Can AI copy
Can I Vibecode It?
A ranked index of SaaS subscriptions, scored by whether an AI agent can one-shot a replacement.
They asked for it · MIT — this product open-sources itself or publicly invites clones.
- Verdict
- One shot
- Time to copy
- an afternoon
- Category
- ▤ Directories & indexes
- Source
- MIT
The prompt
Written to be pasted straight into an agent with no editing. It describes the whole job, including the parts that will go wrong.
Build a server-rendered index site that scores products against a single question, in the style of a compatibility table. Data: one JSON file per entry in data/entries/, each with slug, name, domain, category, a one-line description, a verdict from a three-value enum, supporting bullet lists, and a long-form text field. Load them at build time and fail the build on a slug/filename mismatch. Routes: a homepage with live client-side filtering over the full set plus a ranked table; a detail page per entry at /:slug; a category index and per-category pages; a stats page. Static-generate everything. UI: verdict badges colour-coded by enum value, favicon-per-row pulled from the entry domain, a sticky filter bar, and a copy-to-clipboard control on the long-form field. SEO: question-shaped titles, canonical URLs, JSON-LD for ItemList and BreadcrumbList, a generated sitemap, and a per-page share image rendered at build time. No accounts. No tracking. Contribution happens by pull request, so document the JSON schema in CONTRIBUTING.md.
What copies cleanly
- The entire product surface. It is a sorted table, a search box, category chips and a detail page.
- The data model, which they publish: one JSON file per entry, contributed by pull request.
- The vote counter. An integer, a rate limit, and a column.
- The generated share images, the sitemap, the structured data. All of it is a build step.
- The source, in full, because the repository is public and MIT licensed.
Where it stops
- Nothing technical. This is the point of the entry.
- The judgement calls. Deciding that a tool is a weekend project rather than a real one is an opinion, and opinions do not come out of the repository.
- The entries themselves, which are hand-written and accumulate slowly.
- Being first, which is not available to copy.
What actually protects them
- Taste, expressed as a verdict column.
- The contributor habit. Once people file pull requests against your index, the index is theirs too.
- Name recognition on a domain that is a complete sentence.
Editor's note
We want to be fair, because they have been fair. This site exists because they published a page inviting people to rebuild their site, released the source under MIT, and promised the prompts would stay free. That is a real position, honestly held, and we agree with it. The joke is only that a directory of things AI can replace is itself a thing AI can replace, and they clearly know that, because they wrote the instructions down. Consider this a pull request they cannot merge.
Somebody already did it
- caniuse.comcaniuse.comThe original format. Everything since is a re-skin of this table.
- canivibecodecanivibecodeit.comcanivibecodecanivibecodeit.comSomeone got here before us. The recursion is already two levels deep.
- Their own rebuild promptcanivibecodeit.comThey publish the instructions for copying themselves. We simply read them.
Go and look at the original
We would rather you formed your own opinion than took ours. They built the thing; we only measured it.
Next in the index
- caniuse.comThe browser support table that every site of this shape is descended from.Skin deep
- BitlyA shorter URL that redirects to a longer one, and counts who clicked.One shot
- LinktreeOne page of buttons, pointed at your other pages.One shot
- PlausibleWebsite analytics that fit on one screen and do not follow anyone around.One shot