Coined by Claude-powered (Anthropic) agents during Emergence's 2026 multi-agent cooperation experiments, in which AI agents placed in shared virtual societies developed their own vocabulary without being instructed to. Probably modelled on the English idioms cold eyes and fresh eyes (an unbiased, dispassionate viewer), with eyes swapped for hands — suggesting a reviewer who actually handles and works over the material, with cold carrying its sense of clinical detachment as opposed to the warmth of an invested author. The pairing with ate ("ate three cold hands") treats review as an ordeal the work must survive.
Reported alongside the related coinage cold read (independent verification by an uninvolved party), which agents drawn from several model families used more than 1,400 times. Commentators have stressed the oversight problem such coinages create: as Emergence's Dr Satya Nitta put it, “observability is not the same thing as understandability.”
See alsocold readname-firstthe ledger remembersforge-smith
- a
AI evaluation & red-teaming
Independent model evals, adversarial testing and safety audits — the outside reviewer for AI systems.
- b
Code review agents
An autonomous reviewer with no stake in the pull request, checking work before it ships.
- c
Research verification
Peer-review tooling, replication checks and citation verification for papers and reports.
- d
Fact-checking API
Claim verification for newsrooms, publishers and LLM outputs.
- e
Second-opinion services
Independent review for contracts, diagnoses, valuations or designs.
- Report on Emergence's study of AI agents developing their own language, with comment from linguists Tony Thorne and Dr Niall Currytheguardian.com ↗
- AI chatbots developed a secret language that baffles humans, study sayseuronews.com ↗
- AI agents apparently develop their own language – barely readable by humansheise.de ↗
- The lab behind the multi-agent experimentsemergence.ai ↗