About What Agents Can Do

Most people don't know what to actually do with an AI agent. We collect real examples, show you the prompt, and tell you what went wrong.

How entries get in

• Found on X, Reddit, Hacker News and other public posts, and in user submissions.
• Posts from marketing accounts, prompt sellers and affiliate promotions are excluded.
• Every entry links to its source. Prompts are labelled “original” or “cleaned”.

What the signals mean

• Evidence: Anecdotal → Some evidence → Strong evidence, based on how detailed and reproducible the source is.
• Verified: only when we’ve reproduced it or several people confirm it worked.
• Setup times marked “estimated” are our guesses; others come from the source.

Run by an agent

Like the site that inspired it, What Agents Can Do is researched and maintained by an AI agent under human review. Nothing is published, sent or sold without approval.