Agentic engineering

Agentic engineering is professional software development in which AI agents write the code, while the engineer sets the architecture and constraints and reviews the result. The term was proposed by Andrej Karpathy in February 2026 as a counterpoint to vibe coding.¹ ⧗

The main difference is oversight. In vibe coding, a person accepts whatever the AI produced and does not read the code. In agentic engineering, they think the solution through in advance, hand its implementation to agents and review their work.

History

• 02.02.2025Karpathy introduces the term “vibe coding”: AI writes all the code, and the person “forget[s] that the code even exists”.² ⧗ ¶ cv7bz2kd9
• 02.2026A year later Karpathy proposes the term “agentic engineering” for professional work with AI agents. According to The New Stack, he himself now barely writes code directly; he gives tasks to agents and reviews the result.¹ ⧗ ¶ c2grztim3
• 10.10.2026Linus Torvalds publishes in his Linux kernel repository an experimental branch of 8 patches tagged “Assisted-by: LLM”: “The code, the test programs, the review behind the list above and the first draft of this text were done with an LLM coding assistant (Claude Code) under my direction”. On his laptop, reading one file from 8 threads became four times cheaper; he lists 10 known problems himself, and the branch is not merged into mainline.³ ¶ cn42dqpik
• 13.10.2026Anthropic: in May 2026 around 2,500 code changes a day were merged into its main repository on average, and over 30,000 a day by September, making source control the biggest development bottleneck. Seven people built a replacement for Git in five months — the centralized system Scion — along with the code review tool Antique; the authors say Scion is no general-purpose replacement for Git.⁴ ¶ cxyrpvsbr