Robinhood unveiled Robinhood Agents at its HOOD Summit on Sept. 29, an in-app AI agent builder that can research markets, develop strategies and place stock, options and crypto trades on customers' behalf, according to the company's announcement.

The feature is "coming soon" to eligible U.S. customers. A companion feature called Loops, which will let agents run recurring strategies around the clock, remains in development. Together, the two products move agent creation and automated execution inside Robinhood's mobile app, extending an agentic infrastructure the brokerage first opened to third-party systems in May.

That earlier launch required customers to connect their own external AI agents through Model Context Protocol (MCP), an open standard for linking AI models to external services. Robinhood says more than 150,000 customers have opened agentic trading accounts since then. The Defiant reported in July that Robinhood planned to expand agentic trading to crypto alongside the launch of Robinhood Chain's public mainnet.

Scope and Limits of Agent Authority

Robinhood's setup guide places strict structural boundaries around the new product. Customers need an open primary individual investing account and can create a single built-in agent with its own dedicated trading account. Users select the AI model and set the funding level. They can choose cash or limited margin, though the margin option only permits trading with unsettled sale proceeds rather than actual borrowing.

The trading guide lists equities, options and crypto as supported assets. Trade approvals default to on for built-in Agents, requiring users to review and manually place proposed orders. External-agent MCP accounts default to approvals off. Disabling approvals lets an agent place eligible trades without per-order confirmation, though certain trades still require approval regardless of settings.

Crypto trading carries additional conditions. It requires a corresponding Robinhood Crypto account and acceptance of updated agreements. Agents can trade supported crypto pairs but cannot transfer, stake or lend the assets. Agent-based crypto trading is unavailable in some states, including New York, and built-in agents must retain crypto trade approvals in California and Connecticut — a state-level patchwork that constrains the rollout's geographic footprint.

An agent can view information from a customer's other Robinhood accounts but can trade only in its dedicated account. Robinhood warns that agents can misinterpret instructions or act on outdated information, and the disclosures place responsibility for agent actions and investment losses squarely with the customer.

Loops and the Automation Frontier

Loops would convert a strategy into a standing instruction — checking markets each morning and trading when specified conditions are met, for example. Robinhood says users can stop a Loop at any time, but its disclosure notes the feature may place, modify or cancel trades automatically while a user is away, and pausing it will not reverse positions or orders already placed.

The Regulatory Distinction: Tool Versus Adviser

The disclosures reveal how Robinhood is drawing its line between AI tooling and regulated investment advice. Delegating orders to an AI agent and hiring a portfolio manager are, in the company's framing, different service relationships. The Agents disclosures assign model selection and monitoring to the customer; Robinhood Labs supplies technology, not investment advice.

The contrast is Robinhood Strategies, the firm's advisory product. Its June 2026 adviser brochure states that personnel verify AI outputs used for investment decisions and that the investment team retains portfolio-management responsibility. Supervision — not the mere presence of AI — is the consequential distinction between the two arrangements. Clients can lose money under either model.

With Agents rolling out to eligible U.S. customers in the coming months and Loops still without a launch date, the near-term question is adoption: whether the 150,000-customer agentic base expands meaningfully once creation requires no external MCP setup, and how state-level crypto restrictions shape where the automation takes hold first.