Coding as Craft Is Dying, AI‑Driven Building Is Booming – OpenClaw Founder’s View

Coding as Craft Is Dying, AI‑Driven Building Is Booming

On February 11, 2026, Silicon Valley entrepreneur Brian Norgard tweeted that many tech workers feel extreme anxiety, and xAI co‑founder Jimmy Ba announced that we are entering an era where the right tools can deliver a hundred‑fold productivity boost. The same day, Austrian developer Peter Steinberger, while staying in an Airbnb in San Francisco, built the prototype of OpenClaw – the fastest‑growing open‑source project on GitHub with 189 k stars and two million weekly visitors – in just one hour using Claude Code.

The death of the coding craft : Steinberger says the deep‑flow state of writing elegant code for hours is becoming as rare as hand‑weaving, persisting only as a hobby. He likens the shift to the steam‑engine revolution that replaced manual labor. METR data shows AI‑completed tasks growing exponentially: from ten minutes a year ago to nearly five hours today, doubling every four to seven months. Matt Shumer notes GPT‑5.3 Codex exhibits a “judgment” or “taste” previously thought impossible for AI.

“Programming as a uniquely human cognitive activity is losing its scarcity, and the absurdly high salaries it supported are ending.”

The explosion of building ability : In November 2025, Steinberger built an OpenClaw prototype that forwards WhatsApp messages to Claude Code and back, adding image support within hours. Three months later the project grew to 300 k lines of code, supporting most major chat platforms. Andrej Karpathy called it “the closest thing to sci‑fi take‑off I’ve seen.” Both Meta and OpenAI offered acquisition, and Zuckerberg spent a week experimenting with it.

Steinberger alone made 6 600 commits in January 2026, running four to ten agents simultaneously—some for major features, others for bug fixes and documentation generated by the agents. He interacts with agents via voice, leaving the keyboard for terminal commands, and claims his hands are too valuable for typing.

IBM research scientist Kaoutar El Maghraoui confirms that OpenClaw disproves the assumption that autonomous AI agents require vertical integration by large companies; a loosely‑governed open‑source project with sufficient system permissions can be “extremely powerful.”

Agentic engineering vs. “vibe coding” : Steinberger rejects the term “vibe coding” as an insult and prefers “agentic engineering.” He describes a U‑shaped learning curve: beginners achieve results with short prompts, the middle stage falls into an “agentic trap” with many agents and complex orchestration, and experts return to short prompts backed by deep system understanding. He emphasizes empathy with agents, regularly asking them what to refactor after each merge, and probing their knowledge gaps with the question “Do you have any questions for me?”

One illustrative story: while traveling in Marrakech, Steinberger sent a voice note to his agent without having added voice capability. The agent detected the Opus format, used ffmpeg to transcode, retrieved an OpenAI API key, sent the audio to Whisper via curl, and returned the transcription, choosing the cloud API over a local model for efficiency.

Steinberger argues that large models’ “coding ability” is really a general problem‑solving ability: understanding a task, searching tools, evaluating solutions, and executing the optimal path, which can transfer to any domain.

Cognitive divide : Matt Shumer’s article “Big Things Are Happening” describes a K‑shaped split: those building AI become increasingly aware and anxious, while the majority remain unaware and vulnerable. He warns that judging AI based on free‑tier ChatGPT is like evaluating smartphones with flip phones.

Steinberger’s practice shows that the key is not AI’s intelligence but learning to collaborate with it, turning the developer role from code writer to team leader.

Prediction: 80% of apps will disappear : Steinberger predicts that most apps will be replaced by agents that can perform tasks via chat interfaces, rendering many existing apps obsolete. Surviving apps will need to become agent‑friendly APIs or accept being “slow APIs” that agents can simulate.

How to become a builder : When asked for advice, Steinberger’s one‑word answer is “Play.” He encourages experimenting with AI daily, building small projects, and letting the AI act as an infinite‑patience teacher. Shumer echoes this, suggesting an hour of AI experimentation each day for six months to surpass peers.

Steinberger’s personal journey—from 13 years at PSPDFKit to a year‑long hiatus, then a one‑hour prototype that changed the world—embodies the shift from “coding for the sake of coding” to “building experiences.” He concludes that the future has arrived, but the meaning of coding has changed.