Natural language as an interface
Vibe coding is an approach to software creation in which a person describes desired behaviour in ordinary language and an AI model produces or modifies code. The term was popularised by Andrej Karpathy in 2025.
A HUMAN × MACHINE FIELD EXPERIMENT // 2026
THE MONKEY
DISCOVERED
ENGLISH COULD
COMPILE.
I didn't know I was vibe coding. I thought I was building worlds.
What happens when the distance between an idea and a working prototype becomes a conversation?
01 / INTERACTIVE DEMONSTRATION
Speak an idea.
Watch a prototype appear.
This is a simulated demonstration, not a live code-generation service. Choose a concept to see how a request might map to interface building blocks.
● ● ● /prototype/preview
02 / THE EXPLANATION
What is vibe coding?
From prototype to product
AI can quickly scaffold interfaces, suggest fixes and iterate on designs. But a prototype that looks right is not automatically secure, reliable, accessible, or maintainable.
Vibe coding vs conventional programming
Traditional workflows usually require the developer to write and inspect more of the code directly. AI-assisted workflows shift some effort toward specifying intent, evaluating output and managing integration. Both can be combined.
Who is it for?
Designers, researchers, entrepreneurs, hobbyists and experienced developers can use these tools. Domain knowledge and careful testing remain valuable, especially when real users or sensitive data are involved.
03 / THE REALITY CHECK
The vibe is not
the verification.
01 — Specify. Describe the user, the task, constraints, and what success looks like.
02 — Generate. Ask AI for a small, inspectable change rather than an opaque mountain of code.
03 — Test. Check keyboard navigation, mobile layout, broken links, edge cases and errors.
04 — Secure. Keep secrets out of front-end code, validate inputs and review dependencies.
05 — Maintain. Record decisions, version changes and understand enough to recover when things fail.
Automation becomes more dependable when you understand the process you're automating.
That is an interpretation drawn from building and debugging websites, not a universal law. AI can also help teach the process along the way.
04 / TAKE SOMETHING HOME
Keep the curiosity.
Bring the checklist.
Download the practical field guide, explore related worlds, or write directly to Sahil.
Email fallback: sahil@sahiltayal.com