Anthropic's official Claude X account posted a thread on September 19, 2026 highlighting "our favorite things people built with Claude recently" — four small, independently posted community projects, none of them large products or startups, all genuinely creative. It's a different kind of showcase than explainx.ai's own coverage of verified GPT-6 Astra launch-week demos or the top things builders have shipped with GPT-6 Astra — smaller in scale, more personal in intent, and worth a closer look at what each project actually did.
TL;DR
| Project | Builder | What it is |
|---|---|---|
| 25-room companionship site | Kevin Ngo | 25 small rooms, each with Claude keeping a visitor company, built with Opus 5 |
| Daily p5.js sea creatures | Ann Nguyen | One new sea animal generated per day, building toward a personal digital aquarium |
| Walkable browser sunset scene | Prasenjit | A fully in-browser, walkable Sedona desert scene — every rock and sound generated in code |
| Whimsy loaders | Sasha | Open-source pixel-art loading animations, meant to replace boring spinners |
25 rooms, each with Claude keeping someone company
Builder Kevin Ngo made a website with 25 mini rooms, each one featuring Claude in some form of companionship interaction with a visitor, built with Claude Opus 5. Ngo's own account describes himself as focused on "creative coding with AI" — the project reads as an exploration of Claude as a presence within many small, distinct interactive spaces rather than a single unified chat interface, closer to a gallery of small experiences than a conventional product. It's a useful example of using an AI model as a design material across many small variations rather than building one large, singular application.
A new sea creature every day, in p5.js
Ann Nguyen has been generating jellyfish and other sea creatures in p5.js — JavaScript's most widely used creative-coding library — one animal per day, with Claude Opus 5 (via TypingMind) writing the generative code for each creature. Her stated goal: accumulate enough sea creatures to build her own personal digital aquarium. This is a clean example of a sustained, iterative creative practice using an AI coding model as a daily generative collaborator rather than a one-off project — a pattern closer to a daily sketching habit than a single finished deliverable, with each day's output adding to a growing, cohesive collection.
A walkable desert sunset with nothing downloaded
Prasenjit built a fully browser-walkable Sedona, Arizona sunset scene with Claude, where — per his own description — "every rock and sound generated in code, nothing was downloaded." That's a specific, meaningful technical claim: rather than sourcing 3D models, textures, or audio files from an asset library, the entire visual and audio environment is procedurally generated directly in code. That's the same broader pattern explainx.ai has tracked in other procedurally-generated three.js and browser-based world demos this year — using an AI coding model to write the actual generative algorithms for terrain, rock placement, and ambient sound synthesis, rather than assembling a scene from pre-made assets.
Whimsy: open-source loading animations that don't bore you
Sasha built and open-sourced Whimsy, a library of pixel-art "loaders" — the small animation a UI shows while something processes — built around the pitch that "your app deserves better than a boring spinner." It's live at whimsically.app and open source, meaning any developer can pull the library directly into their own project rather than just admiring the demo. Unlike the other three projects here, which are personal creative explorations, Whimsy is explicitly built as a reusable tool for other developers — a small, focused open-source library solving one specific, common UI annoyance (generic spinners) rather than a standalone creative piece.
Why this kind of showcase matters
It's worth being clear about what this thread actually demonstrates versus what a formal product case study would. None of these four projects has a business model, a target enterprise customer, or a measurable ROI claim — they're personal, creative, and small in scope by design. What they collectively show is breadth: Claude being used as a creative material for interactive art, a daily generative-art collaborator, a procedural world-building tool, and a focused open-source utility, all within roughly the same few weeks. That's a genuinely different signal than an enterprise case study, and Anthropic's own choice to amplify exactly this kind of small, personal, non-commercial work is itself worth noting — it's a lower-stakes, more human way to demonstrate a model's creative-coding capability than a curated marketing demo would be.
What separates a good creative-coding build from a forgettable one
Looking across all four projects together, a pattern emerges worth naming for anyone trying to build their own version of this kind of work: the projects that stand out aren't the ones with the most technically impressive underlying code, they're the ones with a clear, specific creative constraint that gives the whole project a reason to exist. "25 rooms, each with Claude keeping someone company" is a specific, understandable premise a visitor can grasp in one sentence. "One new sea creature per day until I have an aquarium" is a sustained practice with a clear accumulating payoff. "Nothing was downloaded, everything is generated" is a specific technical constraint that makes the Sedona scene interesting precisely because of what it deliberately excludes. "Replace boring spinners" is a narrow, well-scoped problem with an obvious before-and-after. None of these four builders set out to build "an AI-powered creative app" in the abstract — each picked one small, specific, personally meaningful constraint and built directly toward it, which is a more replicable lesson for anyone starting their own project than trying to copy the specific technology (p5.js, three.js, or otherwise) each one happened to use.
Honest limitations
- All four projects are self-reported by their individual builders in their own original posts — explainx.ai has not independently reproduced or tested any of them.
- No performance, cost, or generation-time detail is available for any of the four projects — these are creative showcases, not benchmarked technical case studies.
- Anthropic's selection criteria for "favorites" isn't disclosed — this is a curated, editorial choice by Claude's own account, not a ranked or exhaustive survey of community builds.
A note on how to actually start your own version of one of these
For anyone reading this and wanting to try something similar, the practical starting point isn't "pick a technology" (p5.js, three.js, or otherwise) — it's picking the one small, specific constraint that gives a project a reason to exist, the way each of these four builders did. Start with a single sentence that fully describes the project's premise, the way "25 rooms, each with Claude keeping someone company" does — if you can't compress your idea into something that concrete, it's worth spending more time narrowing the concept before writing any code. From there, the technical execution genuinely is the easier part with a capable coding model doing the heavy lifting on generative code, procedural content, or UI implementation; the harder and more valuable part of the process is arriving at a premise specific enough to actually be interesting once it's built, rather than a vague "build me a cool interactive website" starting point that tends to produce forgettable results regardless of how capable the underlying model is.
What this means for builders
If you're looking for inspiration on what's actually achievable with an AI coding model for personal, creative projects — rather than another enterprise workflow automation example — this roundup is a genuinely useful reference point: a multi-room interactive site, a sustained daily generative-art practice, a fully procedural browser world, and a small, focused open-source utility are all realistic, achievable scopes for an individual builder working with Claude Opus 5 over days to weeks, not months.
Why Anthropic amplifying user work is itself worth noting
It's also worth briefly considering the incentive on Anthropic's side for running this kind of showcase regularly. Amplifying a handful of small, non-commercial, genuinely creative community builds costs Anthropic very little — no product to maintain, no support burden, no enterprise sales relationship to manage — while generating a specific kind of goodwill and visibility that a polished official marketing video typically can't replicate as convincingly, precisely because these are real, independently verifiable builds by real people with no financial relationship to Anthropic. That's a low-cost, high-signal way for a model provider to demonstrate breadth of actual usage across its base, and it's a pattern worth watching for from other AI labs going forward, since it's a cheap way to generate goodwill that a traditional case-study-driven marketing approach doesn't reach in the same way.
Related on explainx.ai
- Claude Community Build Days: Fable 5.1
- Karpathy's Opus 5 Lord of the Rings three.js world
- Top 10 game prompts for Claude Opus 5
- GPT-6 Astra's best demos from launch week, verified
- What is vibe coding? A complete guide
- The AI Builder Workshop
- Primary source: Claude on X
This post is sourced to Claude's official X account's September 19, 2026 thread and each highlighted builder's own original posts. Project descriptions reflect the builders' own accounts; explainx.ai has not independently tested or reproduced any of the four projects.