What AI agents actually did, with receipts.

Real use cases from people running Claude Code, Codex, Grok Bot, OpenClaw and more. Every case has a source, a verdict and a prompt you can try. The wipeouts get the same care as the wins.

Get the 5-minute weekly → Browse all cases

0 case files on the site · every source checked by a human · failures included

Latest case files

See all 0 →

Newsletter

Get 5 Things Agents Did

Five real things AI agents did this week: one win, one wipeout, one thing to try. Two minutes, every Tuesday. Free.

[Placeholder: signup form not connected yet.]

Browse by what you want done

💻 Coding

Shipped features, merged PRs, migrations, and the incidents.

0 cases

🧾 Finance & Admin

Bookkeeping, bills, refunds and disputes.

0 cases

⚙️ Ops

Triage, support, running a one-person business.

0 cases

🧍 Personal

Errands, shopping, negotiating, life admin.

0 cases

🔒 Security

Prompt injection, misuse, agents misbehaving.

0 cases

🔬 Research

Literature, analysis, long research runs.

0 cases

📈 Sales

Outbound, lead research, reply handling.

0 cases

📣 Marketing

Ad copy, content, events.

0 cases

💥 Wipeouts

Agents have deleted production databases, wiped drives and sent forms nobody approved. We document every one with the setting that would have prevented it. See the wipeouts →

How we check a case

Primary source or it doesn't go in. The person's own post, the GitHub issue or the vendor's report.
We open every link and spot-check every quote.
Self-reported and vendor-published cases are labelled.
Prompts are quoted, or marked "recreated". We never present our wording as theirs.

More about how we work → · Submit a case →