What happened
An AI investment assistant just got a lot more literal. Russian brokerage T-Investments (part of T-Bank) has opened up a way to connect AI agents like Claude and ChatGPT directly to brokerage accounts, using a connection standard called MCP (Model Context Protocol). In plain terms: instead of just asking an AI chatbot "what do you think of this stock," you can now let it actually place the trade.
Here's how it works. MCP is basically a universal plug that lets an AI model talk to outside software — think of it as a translator that lets Claude or ChatGPT "speak" the language of your brokerage's systems instead of just spitting out text in a chat window. Once you connect an agent through MCP, you give it a task in plain English ("rebalance my portfolio toward tech" or "sell half my position if it drops 5%"), and the agent converts that instruction into an actual trade order sent to the exchange.
T-Investments didn't just fling the doors open, either. There's a permissions dial, similar to app permissions on your phone. You can restrict an AI agent to read-only access, where it can only look at your trade history and portfolio, or you can grant it authority to execute orders and even move money between your own accounts. The rollout was first reported by Vedomosti, and independently confirmed by VC.ru, a second VC.ru report, trading community Smart-Lab, and T-Bank's own announcement.
What it means for you
This isn't just a broker feature — it's a preview of where AI agents are headed across everyday life. A few concrete scenarios:
At home: instead of manually logging into your brokerage app every few days to check on your investments, you tell an AI agent once — "keep my portfolio at 60% stocks, 40% bonds, and rebalance if it drifts more than 5%" — and it handles the boring maintenance while you go live your life.
At work: a small business owner managing a company's cash reserves in interest-bearing instruments can hand routine treasury moves to an agent with tightly scoped permissions (view and transfer only, no risky trades), freeing up an afternoon a week that used to go into spreadsheet babysitting.
Running a business: a solo founder who doesn't have time to watch markets can set rules — "never let any single position exceed 10% of the account" — and let the agent enforce discipline that's hard to stick to emotionally when you're busy building a product.
As a student: someone learning about investing can use an AI agent in read-only mode to ask plain-language questions about their own (small, practice) portfolio — "why did this position lose money this week" — without risking real capital while they learn.
For extra income: a freelancer with irregular income can automate the unglamorous part of investing — moving spare cash into a savings instrument the moment it hits the account — something people plan to do manually and usually don't.
For creators: anyone whose income depends on unpredictable platform payouts can set an agent to auto-invest a fixed percentage of every payment received, removing the temptation to just let it sit in a checking account.
The common thread: the AI isn't picking stocks for you out of thin air. It's executing rules and instructions you set, faster and more consistently than you'd manage by hand.
How to try it right now
You don't need to be a programmer to test this, but you do need a T-Investments brokerage account, since that's the specific platform offering the connection right now.
1. Start free. Sign up for a Claude account (the free plan is enough to test a conversation-based connection) or use ChatGPT's free tier at ChatGPT if you already use OpenAI's assistant.
2. Open your T-Investments account settings and look for the AI agent / MCP connection option, as described in T-Bank's own post.
3. Set permissions first, before anything else. Start with view-only access — let the agent read your transaction history and answer questions about your portfolio. Don't jump straight to giving it trade or transfer rights.
4. Test it with a harmless task, like "summarize my portfolio's performance this month," before trusting it with anything involving real money movement.
5. Only then, if you're comfortable, widen permissions to order execution — and even then, consider capping how much the agent can move in a single action.
If you want to experiment with AI tools more broadly before connecting anything to real money — drafting the instructions you'll give an agent, summarizing financial news, or just getting comfortable with how AI chat tools respond to plain-language commands — a free, no-signup option like mykreatool.com is a low-stakes place to practice prompting before you hand an AI agent the keys to an actual brokerage account.
Upsides and what changes
The obvious upside is time. Portfolio rebalancing, moving idle cash into interest-bearing accounts, and reacting to price swings are all things people mean to do and often don't, simply because life gets in the way. An agent that executes rules consistently removes that gap between intention and action.
The bigger shift is conceptual: AI stops being just an advice-giver and starts being an actor with real-world consequences. That's a meaningful line to cross. Up to now, an AI chatbot telling you bad information about a stock cost you nothing more than wasted time. An AI agent with execution rights that misreads an instruction can cost you real money in seconds.
Limitations
Be honest with yourself about what this actually is: a permission you're granting a piece of software to move your money, and AI models can still misunderstand ambiguous instructions, get confused by unusual market conditions, or act on a poorly worded prompt in a way you didn't intend — there's no regulator-backed guarantee that an AI agent trades exactly the way a human would in every edge case. The feature is also currently specific to T-Investments' own platform, so if you bank or invest elsewhere, this exact setup won't apply to you yet, though similar integrations from other brokers are a reasonable bet given how fast MCP adoption is moving. Start with the narrowest permission level you can, and only expand it once you've watched the agent behave correctly for a while.
Conclusion
AI agents are moving from "tell me what to do" to "do it for me," and brokerage accounts are now part of that shift. One action for today: if you have a T-Investments account, log in and connect an AI agent in view-only mode — no trades, no transfers — just to see what it can already tell you about your own money.
Comments 0