The Era of Typing Code Is Over — Andrej Karpathy and the Rise of Agentic Engineering

"I Haven't Written a Single Line of Code Since December 2025"

This is not a prediction. This is happening right now — from one of OpenAI's original founding members.

Andrej Karpathy — OpenAI founding member, former VP of AI at Tesla — shared on the No Priors podcast (Episode 154, aired 03/20/2026) something that made the entire tech world pause:

"I haven't written a single line of code since December 2025."

Before December, Karpathy personally wrote roughly 80% of his code. After December — that number dropped to near zero. Everything was delegated to AI coding agents.

He stated plainly: "I don't think normal people realize this has happened, or how serious it is."

---

What Happened in December 2025?

Karpathy explained clearly: before December, coding agents were fundamentally unreliable. But in December, everything suddenly "flipped" — agents began writing dependable code, handling complex multi-step tasks, and iterating through workflows without requiring human intervention on every line.

His exact words on X:

"Coding agents basically didn't work before December, and basically work since."

And earlier, also on X in December 2025, he wrote:

"I have never felt more mass-deskilled as a programmer. The profession is being restructured entirely."

That post received over 22,000 likes and 3.6 million views.

---

How Karpathy Works Now

Instead of typing code line by line, he runs multiple AI agents in parallel on a tiled screen. Each agent receives a distinct task:

• One agent builds a new feature
• Another researches solutions
• A third plans the deployment

Each agent takes roughly 20 minutes to complete its task. He moves between agents, delegates in plain English, then reviews the returned output.

He calls this new working model "agentic engineering" — the discipline of orchestrating agents. Fundamentally different from "vibe coding" (coding with AI by feel), a term he himself coined in early 2025.

---

Vibe Coding vs. Agentic Engineering

---

What Agentic Engineering Demands

Agentic engineering is not "press a button and wait." It requires:

1. Breaking large work into clear, bounded tasks
2. Assigning the right task to the right agent
3. Reviewing output and catching errors before they propagate
4. Managing multiple workstreams simultaneously

In other words: you're no longer a "programmer" in the traditional sense. You're a project manager for a team of AI agents.

Karpathy described the feeling:

"It feels like cheating. Because the code you used to be proud of writing, requiring intelligence and deep knowledge — is now free and instant. Very disorienting."

---

Real Data: Who's Being Affected?

This is the most sobering part.

According to research from Stanford (analyzing ADP payroll data, published August 2025):

Why?

AI is handling precisely the work that junior developers traditionally do:

• Basic code implementation
• Application scaffolding
• Simple function writing
• Boilerplate generation

Meanwhile, those with deep experience — who know how to decompose problems, catch architectural errors, and understand system design — become the most powerful AI operators.

Experience doesn't become obsolete. Experience becomes a multiplier.

---

The Jevons Paradox: Software Demand Will Explode

Karpathy also believes software demand won't decrease — it will surge dramatically.

He cited the Jevons Paradox: when production costs drop, consumption rises.

Applied to software:

• When creating software becomes nearly free → the number of apps people want to build explodes
• Code becomes something that can be rewritten, modified, and customized continuously
• Every business, every individual can "build an app"

This means: demand for developers won't drop — but the role changes entirely.

---

What Karpathy Says About the Future

On X, he wrote:

"You no longer type code into an editor the way things have been done since the computer was invented. That era is over."

This isn't a distant future. This is happening right now — to one of the world's leading AI experts.

---

Our Take

Karpathy isn't alone. This trend is accelerating across the industry:

• Google just launched Antigravity agent in AI Studio
• Anthropic introduced Agent Skills for modular workflows
• The agent workflow loop is becoming a standard architecture pattern

The question is no longer "Will AI replace developers?" — it's "Will you be the one orchestrating AI, or the one being replaced by it?"

Skills to invest in right now:

1. Systems thinking — understand overall architecture, not just individual lines
2. Task decomposition — break large problems into agent-sized tasks
3. Deep code review — catch logic errors that AI misses
4. Prompt & context engineering — delegate effectively to AI
5. Domain expertise — specialized knowledge that AI cannot yet replicate

---

Conclusion

Andrej Karpathy — the man who coined "vibe coding" — has already moved past it. He doesn't vibe code anymore. He agentic engineers.

And the question for you:

Are you preparing for the role of "programmer" — or for the role of "AI orchestrator"?

Source: No Priors Podcast, Episode 154 — Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AI (03/20/2026)