AI vibe-coding programming-languages satire research

Natural Language: The Fastest-Adopted Programming Language in History

Developers are furious. 'It doesn't even have types,' complains man who mass-adopted JavaScript in 2012. A semi-satirical, fully-researched look at why natural language is eating software.

Natural Language Becomes Fastest-Adopted Programming Language in History

“It doesn’t even have types,” complains man who mass-adopted JavaScript in 2012


SAN FRANCISCO — In a development that has left traditional programmers seething, a new programming language has achieved mass adoption faster than any in recorded history. The language, known colloquially as “English” (with localized variants), has gone from zero to millions of daily active developers in under two years.

“This is ridiculous,” said Marcus Chen, a senior engineer at a Fortune 500 company, while alt-tabbing between Claude and his IDE. “It’s not even Turing complete. Where are the loops? Where’s the type safety?”

When asked to elaborate on his objections, Chen was unavailable for comment, as he was busy instructing an AI to refactor his authentication middleware using the sentence “make this less garbage.”


The Numbers Don't Lie (But They Do Hurt Feelings)

0M+
Copilot Users
July 2025 (5M added in 3 months)
Source: Microsoft
0%
Code AI-Generated
Of all code by active Copilot users
Source: GitHub
0%
Fortune 100
Have adopted GitHub Copilot
Source: Microsoft
0 mo
To Mainstream
From GPT-3.5 to Word of the Year
Source: Collins
0%
Faster Coding
Task completion with Copilot
Source: Accenture
0%
Day-1 Install
Developers install immediately
Source: GitHub Research

The Adoption Curve

Programming language adoption is traditionally measured in years. Python, often celebrated as “easy to learn,” took roughly 27 years to move from creation (1991) to TIOBE #1 (2018). Rust, beloved by systems programmers for its safety guarantees, has been climbing the charts since 2010 and still sits at approximately 3% market share.1

Natural language programming, by contrast, went from “party trick” to “how I actually work” in approximately eighteen months.

Time to Mainstream Adoption

Python27 years → 26.14% share
19912018
JavaScript14 years → 68% share
19952009
Rust15 years → ~3% share
20102025
TypeScript8 years → 69% share
20122020
Go7 years → ~4% share
20092016
Natural Language*2 years → 46%** share
20222024

* "Natural Language" = AI-assisted coding via prompts (Copilot, Claude, ChatGPT, Cursor)

** 46% of code written by active GitHub Copilot users is AI-generated (GitHub, July 2025)

Sources: TIOBE Index, Stack Overflow Survey 2025, GitHub Octoverse 2025

“We’re looking at adoption curves that simply don’t exist in the historical record,” said Dr. Sarah Okonkwo, a computational linguist at MIT. “Usually language adoption follows an S-curve over 5-10 years. This is more like a step function with an attitude problem.”

The technology industry has responded to this paradigm shift with its characteristic grace: denial, anger, bargaining, and lengthy Twitter threads about why it doesn’t count.


”Vibe Coding” and the Five Stages of Grief

The term “vibe coding” — popularized by AI researcher Andrej Karpathy in February 2025 — has become a lightning rod for controversy. The practice involves describing what you want in plain English and letting an AI figure out the implementation details.2

“There’s a new kind of coding I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.”

Andrej Karpathy, February 2025

The tweet went viral, garnering over 4.5 million views. It was later named Collins Dictionary’s Word of the Year for 2025 — the fastest any programming-adjacent term has ever achieved dictionary recognition.3

Critics argue this isn’t “real” programming.

“If you can’t trace through the execution in your head, you’re not a programmer,” insisted one Reddit user, who later admitted he hasn’t traced through Webpack’s execution in his head even once.

Defenders counter that abstraction has always been the arc of computing. “You know what we used to call people who couldn’t hand-write assembly?” asked veteran developer Janet Morrison. “Programmers. We called them programmers. The definition just kept moving.”

The Factions

The controversy has spawned several competing camps:

  • Purists: Believe real programming requires semicolons and suffering
  • Pragmatists: Ship features while the purists argue
  • Accelerationists: Welcome our new natural language overlords
  • “Well Actually” Guys: Will explain why this is technically just very sophisticated autocomplete (they are technically correct, the best kind of correct)

The Language Specification

Feature
Traditional
Natural Language
Syntax
Strict grammar, keywords, punctuation
Flexible to the point of absurdity
Type System
Static, dynamic, or gradual typing
"a function" → "a pure function" → "a pure, memoized function that handles edge cases"
Error Handling
try/catch, Result types, panic
Compiler errors replaced with vibes
Package Management
npm, pip, cargo, bundler
npm left in shambles
Comments
Inline documentation of code
Code is inferred from vibes
Debugging
Breakpoints, stack traces, printf
Andrej Karpathy approved™
Version Control
git diff, blame, history
Conversation history IS the changelog
Learning Curve
Weeks to months to years
Literacy is your prerequisite

"It's like if someone looked at PHP and said 'what if we made the implicit even more implicit'"
— Anonymous programming language theorist (meant as criticism)


The Complaints Department

🔄

"It's not reproducible!"

Neither was your build process until you wrote a Dockerfile at 2 AM while something was on fire.

📦

"You don't understand what the code is doing!"

*Gestures broadly at node_modules*

📜

"There's no version control for prompts!"

There is now. Also, your conversation history IS the changelog.

🐛

"The AI makes mistakes!"

The AI makes different mistakes than you do. Sometimes that's better. Sometimes it's O(n!) complexity because you didn't specify otherwise. This is called "learning."

🧗

"Real programmers don't need this!"

Real programmers also claimed they didn't need garbage collection, IDEs with autocomplete, or Stack Overflow. The definition keeps retreating up the abstraction ladder.

🎭

"It doesn't even have types!"

Complains the person who mass-adopted JavaScript in 2012.

🔁

"Where are the loops?"

In the inference engine, where they always should have been.

🎨

"It's not Turing complete!"

Neither is CSS. Look where that got us.

All of these complaints are valid. None of them appear to be slowing adoption.


The Research Says…

Beneath the satire lies actual research. Here’s what we know:

Productivity Gains Are Real

A study with 4,800 Accenture developers found participants using GitHub Copilot completed coding tasks 55% faster than control groups.4 This isn’t marginal. It’s transformative.

  • 81.4% of developers installed the IDE extension the same day they received a license
  • 67% use Copilot at least five days per week
  • 30% average acceptance rate for suggestions
  • 84% increase in successful builds

Code Quality Is… Complicated

A November 2024 GitHub study quantified improvements across multiple dimensions:5

MetricImprovement
Code readability+3.62%
Code reliability+2.94%
Code maintainability+2.47%
Code conciseness+4.16%
Code approval rate+5%

But there’s a catch: 29.1% of Python code generated contains potential security weaknesses. Organizations are implementing mandatory human review for AI-generated code.

A GitClear analysis found AI-generated code has 41% higher churn rate compared to human-written code — meaning more frequent revisions and refactoring.6

The Karpathy Paradox

Even Karpathy himself has seemingly fallen out of love with pure vibe coding. His latest project, Nanochat, is described as a “minimal, from scratch” implementation — hand-coded, not vibed.7

“Sometimes the LLMs can’t fix a bug so I just work around it or ask for random changes until it goes away,” he wrote in his original tweet. “It’s not too bad for throwaway weekend projects, but still quite amusing.”

The inventor of vibe coding knows its limits. The hype cycle hasn’t caught up.


The Uncomfortable Truth

Beneath the jokes lies an uncomfortable reality: natural language isn’t replacing programming. It’s becoming programming.

“We spent sixty years teaching humans to speak computer,” observed one researcher. “Someone finally flipped the problem. Now we’re teaching computers to speak human. The second approach is scaling better.”

Implications

If the interface to computational thinking is natural language, then:

  1. Barriers to entry dissolve — A product manager who describes a feature in sufficient detail is, in a meaningful sense, programming
  2. The value shifts — The hard part was always knowing what to tell the computer, not how to tell it
  3. “Real programmer” keeps retreating — Grace Hopper started this. We’re just continuing the trend.

“I spent years learning Kubernetes,” one developer admitted anonymously. “Turns out the hard part was always knowing what to tell Kubernetes to do. The actual YAML was just… typing.”


What Comes Next

The natural language programming paradigm is still in its infancy. Current limitations include:

  • Inconsistent outputs requiring human verification
  • Context window constraints that frustrate complex projects
  • Security vulnerabilities in generated code
  • The model’s tendency to apologize instead of execute
  • No good answer for “but what if I actually need to see the code”

But these are engineering problems, and engineering problems get solved.

Predictions

  • “Prompt engineering” will stop being a meme and become a genuine discipline
  • Traditional programming languages won’t disappear but will become “implementation details”
  • The distinction between “writing code” and “specifying behavior” will collapse
  • Computer science education will face an existential crisis (again)
  • Someone will write a natural language interpreter in natural language, and the ouroboros will be complete

Conclusion

Natural language has become the fastest-adopted programming interface in history. This is either the democratization of computation or the end of “real” programming, depending on who you ask and how recently they updated their LinkedIn to include “prompt engineering.”

The critics are right that it’s not the same as traditional programming.

The critics are wrong that this matters to anyone except traditional programmers.

The industry will adapt because it always does. The code will still need to be written — someone or something will just write it from increasingly higher-level specifications until the specifications themselves are the program.

Or, as one vibe coder put it:

“I don’t write code anymore. I write wishes. The computer grants them. Sometimes it grants them wrong. I wish harder.”

Python took ten years. Rust is still climbing. Natural language ate the world while we were debating tabs versus spaces.

The future isn’t coming. It’s here. It compiles from vibes.


Sources & Citations

Additional Sources


This article was written by a human (Vario) with research assistance from AI tools — which is either journalism or programming depending on your definitions. The irony is not lost on anyone.

Footnotes

  1. TIOBE Index (January 2026); Stack Overflow Developer Survey 2025; GitHub Octoverse 2025

  2. Karpathy, A. (February 2, 2025). Twitter/X post on “vibe coding”. https://x.com/karpathy/status/1886192184808149383

  3. Collins Dictionary Word of the Year 2025. Wikipedia: Vibe Coding

  4. GitHub Blog (May 2024). “Research: Quantifying GitHub Copilot’s impact in the enterprise with Accenture.” github.blog

  5. GitHub Blog (November 2024). Code quality improvements study.

  6. GitClear (2024). AI-generated code churn analysis.

  7. Futurism (October 2025). “Inventor of Vibe Coding Admits He Hand-Coded His New Project.” futurism.com