea. He also sees where it breaks.

A company does not always know how to describe its own problem. Users do not always say what they mean. Business processes are full of exceptions, habits, politics, old tools, silent workarounds, and data that only makes sense after someone asks the right questions. A system can generate code quickly and still miss the reason the product should exist.

“A model can write a feature,” he says. “It does not automatically understand why that feature matters to a customer.”

Konstantin’s perspective comes from working close to enterprise problems. At Echelon, he joined as engineer number four and helped build core AI pipelines for a startup that grew from him and the founders to a team of 12, with contracts across multiple enterprises. He led several deployments that received strong customer feedback and helped the company expand into larger contracts.

The work involves more than adding AI to a workflow. Konstantin has built pipelines designed to discover savings worth millions of dollars for large enterprises. Those systems involve document processing, ERP discovery, signals from employee calendar and email data, and the technical wiring needed to connect those inputs into something useful.

That is where the “AI will build everything” argument starts to look thin to him. The code matters, but it is only one layer of the work. The larger challenge is understanding the organization well enough to know what should be built in the first place.“

Enterprise software lives inside the company’s reality,” Konstantin says. “If you do not understand that reality, the product can look smart and still be wrong.”

He has seen a version of the same problem from the consumer side. Before Echelon, Konstantin founded and operated a mobile gaming company at 15, running it for four years with employees and revenue. That experience taught him how unforgiving product discovery can be. In mobile gaming, he saw success rates around one in 60 to one in 100 in testing. The issue was not that teams could not build. The issue was that nobody fully knew, in advance, what customers would want.

That lesson stayed with him. Consumer markets and enterprise markets look different, but both punish guesses. A product can be technically impressive and still fail because it solves the wrong problem, reaches the wrong user, or misses the emotional and practical details that shape adoption.“

In gaming, you learn very quickly that being able to build something does not mean people will care,” he says. “That changed how I think about software.”

He believes foundation models will become the basis for many AI products. His disagreement is with the idea that large labs can replace the entire process of discovering, shaping, and delivering software for every specific market. Models may write code. They may automate parts of development. They may make small teams far more productive. Yet the work of understanding a business problem still requires contact with people.

Customers often explain the surface issue first. The deeper problem may appear later, after conversations, observation, and repeated testing. In enterprise settings, people are also more likely to discuss sensitive business challenges with other humans than with an AI system. That matters because the most valuable software often begins with trust.

“People talk differently when they believe another person is trying to understand their business,” Konstantin says. “That conversation is part of the product.”

This is one reason he believes good product people and engineers will remain important. The role may change as AI writes more code, but the center of the work will move toward discovery, judgment, and translation. Someone still has to find the real problem, invent the right solution, and turn it into a product that fits the user’s world.

Konstantin’s own path prepared him for that kind of work in an unusual way. He started programming at 8, completed a course with adults at 12, got his first job at 14, and skipped college because he was already working through high school. That early start gave him technical speed, but startup work gave him something different: the habit of building under uncertainty.

He also learned that software engineering is more social than many people assume. Good engineers do not only sit alone and write code. They ask questions. They talk to peers. They listen to customers. They find the hidden assumptions inside a request before turning it into a system.“

Engineering is less isolated than people think,” Konstantin says. “The code is only good if you understood the problem well enough.”

That belief became more important as AI made programming more powerful. Konstantin has always been drawn to the feeling of writing text and watching something work because of it. With AI, that feeling has expanded. Programs can now handle more ambiguous tasks, reason across more information, and participate in work that used to require narrower automation. To him, that makes discovery more important, not less.

As software generation becomes easier, building the wrong thing may become easier too.

“The companies that win will not only have better code,” he says. “They will know more about their users.”

Konstantin’s work at Echelon is one example of that shift. Enterprise AI is not only a question of models or pipelines. It is a question of whether a system can uncover value inside a company’s real operations and make that value usable. Documents, ERP data, calendars, emails, and business context all have to be interpreted carefully. The product has to fit into the way people already work while showing them something they could not easily see before.

That is the future of software Konstantin is betting on. AI may change how much code a team can produce, but it will not remove the need for builders who can ask better questions, read the room, and discover where the real opportunity sits.

For Konstantin Marunchenko, the next era of software will not be won by the teams that simply generate the most. It will be won by the ones that understand the most, then build with that understanding.