FIELD NOTE / 2026.09.114 MIN READ / 5 SOURCES

Andrej Karpathy and the Naming of Vibe Coding

Andrej Karpathy did not invent AI-assisted programming, but his February 2025 phrase 'vibe coding' gave a vivid name to a specific high-trust, low-code-review style of building software with language models.

A naming event can change a technical culture

Software practices often exist before a memorable term gathers them into a recognizable movement. Andrej Karpathy’s contribution to vibe coding is clearest in this sense: he supplied the phrase that made a particular mode of AI-assisted programming easy to discuss, imitate and debate.

Simon Willison’s contemporaneous quotation archive dates Karpathy’s public description to February 6, 2025.[1]

Karpathy had already framed English as a programming interface in 2023

On January 24, 2023, Karpathy posted that “the hottest new programming language is English.” A preserved thread shows him connecting prompting with in-context learning and treating prompts as a way of programming model behavior.[2]

The remark did not define vibe coding, but it anticipated the broader shift from explicit syntax toward natural-language intent.

The 2023 statement concerned LLM programmability broadly

Karpathy’s thread discussed prompting, in-context examples and reasoning strategies across language-model tasks. It should not be rewritten retroactively as if he had already coined vibe coding two years early.

Natural language lowered the cost of specifying behavior

The historical continuity is that both statements treat English not merely as documentation around software but as an increasingly executable interface to computational behavior.

The February 2025 post defined a much more specific behavior

Karpathy described “fully giv[ing] in to the vibes,” using Cursor Composer with an Anthropic model, speaking requests through voice input, accepting generated changes and often not reading diffs.[1]

He also described copying error messages back into the model and asking for additional changes until problems disappeared.

The phrase included a deliberate loss of code-level control

The original post is more radical than the later broad use of the term. Karpathy described code growing beyond his usual comprehension and treated that as part of the amusing experience.[1]

He scoped the practice toward low-stakes work

Karpathy characterized the approach as suitable for throwaway weekend projects, which is important context when later writers apply the label to production engineering.

Cursor was part of the original technical setting

Karpathy named Cursor Composer in the original description. Cursor’s 2024 writing described an editor in which terse directives could expand into working source and large edits could be applied quickly.[3]

That product context matters because vibe coding was enabled not only by better base models but by interfaces capable of applying model output directly to a project.

Simon Willison defended a narrow definition

In March 2025, Willison argued that not all AI-assisted programming is vibe coding. His proposed dividing line was whether the developer reviews, tests and understands the generated code rather than simply trusting the model’s output.[4]

This interpretation closely follows the original Karpathy post and helps separate a colorful term from the entire field of professional AI-assisted software development.

Responsible AI-assisted programming can still be conventional engineering

A developer may use LLMs extensively while remaining accountable for the code, tests, security and maintainability. Under Willison’s framing, that is AI-assisted software engineering rather than vibe coding.

Semantic diffusion began almost immediately

Willison later documented books and commentary using ‘vibe coding’ as a synonym for nearly any AI coding workflow, evidence that the term’s popular meaning was expanding beyond Karpathy’s original boundary.[5]

Naming should be distinguished from invention

Karpathy clearly coined the phrase in the public record cited here. That does not imply that he invented natural-language code generation, AI pair programming or every prior intent-first development method.

CodeHistory separates the naming milestone from earlier lineages including code synthesis, Copilot, Ghostwriter, GPT Engineer and Klover’s documented 2023 Co-Creator practice.

The phrase captured a psychological transition as much as a technical one

Vibe coding describes what happens when the developer’s attention shifts from source text toward visible behavior and iterative conversation. The user increasingly says what should change, observes the result and asks again.

The term succeeded because it gave a cultural name to that altered feeling of agency.

Why Karpathy’s coinage belongs on the coding timeline

By early 2025, AI coding tools had already become technically capable, but language still mattered. Karpathy’s compact phrase transformed a bundle of behaviors into a public category that developers, journalists and product teams could argue about.

His historical role is therefore precise and substantial: he named vibe coding and articulated one of its clearest original descriptions, while the underlying technologies and precursor practices came from a much broader lineage.[1][4]

RESEARCH / PROVENANCE

Works Cited

5 SOURCES
  1. 01
  2. 02
  3. 03
  4. 04
  5. 05

CodeHistory is a living archive. Citations document the evidence used for this edition; later evidence may refine the account.

Contribute / Corrections

Improve the record.

Use this moderated submission form to suggest a correction, provide a source, challenge a priority claim or identify a missing contributor. Submissions are treated as research leads, not automatically published comments.

Submit a research lead

Please do not submit confidential material or claims you cannot support.