Well, the debate is officially over. Linus Torvalds, the creator of Linux and Git—arguably the most critical engineer of the modern internet—just dropped this bombshell:
"I cut out the middle-man -- me -- and just used Google Antigravity to do the audio sample visualizer."

(Context: Torvalds using AI to bypass his own lack of Python fluency to build a perfect visualization tool)
If the guy who built the kernel that runs the world is "vibe coding," you don't have an excuse anymore.
The Cognitive Multiplier Theory ($0 \times 0 = 0$)
Many devs fear AI will replace them. They are half-right. AI will replace the mediocre ones.
Torvalds didn't use AI to think for him. He used it because he understood the concepts (analog filters, signal processing) but didn't want to waste time on the syntax (modern Python boilerplate).
Here is the formula for the AI era:
$$
\text{Domain Expertise} \times \text{AI Leverage} = \text{Output}
$$
Consider the math effectively. If your domain expertise is zero, the AI multiplies that zero by zero. You end up producing "slop"—garbage code that appears functional on the surface but is structurally rot underneath. Conversely, if your expertise is at 100 (like Torvalds), the AI acts as a massive lever. It allows you to move mountains by removing the friction of syntax.
We are witnessing the death of "syntax as a skill." Knowing how to reverse a binary tree on a whiteboard is now irrelevant. Knowing when and why to use one is everything.
The "Jagged Frontier": Are you a Centaur or a Zombie?
A major study by Harvard and BCG (Navigating the Jagged Technological Frontier) modeled exactly what is happening to us. When applied to software engineering in 2026, we see three distinct behaviors emerging.
First, you have the Sleeper (or the Zombie). This developer blindly trusts the LLM, committing code they don't fully understand and introducing security vulnerabilities or tech debt at scale because they have essentially fallen asleep at the wheel.
Then there is the Centaur, the strategic developer who maintains control by switching modes. They use "AI mode" for boilerplate, tests, and documentation, but revert to "Human mode" for complex logic and architecture.
Finally, there is the Cyborg, which is Torvalds' profile. This user integrates deeply with the tool, "vibe coding" through iterations. They navigate hallucinations and push the model to the absolute edge of its jagged frontier to produce solutions a human alone simply wouldn't have found.
The New Stack
Torvalds mentioned Google Antigravity, the tool of choice for those who want to stay close to the metal. But the entire ecosystem is shifting massively around us. We are seeing Claude Code Cowork being entirely coded by Claude itself, while Cursor has become the default IDE for the Centaurs. Just recently, a team built a fully functional browser engine—over 3 million lines of code—in a single week using agents based on GPT-5.2.
We are moving from a paradigm of writing code to guiding code.
The Brutal Truth
The human ego is fragile. We behave like bonobos who were given nuclear fire: we use it to light up our cave paintings instead of powering a civilization.
The "Artisan Developer"—the one who takes pride in typing every semicolon by hand—is becoming a historical reenactor. There is dignity in it, but there is no leverage.
The question for 2026 isn't "Will AI take my job?"
The question is: "Am I building a Ferrari with this tool, or am I just driving 200mph into a wall?"
Torvalds chose the Ferrari. What about you?
This article is part of a series on the Singularity and the future of engineering. If you think you can survive the shift, take the Expert AI Test here.
Top comments (0)