Karpathy Declares "Agentic Engineering" as the New Default for Software Development

By Andrej Karpathy | One year after coining "vibe coding"

A New Term for a New Reality

On the one-year anniversary of introducing "vibe coding" to the world, Andrej Karpathy — former AI lead at Tesla and OpenAI co-founder — is already retiring the framing in favor of something more precise: agentic engineering.

The term is deliberate on both counts. Agentic because the new default mode of software development is no longer writing code directly. Developers now spend 99% of their time orchestrating AI agents and acting as oversight — reviewing, directing, and approving rather than typing. Engineering because Karpathy wants to make clear this is not passive or trivial. There remains a genuine art, science, and hard-won expertise to doing it well.

The Profession Is Being "Dramatically Refactored"

Karpathy didn't soften the implications. He described the programming profession as being dramatically refactored, with direct human code contributions becoming increasingly sparse. The role is shifting from builder to director — from hands on keyboard to judgment in the loop.

Yet he sees enormous upside for those who adapt. His view: developers who properly string together the tools and agents already available today could be 10x more powerful than those working in traditional ways. The gap between those who embrace agentic workflows and those who don't is widening fast.

"This Is What Post-AGI Feels Like"

In a separate post, Karpathy described observing a fully autonomous AI workflow execute end-to-end without any human intervention — and captioned it simply: "ah yes this is what post-AGI feels like."

He also admitted, candidly, to never having felt this far behind — a striking signal from one of the most respected AI researchers in the world. If Karpathy feels the pace of change is outrunning him, it's a useful calibration for anyone assessing how quickly their own teams need to adapt.

What This Means

The shift Karpathy is naming is not a future prediction — it's a present observation. The question for engineering teams and the leaders who manage them is no longer whether AI will change how software gets built, but whether they are building the organizational muscle to operate in an agentic mode before competitors do.