Jev Daily

Illyism points an agent at a clone of his product

A refund email from a confused customer turned into a terminal session of OSINT and auto-filed abuse reports, none of it independently verified.

ILIAS ISM says a customer emailed him for a refund after accidentally buying a clone of one of his products. He pointed an AI agent at the clone's domain and let it run what he describes as full OSINT and takedowns in the terminal. OSINT here means open source intelligence gathering, which is to say pulling together public records until they point at a person.

ILIAS ISM
@illyism
X
wild what one guy with an agent can do in 20 mins now tbh
Nov 11, 2026 · View on X

The list he posted is specific. The agent pulled WHOIS and Cloudflare DNS records across 10 linked domains, used Google Ads Transparency to surface what he calls the operator's real KYC name and a Chinese ad agency, traced JS bundles and public events back to the developer's personal GitHub, and decoded live Stripe merchant IDs. It then auto-filled and submitted abuse reports to Stripe, Google Ads, Google Sign-In OAuth, Cloudflare and the registrar, driving Chrome to do it. He puts the whole run at 20 minutes.

What is actually evidenced here

One post from the person who ran it. He does not say which model, which harness, or which browser automation layer filed the forms, so there is no way to reproduce this from the description. He does not say whether any of the abuse reports were accepted, whether the clone came down, or whether the identification was correct. A name surfaced by cross-referencing ad transparency records and a GitHub account is a hypothesis until someone acts on it, and the only confirmation offered is his own confidence.

For people building on it, the interesting part is not the investigation, it is the last step. An agent that fills in and submits abuse reports to five platforms is an agent with write access to other companies' enforcement queues, and the cost of a wrong identification lands on a third party rather than on your own infrastructure. The same account has previously argued for a two step spend confirmation gate on write-enabled tools, as covered earlier. Submitting a takedown is a write with no refund path.

Know what Jev is good at before you bet on it

One email a day. Under three minutes. What builders found out about Jev yesterday.

Get the next one by email

Jev, read daily so you do not have to. The builds, the benchmarks, the criteria that worked and the cases where it lost, from the people shipping on TypeSafe AI's System One model.