All articles
Automation

Agent-Executed Trades: What Changes When AI Places the Order

Coinbase, Deribit, and the shift to permissioned automation

HBy FNB AI Team ·November 2, 2026 ·4 min read
Agent-Executed Trades: What Changes When AI Places the Order

Not financial advice. This article is for educational purposes only and does not constitute investment, trading, or financial advice. Cryptocurrency trading carries a high level of risk, including the possible loss of capital. Past performance does not guarantee future results.

01Coinbase Opens Derivatives to AI Agents

On November 2, 2026, Coinbase announced that AI agents connected to user accounts can place trades in its Deribit-powered global derivatives markets, available to eligible customers outside the United States. Users connect their Coinbase account to an AI platform—including Claude, ChatGPT, and Grok—to research markets, develop strategies, and place trades within user-set limits The Defiant.

The mechanics are important: the agent cannot send funds off Coinbase, users choose what the agent can access, and they can revoke the connection. Most agent apps require approval before each trade. The global derivatives gateway lists at least 125 perpetual contracts with maximum leverage up to 50 times The Defiant.

125+
Perpetual contracts on the gateway
50x
Maximum leverage
0
Funds the agent can send off Coinbase

02What Actually Changes for the Trader

The shift is from manual execution to parameterized execution. Instead of watching a chart and clicking buy, the trader defines limits: which assets, what leverage, what maximum position size. The agent operates within those parameters. Coinbase explicitly warns that AI agents can make errors and that users remain responsible for reviewing and authorizing their trades The Defiant.

This is a meaningful change in workflow, not a change in accountability. The trader still owns the outcome.

03The Risk That Automation Introduces

Speed cuts both ways. An agent that can prepare and execute a trade in seconds can also execute a mistake in seconds. Virtuals' Butler was designed with a security-event monitoring function—if a held asset is affected by a security incident, Butler analyzes the impact and proposes a response KuCoin/Foresight News. But the existence of a response mechanism does not guarantee the response is correct.

The practical question for anyone considering agent-executed trades is not whether the AI is smart. It is whether the parameters the human sets are conservative enough to survive an AI error. If a 50x leverage position is within the agent's permitted range, an error at that leverage is not a minor inconvenience.

04Agentic Copy Trading as the Next Step

Virtuals went further in October 2026, launching agentic copy trading through Butler AI in closed beta. Butler tracks wallets of user-selected traders and automatically executes copy trades across multiple chains within limits set by each user TradingView/CoinMarketCal.

For the trader, this raises the question of where research ends and delegation begins. Copy trading through an AI agent is not the same as making your own entry decision. It is handing the timing decision to an agent that follows another trader.

Stays with the user
  • What percentage of capital
  • Which assets
  • Maximum position size
Moves to the agent
  • The “when” decision — entry timing, following another trader
Sources for this article
H
FNB AI Team

Research and educational content on AI-assisted crypto investing, trade execution, and risk management.

Want to put these ideas into practice?

FNB AI explores how AI and automation change the way traders approach crypto—from research and signal interpretation to execution and risk control. See how it works.

Get started