Vibe Coding: How Far the Success Stories Actually Go

Somewhere between “I built an app without writing a line of code” and “$10,000 a month from a weekend project,” a real technology shift and a marketing genre started sharing the same headlines.

Vibe coding is both of those things at once, and separating them matters more than either side of the argument usually admits.

Where did the term actually come from?

Andrej Karpathy posted it in February 2025, describing a way of working where he gives in to the vibes, forgets the code even exists, and lets an AI model handle implementation while he reviews and redirects. Karpathy isn’t a random voice in this space: he co-founded OpenAI, ran Tesla’s Autopilot AI team for close to five years, later left OpenAI a second time to start Eureka Labs, an AI education company, and in May 2026 joined Anthropic’s pre-training team. His YouTube series on building GPT from scratch has millions of views among people who take machine learning seriously, which is exactly why one of his tweets was strong enough to reshape how an entire industry describes its own work.

By November 2025, Collins Dictionary named “vibe coding” its Word of the Year, citing the term’s rapid uptick in everyday use well beyond software circles.

Did the tools actually catch up to the hype?

They did, and fast. Cursor’s parent company Anysphere went from $100 million in annualized revenue in January 2025 to $4 billion by May 2026, climbing through $500 million, $1 billion, $2 billion, and $3 billion in the months between, with funding talks reportedly circling a $50 to $60 billion valuation. Claude Code processes roughly 195 million lines of code a week. Lovable, built for people who don’t write code at all, crossed $500 million in annualized revenue in 2026 after reaching its first $100 million in just eight months. Replit hit $525 million in April 2026, up more than seventeenfold year over year, and is aiming for $1 billion by year end.

The model side moved just as fast. SWE-bench Verified, which measures whether an AI can fix a real GitHub issue and pass the tests, sat around 50% for the best models in mid-2024. By 2026, the frontier is past 95%. That’s not a demo statistic anymore; it’s a measurable capability shift.

So how much faster does this actually make a developer?

Real, but not evenly distributed. A randomized controlled trial across 4,867 developers at Microsoft, Accenture, and a Fortune 100 company found an average 26% productivity gain, concentrated almost entirely in junior developers. Senior engineers who already knew their own codebase well showed little measurable speed-up in that study. A separate Google-run trial found the opposite pattern, developers 21% faster on average with more senior engineers gaining more, once seniority and prior AI usage were controlled for. Both studies are real. They just don’t agree, which is itself the honest answer: the effect depends heavily on who’s using the tool and on what.

For non-developers, the shift is more structural than a speed number captures. Roughly 90% of developers now use an AI coding tool regularly at work, but the more telling number is how many of vibe coding’s users were never developers to begin with, building a scheduling tool for their own small business instead of hiring someone or learning to code from scratch.

What do the social media success stories leave out?

Pieter Levels built fly.pieter.com, a browser flight simulator, with Cursor and Three.js, and took it from zero to $1 million in annualized revenue in 17 days, a number he posted himself in real time on X. That’s true. It’s also true that he started that 17-day clock already carrying hundreds of thousands of X followers built up over years, on top of a portfolio of other projects already running at several million dollars annually. The tool didn’t build that audience. Marc Lou crossed just over $1 million in 2025 revenue across three products with zero employees, a genuinely strong outcome that gets flattened in retelling into a bigger number than it actually is.

The retrospective posts that don’t go viral tell a different average story. The founder behind Memex, an agentic coding tool for scientific researchers, wrote candidly that it took 26 months from quitting his job to reach $10,000 in monthly recurring revenue, after pivoting twice. He signed off the post with “I’m off to vibe code something,” the same optimism, after describing two years most success threads skip entirely.

Speed of shipping doesn’t cancel out what happens after launch, either. Two critical vulnerabilities were disclosed in Cursor in 2026, tracked as CVE-2026-50548 and CVE-2026-50549, both rated 9.8 out of 10, letting a single crafted prompt escape the editor’s sandbox and execute arbitrary commands on a developer’s machine. Shipping fast and shipping something that survives contact with real users and real attackers are not the same achievement, and the gap between them is exactly the part that doesn’t make it into a launch-week screenshot.

What actually changed, then?

Not the odds of success. The cost of finding out. A failed attempt used to cost six months of building before anyone knew if the idea worked. Now it costs a weekend, and a working prototype is achievable at a speed that simply wasn’t available three years ago.

That’s a real, structural change, and it’s also a smaller claim than the SNS success stories imply. The tools got dramatically better at removing one specific bottleneck, the time it takes to build a first version. Everything that happens after launch, distribution, security, retention, the operational grind that separates $10,000 a month from $500,000, is exactly as hard as it always was, and it’s still where almost every one of these stories was actually won or lost.

References

• Collins Dictionary: Collins’ Word of the Year 2025
• Levels.io: fly.pieter.com, my first vibecoded flight simulator, reached $1M ARR in 17 days
• The Hacker News: Critical Cursor Flaws Could Let Prompt Injection Escape Sandbox and Run Commands
• TechCrunch: OpenAI co-founder Andrej Karpathy joins Anthropic’s pre-training team