$ vibe-coding –list-top-50
The shift from line-completion to full agentic builds is the defining story of 2026. Tools now plan multi-step work, execute it, recover from their own errors, and ship. The projects below are the proof — pulled from Claude Code and OpenAI Codex ecosystems, grouped by category, every entry traceable to its source.
Sources: Claude Code and OpenAI Codex adoption reporting, 2026.
Every entry links to its original write-up, repo, or a live search you can browse. Nothing here is unsourced — the point is that you can verify each one yourself rather than take our word for it.
01–06
Enterprise & Scale-Proof Builds
The examples that answer “but does it hold up at scale?” — real companies putting agentic tools on production-critical work.
- OpenAI’s zero-manual-code product
Application logic, tests, CI config, and docs all written by Codex — roughly a million lines and ~1,500 PRs merged by a three-engineer team.
- Harvey’s legal-engineering workflow
Codex cut early iteration time 30–50%, freeing engineers for system design instead of boilerplate.
- Wonderful’s core architecture
Codex CLI reportedly replaced every other agentic harness the team used for deep-reasoning technical work.
- Tobi Lütke’s MRI viewer
The Shopify CEO built a working cross-platform viewer for a locked hospital MRI file format in a single sitting with Claude Code.
- Pietro Schirano’s DNA analysis tool
The MagicPath founder fed Claude Code a raw consumer DNA file; it spun up multiple sub-agents to analyze genome regions in parallel.
- Deedy’s Hermès spec-ad pipeline
A Menlo Ventures partner prompted Claude Code to script, voice, source visuals, and stitch a full 30-second ad from scratch.
07–19
Games, Tools, and the Everything-In-Between
The middle of the list is where vibe coding gets fun — and where the ecosystem builds for itself.
Ten playable browser remakes of arcade classics in a weekend. A live physics game with sound, sparks, and a conquest mechanic shipped in one afternoon. A full chaotic word game, published free the same day.
PeonPing — a Claude Code notification plugin with Warcraft III voice lines — hit 1,200+ GitHub stars and 40+ community sound packs within days. awesome-claude-code has become the ecosystem’s de facto directory at 45,000+ stars.
Status-line utilities showing rate-limit windows and per-session cost; “swarm” desktop IDEs running multiple sessions in a grid; and custom MCP servers wiring Claude Code into a team’s own APIs and data.
Retro space-shooter tutorials and full-stack deploy walkthroughs — backend, auth, database, and deployment built end-to-end in a single session — have become the on-ramp for new agentic builders.
20–50
The Patterns Behind the Rest
Most vibecoded projects never get a viral write-up — they live quietly on Indie Hackers, Show HN, and #buildinpublic threads. The categories repeat with striking consistency: micro-SaaS billing tools, one-task Chrome extensions, internal ops dashboards, Slack and Discord bots, browser automation, lightweight API wrappers, habit trackers, README and doc generators, automated code-review agents, support-ticket triage, podcast show-note generators, voice-memo transcribers, local-first note apps, CLI productivity tools.
Each of those is now a browseable live GitHub search — real, current examples rather than a list you have to take on faith.
The takeaway
The Through-Line Is Speed
Across all 50, the story isn’t the tech stack — it’s compression. Projects that would have needed a contractor, a sprint, or a “someday” slot on a roadmap are now built and shipped in a weekend, sometimes an afternoon.
That compression is exactly why the AI token award exists at all: when the milestone arrives faster than ever, the recognition has to catch up. A first user, a first dollar, a first “wait, it actually works” — those are the moments a dashboard forgets and a shelf remembers.
Leave a Reply