Van Coder naar Architect/Builder

From "Coder" to "Conductor": How AI is Rewriting Software Development

Imagine sitting at your desk, but instead of staring at a blinking cursor and manually typing out lines of code, you feel more like an orchestra conductor. You aren't writing the notes anymore; you are guiding an entire crew of AI agents.

That is precisely the shift software creator Peter Steinberger highlights in his work. Rather than casual "vibe coding," he calls his approach agentic engineering. He barely writes code by hand anymore; instead, he designs systems, establishes the core architecture, and lets AI agents handle the boring plumbing work.

From Pull Requests to Prompt Requests

The most radical part of his workflow? He often doesn't even read the code he ships. While traditional software engineering relies heavily on meticulous code reviews and pull requests, Steinberger views contributions now as prompt requests. The real value lies in the underlying intent and how effectively you prompt the AI, not the raw text generated. As long as a robust feedback loop—where the system automatically runs tests and catches errors—is in place, the output becomes reliable.

The "Black Eye Club"

Working this way is intensely addictive. Steinberger talks about the "Black Eye Club"—developers staying up until the early morning hours because prompting AI feels like playing a slot machine. The thrill of entering a prompt and getting an ingenious solution back is hard to put down. However, it is also mentally exhausting, requiring you to constantly context-switch while managing five to ten active agents in parallel.

A Word of Caution

Before you throw away your IDE, Steinberger offers a reality check. There is a steep learning curve to mastering the "language of the machine," and many experienced developers burn out early because they expect instant magic without learning how to steer an agent properly. Furthermore, building a side project is entirely different from an enterprise environment; large companies need a massive cultural and structural refactor before AI can truly scale efficiently.

Ultimately, tomorrow's developer looks less like a manual bricklayer and more like an architect—someone who commands an army of digital builders with a single prompt.

Source: The creator of OpenClaw: "I ship code I don't read"

Original video: https://youtu.be/8lF7HmQ_RgY?si=Oc5ohlXWv9MGBYp0