- What is vibe coding?
- Where did vibe coding come from?
- How does vibe coding work, and how is it used today?
- Vibe coding vs. prompt engineering: What's the difference?
- Does vibe coding matter?
- TL;DR
What is vibe coding?
Vibe coding is an emerging AI-driven approach to software development where programmers use natural language prompts to generate working code through large language models (LLMs) like ChatGPT, Claude, or Code Llama. Rather than writing code line by line, developers describe what they want, and the AI generates executable code in response. This enables a faster, more exploratory coding process.
Think of it as programming by vibe: you describe the idea or intent, and let the AI fill in the technical details.
While this may sound casual, vibe coding is quickly becoming a serious paradigm, especially for:
- Rapid prototyping
- MVP development
- Low-code/no-code enablement
- Teaching and onboarding new developers
Where did vibe coding come from?
The term "vibe coding" was coined in early 2025 by Andrej Karpathy, an AI researcher and former Tesla AI director, who described it as “where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.” The phrase quickly resonated within dev and AI communities, encapsulating a growing shift toward conversational development using powerful LLMs.
While earlier forms of AI-assisted coding auto-completed code fragments, vibe coding marks a different shift: coding by idea, not by syntax. This is part of a larger wave of "natural language programming" that began gaining traction with models like GPT-3 and exploded with GPT-4 and Claude 2+.
Online communities and emerging AI tools have accelerated the adoption and refinement of vibe coding practices.
How does vibe coding work, and how is it used today?
In a simplified process, vibe coding works like this:
- Describe what you want in natural language (e.g. “Build a Python script that scrapes product names and prices from an e-commerce site and stores them in a CSV”)
- The LLM responds with code that aims to fulfill the request
- You run and debug the code, refining your prompts based on results
- Iterate conversationally, adjusting tone, constraints, edge cases, or tools used
What distinguishes vibe coding from simple prompting is the intentional collaboration between human and model—where the human guides the vibe, and the AI handles the grunt work of syntax, boilerplate, and scaffolding.
Common use cases include:
- MVP building with frameworks like Next.js
- API prototyping
- Data analysis via pandas or SQL generation
- Agent workflows bootstrapped from high-level instructions
- Internal tool creation (dashboards, scripts, automations) with minimal frontend/backend config
Vibe coding vs. prompt engineering: What's the difference?
Though similar in spirit, vibe coding does not equal prompt engineering.
Prompt engineering | Vibe coding |
|---|---|
Focuses on AI content generation (text, images) | Focuses on generating working code |
Usually optimized for one-shot answers | Built around iteration and execution |
Goal: steer LLM behavior | Goal: build functional software via dialogue |
Often used in marketing, customer support, RAG apps | Often used in dev workflows, toolchains, automations |
Vibe coding is often described as "programming by vibes, testing by execution." You don’t always write perfect specs—you try something, see what breaks, then guide the AI to refine. The feedback loop becomes part of the development process.
Vibe coding limitations
- Reliability: LLMs can hallucinate APIs, misapply libraries, or produce insecure patterns
- Debugging gap: If something breaks, users may not know how to fix it
- Tooling immaturity: Version control, test coverage, and dependency management are still manual
- Scalability: What works for a 200-line script may fall apart in a production-grade system
Does vibe coding matter?
Vibe coding reflects a broader transformation in how people are learning how to write software. For individuals, vibe coding can offer a more intuitive entry point into the world of development—especially for people from creative, analytical, or non-traditional backgrounds.
TL;DR
Vibe coding is AI-assisted development driven by natural language prompts, not syntax. You describe what you want; the AI builds it; you refine through conversation. It’s fast, experimental, and slightly disorganized—but it works. As LLMs get smarter and AI-native toolchains improve, vibe coding may become a skill for those with non-traditional engineering backgrounds.
“When our teams need something, they usually need it right away. The more time we can save doing all those tedious tasks, the more time we can dedicate to supporting our student-athletes.”
Sarah Harris
Secretary, The University of Tennessee Athletics Foundation, Inc.
“Ramp had everything we were looking for, and even things we weren't looking for. The policy aspects, that's something I never even dreamed of that a purchasing card program could handle.”
Doug Volesky
Director of Finance, City of Mount Vernon
“Switching from Brex to Ramp wasn’t just a platform swap—it was a strategic upgrade that aligned with our mission to be agile, efficient, and financially savvy.”
Lily Liu
CEO, Piñata
“With Ramp, everything lives in one place. You can click into a vendor and see every transaction, invoice, and contract. That didn’t exist in Zip. It’s made approvals much faster because decision-makers aren’t chasing down information—they have it all at their fingertips.”
Ryan Williams
Manager, Contract and Vendor Management, Advisor360°
“The ability to create flexible parameters, such as allowing bookings up to 25% above market rate, has been really good for us. Plus, having all the information within the same platform is really valuable.”
Caroline Hill
Assistant Controller, Sana Benefits
“More vendors are allowing for discounts now, because they’re seeing the quick payment. That started with Ramp—getting everyone paid on time. We’ll get a 1-2% discount for paying early. That doesn’t sound like a lot, but when you’re dealing with hundreds of millions of dollars, it does add up.”
James Hardy
CFO, SAM Construction Group
“We’ve simplified our workflows while improving accuracy, and we are faster in closing with the help of automation. We could not have achieved this without the solutions Ramp brought to the table.”
Kaustubh Khandelwal
VP of Finance, Poshmark
“I was shocked at how easy it was to set up Ramp and get our end users to adopt it. Our prior procurement platform took six months to implement, and it was a lot of labor. Ramp was so easy it was almost scary.”
Michael Natsch
Procurement Manager, AIRCO
