Vibe Coding
Accepting AI-generated code without reading or understanding it.
Simple Definition
Vibe Coding is when you accept AI-generated code based on "the vibe" without actually reading or understanding what the code does. You click "Accept" because it looks right, not because you verified it.
Origin
Coined by Andrej Karpathy (former Tesla AI director, OpenAI founding member) in February 2025. He described it as:
"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."
When It's Problematic
High Risk: Using vibe coding on production systems, security-sensitive code, or new features. Veracode 2025 found 45% of AI-generated code contained vulnerabilities in controlled testing.
Why It Fails: AI generates plausible-looking code that compiles and runs—but may:
- Introduce security vulnerabilities
- Silently mock dependencies instead of using real ones
- Create subtle logic bugs that only surface in production
When It's Acceptable
Karpathy himself noted it's "not too bad for throwaway weekend projects." Vibe coding has lower risk for:
- Prototypes you'll discard
- Learning/exploration code
- One-off scripts
- Non-production experiments
The key is knowing when to switch gears—before promoting code to production, review every line.
Key Takeaways
- Vibe coding = accepting AI code without reading it
- Coined by Karpathy in Feb 2025
- High risk for production code (45% vulnerability rate)
- Acceptable for throwaway prototypes and learning
- Always switch to review mode before shipping