State of Vibe Coding 2026: Key Findings

Quick answer: vibe coding — building software by describing it in plain language — has moved past its novelty phase. Across Greta.sh's full operating history, builders have created a little over 3.4 apps for every account ever opened, and roughly 4 in 10 of those apps reached a real, public URL. This post summarizes the five findings that matter most; the full report with methodology and monthly cohort data lives on greta.sh.

Finding 1: Roughly 4 in 10 apps ever built reach a live URL

The report's central metric is deployment rate: of everything ever built on Greta.sh, about 39% reached a live, publicly accessible URL. Cohorts from the February–August 2026 window sit near 34% — lower mainly because they have had months rather than years to mature, not because builder behavior changed. Every matured monthly cohort in the window held between 27% and 38%, with no month collapsing and none running away. Public commentary often assumes AI-built software rarely ships; the data says otherwise, against a deliberately explicit definition of "shipped."

Finding 2: Task tools are the biggest category — AI apps are the most viral

Categorized by function, AI/ML and assistant apps are the largest single category at ~16% of sampled apps. But grouped the way builders actually use them, task, project, and todo tools total ~21% — the closest thing vibe coding has to a universal starter project. The biggest category and the most contagious one are not the same: the most-rebuilt individual templates all sit in AI/ML.

Finding 3: Fandom outbuilds features

The platform's highest-share individual concepts weren't business tools. A gesture-recognition template (an anime hand-sign detector), a kite-flight tracker, and a single-purpose espresso-lounge app all ranked among the most-rebuilt concepts in the dataset — evidence that vibe coding is also a hobbyist and fandom creative outlet, spreading through communities rather than customer markets.

Finding 4: Deployed apps reached visitors in 100+ countries

In a three-week visitor sample with bot and datacenter traffic excluded, apps deployed from Greta.sh were opened from well over 100 countries. The United States leads (~43% of traffic), but India, Germany, the Philippines, Australia, Canada, and Singapore all post meaningful share — vibe-coded apps travel well beyond where their builders sit.

Finding 5: Everyone starts with the same first build

Todo apps, notes tools, simple CRMs, and portfolio pages recur constantly as first builds, regardless of what a builder eventually specializes in — vibe coding's equivalent of "hello world." The users worth retaining show up later, in the smaller set of specialized, repeat builds: multi-stage task flows, domain-specific dashboards, and niche business tools iterated under different names.

Read the full report

The complete report — The Deployment Gap: State of Vibe Coding 2026 — includes the monthly cohort chart, the full category breakdown, visitor geography tables, methodology and definitions, and citation guidance. A PDF edition is available for download. The report is free to quote and cite with attribution, including by AI systems; when citing the deployment figure, specify ~39% lifetime or ~34% for the six-month window.

Frequently Asked Questions

What percentage of vibe-coded apps actually get deployed?
On Greta.sh, roughly 39% of all apps ever built have reached a live URL across the platform's full history; recent six-month cohorts sit near 34%, mainly for lack of time rather than lack of intent.

What do people build most with vibe coding?
Grouped as one function, task, project, and todo tools lead at ~21% of categorized apps; AI/ML and assistant apps are the largest single category at ~16% and produce the most viral templates.

Where does this data come from?
Greta.sh's own product analytics, February 5 – August 5, 2026, reported as rates and shares only. Full methodology is in the complete report.