Money on Autopilot

Robinhood announced Tuesday that AI agents can now trade stocks autonomously on its platform. Not suggest trades. Not flag opportunities. Buy and sell equities with real money in a dedicated account you fund yourself.

The setup: you create a separate brokerage account, deposit however much you're comfortable losing (Robinhood's word choice, not mine), and connect an AI agent via MCP, the Model Context Protocol that Anthropic open-sourced last year. The agent sees your portfolio. The agent makes decisions. The agent executes. You get push notifications after the fact.

Robinhood's official risk warning, buried in the fine print: “Agentic trading involves significant risk, including the possible loss of your entire investment.”

That sentence should be on a billboard.

The feature launches in beta with equities. Options, crypto, event contracts, and futures are on the roadmap. Robinhood Gold Card holders also get an AI shopping agent: set a budget on your virtual credit card, tell the agent what you want, and it scours the web for deals. You can opt into approving each purchase manually. You can also just let it run.

Two years ago, in Issue 006, we covered Anthropic giving AI agents a credit card. That was a curiosity. This is the sequel where the agent manages your brokerage account, and 27.4 million Robinhood users can try it.

The Protocol Behind the Curtain

MCP is worth understanding because it's going to show up everywhere.

Model Context Protocol is an open standard that lets AI systems plug into apps and data sources. Think of it as USB-C for AI: one connector, many devices. Anthropic released it as a fully open standard. Any AI agent built on any model can use it to talk to any platform that supports it.

That's the part Robinhood is banking on. They built a port that any AI agent can dock into. Your personal Claude, a custom GPT, an open-source agent you built yourself. All of them could, in theory, connect to your Robinhood account and start making trades.

The pitch is customization. The reality is that Robinhood just made itself the default brokerage for the AI agent economy, and every agent builder now has a reason to integrate with them.

The Track Record Nobody Wants to Talk About

AI-powered stock picking has a credibility problem.

Investopedia published research on May 19 showing that AI-generated portfolios overwhelmingly concentrated in mega-cap tech stocks. The recommendations tracked media coverage volume, not fundamentals. The portfolios beat the S&P 500 in raw returns, but the alpha vanished once researchers factored in trading costs and the heavy tech tilt. In other words: the AI wasn't finding hidden opportunities. It was reading the same headlines you are and buying whatever CNBC talked about most.

Robo-advisors (Betterment, Wealthfront, the Schwab and Vanguard versions) have been around for a decade. The market is real: $11 billion in 2025, projected to hit $158 billion by 2035. But the hybrid models, where a human reviews the AI's decisions, hold 58% market share. Pure robo-advisors are growing fastest among younger investors, which makes sense. They're cheapest. Whether “cheapest” and “best” overlap here is a different conversation.

Robinhood's own numbers tell an interesting story. Q1 2026 revenue hit $1.07 billion, up 15%. But net income grew only 3%. As one Zacks analyst put it: “They are running faster but not getting much further.” Prediction market revenue exploded 320%. Crypto revenue cratered 47%. The company is replacing its lost momentum with newer, more speculative revenue streams.

Adding AI agent trading to that mix is consistent with the strategy. Whether it's consistent with your financial health is the question.

The $15 Trillion Regulatory Void

Gartner projects AI agents will intermediate $15 trillion in purchases by 2028. McKinsey puts the retail agentic commerce figure at $3 to $5 trillion by 2030. Visa predicts millions of consumers will use autonomous shopping agents by holiday 2026. Keyrock found AI agents already settled $73 million across 176 million blockchain transactions in the past year.

Those numbers would be reassuring if anyone were regulating this.

The SEC's predictive analytics rule, proposed under Gary Gensler, would have required brokers to identify and eliminate conflicts of interest when using AI to interact with investors. Under the current administration, the rule has stalled. Europe's MiCA framework, the GENIUS Act, and the EU AI Act are all expected to take effect mid-2026. None of them directly address autonomous machine-to-machine transactions. None of them answer the basic question: when an AI agent loses your money, who's liable?

Visa reported a 450% increase in dark web posts mentioning “AI agent” over six months. Traditional fraud detection was built around human behavior patterns. AI agents don't behave like humans. They move faster, they trade in patterns that look nothing like a person clicking “buy,” and the adversaries targeting them are using AI too.

“Know Your Agent” frameworks are being discussed the same way “Know Your Customer” was discussed in the early 2000s. Discussed. Not enforced.