Agentic Engineering & OpenClaw: Why Software Development Becomes a Matter of Taste

In the latest episode of the Lex Fridman Podcast, Peter Steinberger (founder of PSPDFKit) shares fascinating insights into the future of software development. At the center is a concept that radically changes how we produce code: Agentic Engineering.

With his open-source project OpenClaw, Steinberger demonstrates how autonomous AI agents generate not just code snippets, but implement whole features, fix errors on their own, and validate tests. For us as a software company, this raises an exciting question: What remains for the human when the agent gets the work done?

The Paradigm Shift: From Coder to Architect

Traditional software development consists largely of so-called “plumbing” — the necessary but often repetitive work of moving data from A to B, connecting interfaces, and writing boilerplate code.

Steinberger’s approach to Agentic Engineering aims to free us from this mechanical layer. That does not mean code quality or syntax stop mattering. On the contrary, the requirements rise, but the developer role shifts.

Instead of typing every syntax line by hand, the modern developer acts as orchestrator and architect. The focus is on two human core competencies:

1. System architecture: Understanding how complex components work together in a scalable and secure way.

2. Taste: A term also used by Andrej Karpathy (OpenAI/Tesla) in this context. Taste is the ability to intuitively distinguish between a working solution and an elegant, maintainable one.

OpenClaw: The Agent in the Sandbox

OpenClaw is Steinberger’s answer to how far this autonomy can go. The tool acts as an “agentic engineer” working in a safe environment (a Docker container). The process is fascinating:

• The agent analyses a task (Issue).
• It writes the necessary code.
• It runs the build process and tests.
• The kicker: If a test fails, the agent reads the failure message, “thinks” it through, corrects its own code, and tries again — until the tests are green (“Green Loop”).

This enables a way of working that Steinberger and Karpathy call “Vibe Coding”: the human sets the direction and quality standards, while the agent handles iterative execution.

Conclusion: Efficiency Through Abstraction

For us at P-CATION, this trend does not mean the end of classical engineering, but its next evolutionary stage.

When we use agents like OpenClaw to automate “plumbing,” we create space for what matters: solving real business problems and building ambitious product visions. Code is the foundation, but our taste and architectural expertise are what ultimately create software value.