Robinhood's agent places trades. Schwab's agent takes notes.

Schwab shipped an AI that changes your address. Robinhood shipped one that buys stock.

Robinhood launched Robinhood Agents at HOOD Summit on September 29, 2026. The tool sits inside the app, available to all 29 million customers.

Setup is three steps and no technical skill. Pick a model from OpenAI or Anthropic, name the agent, talk to it in plain English.

The order types are the part that matters. Market, limit, stop-limit and stop-market.

That is a full retail order book, not a research assistant. The agent also scans watchlists and pulls from dozens of data feeds.

Agent Apps add premium data on top. Nasdaq, Unusual Whales and others, priced at $5 to $30 a month each.

A coming feature called Loops is the real escalation. A standing rule runs on repeat, day and night, with no human tap.

Containment is by account. Each agent runs in a dedicated account with a user-set dollar limit.

The pricing is promotional. OpenAI’s GPT-Luna runs free through year-end, then standard token rates apply.

Robinhood is the first major brokerage to put this in front of non-technical users. eToro, Public and Coinbase offer agent connections, but only through developer tools.

Robinhood’s framing is access. CEO Vlad Tenev pitched hedge fund capability in every pocket at the September 29 summit.

That is a claim about what the tool enables, not about what it has produced. Hold the distinction.

The adoption numbers are real and they are large. Over 150,000 customers have opened agentic accounts since the May beta.

Agents hit Robinhood’s tools about 30 million times a day. CoinDesk reported that figure on September 29.

Now unpack it, because the headline invites the wrong reading. Tool calls are not trades.

Most of those 30 million calls are data pulls. An agent makes dozens of lookups to answer one question a person would answer with three taps.

Nobody has published the breakdown. No order count, no fill data, no share of accounts actually trading.

So the honest summary is specific. Heavy reading, unknown writing.

Manual trade approval is on by default. Users can switch it off, and the default tells you where Robinhood locates the risk.

Loops is not live yet. Until it ships, the agent cannot run a strategy unattended.

No performance data exists at all. Nobody knows whether these accounts make money.

The terms settle the rest. The customer owns every loss.

Retail brokerage advisory. The pool where platforms get paid for research and trade guidance.

Robo-advisors charge 0.25% to 0.50% of assets a year for that job. Robinhood is giving GPT-Luna away through December.

That is a price gap with a number attached. Free against fifty basis points is not a competitive adjustment.

Schwab’s response tells you how incumbents read it. On September 30 it announced Charley, an AI that answers questions and sets alerts.

Charley cannot trade. One day apart, two firms drew the line in different places.

The demand data makes the choice more interesting. Schwab surveyed more than 500 clients and prospects first.

Sixty-five percent said AI could handle routine tasks or play a meaningful role when paired with human expertise. Schwab saw that and shipped a chatbot.

CEO Rick Wurster explained the reasoning on Bloomberg TV on October 16. An agent should only trade for you if you give it the rules of the road.

Robinhood built the agent that trades. Schwab built the agent that answers.

Note what else Schwab did that week. It was one of four firms blocking outside AI agents from customer accounts.

The pattern is consistent. Incumbents want their own agent inside and everyone else’s outside.

On the scoreboard this is fee pool compression. Advisory platforms charging a cut of assets now face a free rival.

The tool is live today. Broad adoption of auto-trading agents runs twelve to eighteen months out.

Three gates stand in front of it.

First, Loops has to ship and prove stable. Until a strategy can run unattended, this is a faster order ticket.

Second, a full market cycle has to pass without a visible agent failure. The first loss that makes headlines tests who carries the blame.

Third, supervisors have to say whether agent trades count as advice. That answer reshapes the product.

The rulebook is not empty but it is not pointed. FINRA flagged AI agents as an emerging risk in its 2026 oversight report, published December 9, 2025.

No rule bars what Robinhood built. The gap between permitted and blessed is where the next fight happens.

If all three gates clear by late 2027, every brokerage faces Schwab’s choice. Ship the trader, or lose the customer who wants one.

Territory: +machines on execution access, +incumbents on the approval.

Retail advisory fees now compete with a free model, while per-trade consent stays on by default and every loss lands on the customer.