When Domain Experts Start Building

In early 2025 we met a consultant who had just left the world of large energy enterprises to start his own business.

He brought deep expertise in decarbonization and energy transition strategy, understanding how grid limits, volatile energy prices, subsidies, and emission targets interact in practice. His consulting work centered around techno-economic simulations comparing pathways toward net zero. Modeling complex system, calculating emissions, return on investment and sensitivities. All of it lived in Excel in huge files. It worked for years until the growing system complexity turned Excel into a bottleneck.

Curious about recent AI progress, he began experimenting. He described his models and assumptions in conversations with ChatGPT and gradually translated them into code. Scripts became structured simulation modules. Those modules evolved into a small web platform where clients could configure scenarios and explore transition pathways interactively.

He was not a trained software engineer. But he had absolute clarity about the problem space. He knew which variables mattered and which trade-offs were realistic. That precision allowed him to build something far more powerful than another spreadsheet.

As frontier models improved, so did the quality of his code. He learned how to describe requirements more precisely. The system became more stable, and it simply worked. Still, he was unsure about the invisible layers like Security, Performance and the Deployment. All the parts that turn a working prototype into something you can confidently put in front of customers.

That is when he approached us.

We did not replace his work, but we refined it together. We strengthened the architecture and introduced proper deployment pipelines. We reviewed critical parts and improved the structure and boundaries. Of course we also approached this in an AI supported way.

One year later the situation looks completely different. He can formulate new features clearly and break them down into structured tasks. Using modern coding environments like Codex or Claude Code he implements large parts himself. Our role has shifted toward architectural guidance and long term thinking.

Agents as a New Interface

Recently, we met another consultant who also wants to build a product around his expertise.

This time we chose a different starting point. Instead of letting him experiment in isolation, we provided a full coding platform based on PI and a clean application foundation. A structured environment for agent collaboration with clear conventions and clear architectural guardrails.

With this agentic coding platform the workflow feels fundamentally different. Day by day we watch his product take shape. He formulates features in clear language and the agents translate vision into working software. Again, we support on architectural decisions and system design where experience still matters most. This remains an important process, but the real core lies elsewhere. It lies in combining ideas, domain expertise and practical know-how with the individual taste and vision of the person behind the product, and turning that combination into a working system.

The crucial shift is not that agents write code. The shift is that domain experts can directly shape software.

This does not diminish software engineering. It changes its center of gravity.

Why I Am Not Afraid

I want to end this on a personal note.

I am a software developer out of passion. I love watching something grow step by step. Writing code, refactoring it and seeing a system respond for the first time. That feeling started in school and followed me through employment and self-employment.

To be honest, I am not particularly skilled with physical tools, so software became my way of building things. My very personal way of creating.

Over the past month, my emotions have been mixed. When models improve this fast, you inevitably ask yourself what your role will be. Whether the craft you invested decades in is slowly dissolving or not.

But the more I work with agentic tools and our own agent framework, the clearer it becomes.

I am still building. I still make architectural decisions. I still define taste and direction. The difference is that I am amplified. The agent accelerates execution and it explores variations faster than I ever could. I remain at the steering wheel.

It is a different form of programming, but it is not an inferior one.

I am convinced that in the next one to three months, we will see another major leap in model capabilities. Security reasoning will improve. Performance considerations will become more integrated. Architectural guidance will grow stronger. Many of the current weaknesses will shrink.

That does not frighten me. => It energizes me.

Because what I truly love was never typing syntax. It was turning ideas into reality and that remains unchanged!