About
You can. That’s not the question.
canivibecodeit.com asks whether an AI agent can replace a piece of paid software. It is a good question, the answers are careful, and it is right far more often than not. An agent really will produce a working password vault in one session.
This site asks the other one. Should you ship it?
Those are different questions with different answers, and the gap between them is where the expensive mistakes live. Difficulty and consequence are independent axes. A URL shortener is fifty lines of code and a permanent promise to keep a server alive. A password manager is an afternoon and your entire digital life. Nobody is confused about whether these are buildable — they are confused about whether building them is a good idea.
How the list was made
We took every entry canivibecodeit rated YES or KINDA — the things that already cleared the feasibility bar — ranked them by how popular they are over there, and took the top 149. If an agent cannot build it, whether you should is not an interesting question, so the NOT REALLY entries are out of scope entirely.
Their verdict then becomes the horizontal axis of our map. This list is the filter that runs after theirs.
How verdicts are made
Every entry is scored by a human on six dimensions — blast radius, data sensitivity, security surface, regulatory exposure, irreversibility and maintenance tax — which combine into a regret score out of 100. The verdict shown on the page is editorial and can override that number, but never by more than one band; a validator fails the build if it does.
Anything claiming regulatory exposure has to cite a source. Verdicts are opinions, but opinions with consequences attached should show their working.
The two prompts
Build prompts are canivibecodeit’s half of this, they are good, and every entry here links to theirs. Ours is a different artefact: the guardrail prompt, which you paste before you build so the agent builds the risky thing less badly. One gets the thing running; the other keeps it from taking you down with it.
What we don’t do
No sponsorships. No affiliate links. Nobody has paid to be rated SHIP IT and nobody can pay to be moved off ABSOLUTELY NOT.
Credit
Entry metadata — names, categories, pricing, prior art — is seeded from the canivibecodeit repository, which is MIT licensed. The category taxonomy is theirs, used verbatim so the two sites cross-link cleanly. The “should” verdicts, the risk taxonomy and the scoring model are ours.
Start there if you haven’t. It tells you what is possible; this site picks up where that answer stops.
Good — most of these are judgement calls. Entries live as JSON in a public repo and every verdict is a pull request away from changing.
Can you? → canivibecodeit.com
Should you? → you’re here.