← All posts
Vibe coding: real or hype?
·4 min·needavibecoder

Vibe coding: real or hype?

25% of YC W25 ships 95% AI-generated code. The trend is real. The term is being abused. Here is what separates the shippers from the LARPers

opinionvibe-coding

25% of YC W25 startups have codebases that are 95% AI-generated. The trend is not hype. But the word "vibe coding" is being stretched to include everything from real senior work to avocado demos on Fiverr

Real vibe coding

A senior engineer with ten years of instinct, using Cursor and Claude Code to deliver production functions in hours rather than weeks. They read every diff. They know why the model chose that pattern. They reject half of what it suggests

Fake vibe coding

A junior with a free trial, ask until the tests turn green, go to main and hope. The code will be sent. The bug appears in production three days later. A senior cleans it up over the weekend

The line

The tools are the same. The flavor isn't. Vibe coding is real when people still have the architecture in their heads. It is hype when humans are just typists for the model

Why the term gets abused

Because it sells. "Vibe coder" on a Fiverr job brings clicks. "Senior engineer using Claude Code in agent loops with custom rule files" doesn't fit a headline. So the term is stretched until it includes everyone who has ever opened Cursor

Meanwhile, real companies quietly ship production functions in an afternoon while LARPers bicker on Twitter about whether vibe coding is "real programming." It is. They just don't do it

The 92% number lies a little

92% of US developers use AI tools at work. It sounds like everyone is now a vibe coder. They're not. Most use it for autocomplete and the occasional "explain this regex." Tab-complete is the basics. Vibe coding starts where tab-complete ends, in agent loops, multi-file refactors, terminal-driven workflows

The difference between "uses Copilot" and "sends functions in agent mode with Claude Code" is the difference between a typist and a writer. Same tools, totally different output

How to tell the difference

Ask them to go over a recent PR with you. Not the happy path, the edge case that they had to write by hand because the model couldn't see it. If they can't show you one, they're not a vibe coder. They're a prompt-and-pray

The verdict

Vibe coding is real. The output gap is real. The 25% of YC W25 startups shipping 95% AI-generated code are not faking it. But the term has been hijacked by the laziest 80% of the market, and it's your job as a hiring manager to filter out

Read the differences. Watch thirty minutes of how they work. The difference between real and hype is clear in the first ten minutes

See how we →