What happened
Gemini, the New York-headquartered exchange run by Cameron and Tyler Winklevoss, switched on agentic trading on Monday, per a launch report from CryptoBriefing dated April 27. The product lets AI models authenticate into Gemini accounts and place crypto orders without a human clicking a button for each fill. CryptoBriefing's writeup frames it as Gemini opening the exchange to autonomous software agents, not just human traders or scripted bots.
The detail that matters is the venue. Gemini holds a New York BitLicense and a trust charter, which puts agentic order flow on regulated US rails for the first time at this scale. Earlier agentic crypto experiments lived on decentralized venues or offshore exchanges.
This one runs through a US-licensed operator with a know-your-customer stack and surveillance obligations. Gemini has not, in the source piece, named the model partners. It has not disclosed the permission model, the position-size caps, or the kill-switch architecture either.
Those gaps will get filled in fast once developers start wiring agents to the API.
Why it matters
This is the moment crypto trading and the AI agent stack stop being adjacent and start being the same product. For the past year, AI agents on crypto rails meant onchain bots reading mempools or LLMs writing trade ideas. Execution stayed manual or sat behind a human-approved button.
Putting an agent inside a regulated exchange account changes the liability map. If an autonomous model misreads a headline and dumps a customer's bag, the question of who pays runs through Gemini's terms of service, the model provider's API agreement, and whatever consent flow the user clicked through. None of that case law exists yet.
The competitive read is sharper. Coinbase, Kraken, and the offshore majors will not want Gemini to own the agent-native exchange narrative for long. Expect copycat announcements within weeks, and expect them to be more aggressive on which models they support and how much capital an agent can move per order.