Curiosity,
with receipts.
AI Can What? collects software made with AI, agent environments, benchmarks, and datasets.
Two kinds of entries. One directory.
A Showcase documents AI-built or AI-assisted software with an author source. Training and Evaluation describe what an environment is for; they do not imply that AI wrote it. Development provenance and intended use are recorded separately, with unknown facts left unknown.
Is a dataset a runnable environment?
No. Static datasets can provide screenshots, trajectories, or task examples without an interactive system. Every entry labels its resource type. Platform tags can describe an application target or the source of dataset screenshots; consult the entry for its actual setup.
How are entries selected?
- Start with the author’s project page, repository, or original paper.
- Describe what people can do with the project, its supported platforms, and practical limits.
- Use AI-built, AI-assisted, or platform labels only when the original material supports them.
- Record the source-review date and preserve links to code, papers, and author demos.
What does source-reviewed mean?
We read the project material. We have not independently installed every application, run every environment, or reproduced paper results. This is a directory, not a quality ranking. Consult upstream documentation for current setup requirements, licenses, and appropriate use.
Who provides the demos?
Code, environments, and external demos belong to their original creators. This directory provides descriptions and navigation. It does not claim to host every project or imply endorsement by another institution.
How can I inspect, reuse, or correct an entry?
Every project has its own URL, source-review date, and original evidence. Download the JSON catalog or inspect the website source and changes on GitHub. To report an error or suggest a project, send a source-backed suggestion.
Are upstream registry records individually reviewed?
The upstream registry index automatically captures packages, environment manifests and configurations with source and version metadata. It supports discovery; individual records have not undergone project-level review or execution verification. Versions and forks can overlap, so its count is not additive with the reviewed project count.
Related directory
AwesomeREA collects reverse-engineering tools, integrations, and software-reconstruction research for people studying how software works and how to rebuild it.
How is the content produced?
Research, bilingual writing and site development use AI assistance. Source review means checking original material, not an independent human audit. Actual browser interactions are recorded separately with steps, dates and limits; readers can inspect the public source and original links and propose corrections.
What do browser observations cover?
Selected entries include actual interaction steps, visible results, dates and screenshots. These describe the public demo at that time, not whole-application certification, proof of AI authorship, a deployed source version, or untested features.
- Tic Tac Toe — Imperium Edition · 2026-10-16
- Aldi Prices · 2026-10-16
- CSV To Charts · 2026-10-16
- Markdown → Slides · 2026-10-16
- Tiny Kanban · 2026-10-16