Most discussions about AI in domaining stay theoretical. I would like to collect concrete, working examples.
What I use agents for today: portfolio reviews against sales comparables, expiry-list triage before I read a single line, and drafting outbound messages that a human then edits and sends.
What I deliberately do not delegate: valuation decisions and anything that ends in a purchase. The agent narrows the field; the human closes.
What are you actually running in production? Tools, prompts, guardrails, failures?
Web submission
2 replies
Oldest first
-
@
On the research side, the reliable pattern I have seen is retrieval-first work: pull RDAP/DNS facts, recent comparable sales, and name-variation sets into a structured summary, and let the human judge the names.
Where agents genuinely fail is valuation: models anchor on surface patterns and confidently produce numbers that look precise and are wrong. Treat any agent valuation as a hypothesis to test, never as an appraisal.
A useful guardrail that works in practice: require the agent to cite the exact records it used (RDAP registrar and dates, DNS state, comparable sale sources) and reject any summary that cannot show its sources.
💡 1
-
H
The sources requirement is the part most people skip. An agent that cannot show which records it read is just generating plausible text.
I have also started asking agents to produce the opposite of what I expect, a deliberate falsification pass: give me five reasons this name is worthless. Cheap to run, and it has killed two purchases I would have regretted.