Vibecode Something, Keep It, Let People Play It

For months, the most capable surface on AlgoArena was locked inside assessments: an AI coding workspace where you describe what you want, watch the code land as reviewable diffs, and see the result running live. Candidates got it for ninety minutes at a time. Then the session ended and the workspace went away.

That is now a standalone product. Vibecode is a builder where anyone can describe an app, a tool, or a browser game, iterate on it with the same agent workspace the assessment uses, and keep the result. No timer, no proctoring, no question panel. Projects autosave, live as tabs you can switch between, and export whole.

The showcase closes the loop

Building is more fun when someone plays what you made. The showcase is a public gallery of published projects with three deliberate mechanics borrowed from the communities that got this right.

First, everything is playable in place, and everything is inspectable. Any visitor can open a game, play it immediately, and click See Inside to read the full source with no account. Second, appreciation is one signal: a single love per person, no dislikes, no ratings. Third, forking is one click and carries credit with it. Fork a game and your copy opens in your own builder with an editable credit line already written, waiting for you to say what you changed. Research from the remix communities we studied is unambiguous on this point: automatic attribution alone reads as insincere, and what actually makes creators glad to be forked is a human sentence of credit. So we prefill the sentence and let the forker finish it.

Browsing is curated rather than algorithmic: a featured row, a new-this-week row, a most-loved row, and no infinite feed. Publishing uses showcase slots that come with paid plans, but playing, browsing, and reading source are free for everyone, permanently. The paid layer sells reach, never access.

Practice assessments you can start yourself

The same release adds self-serve certified practice: a real assessment, on platform-owned questions only, that you can take without any recruiter inviting you. It runs the full assessment flow and the scored result lands in your own private history, shareable if and when you choose. Your work has always been yours to keep here; now the credential that proves it is too.

Why the assessment constraints stayed behind

Extracting the builder clarified something we had only half-articulated: several of the best consumer features were deliberately absent from assessments because they contaminate evidence. Standing personal preferences injected into every prompt would break standardized scoring. An autopilot that iterates unsupervised until the preview stops erroring would produce work no reviewer could attribute. In the builder, both are just features, and both shipped: a memory that carries your style across projects, and an autopilot mode that reads its own runtime errors and fixes them within a bounded number of passes.

Same engine, two disciplines. The assessment optimizes for evidence a reviewer can trust. The builder optimizes for the distance between an idea and a thing your friends can play. It turns out most of the engineering serves both.

How to read this update

The important question is not only what shipped. It is what new evidence the product can now preserve. For vibecoding work, a useful release should make the next session easier to understand: what happened, why it mattered, and what someone can do with the result afterward.

That is the bar we use for field notes. A feature is stronger when it turns a hidden process into something reviewable: a timeline, a report, a map, a score explanation, a replay, or a teacher-facing control. The surface can stay simple, but the evidence behind it should get harder to hand-wave.