AI Agents in Trading: From Recommendations to Autonomous Management
OpenAI
Anthropic
AI trading agents are evolving from recommendation engines to autonomous portfolio managers. Companies like Podium Markets AI, Robinhood, and Public are building systems where AI can execute trades, though human oversight remains. Key challenges include operationalizing natural language goals into strict rules and ensuring safety. The Russian broker Finam has implemented an MCP server to connect its trading platform to AI assistants like ChatGPT and Claude.
The article discusses the trend of AI agents in trading, where AI not only recommends investments but also executes them. Podium Markets AI's assistant Ivy analyzes portfolios and provides recommendations, but users still decide whether to trade. Robinhood lets third-party AI agents connect to client accounts, and Public develops agents that automate investment processes. Retail investors use general AI tools like ChatGPT and Claude for research with mixed results. A major challenge is explaining investor intent to AI agents; for example, 'aggressively grow my portfolio' can be interpreted in various ways. Companies implement safeguards, such as requiring user approval of agent workflows. The Finam broker has implemented an MCP server for FinamTrade, allowing users to connect their accounts to ChatGPT, Claude, and other services via tokens. This provides secure access to portfolio data and quotes through natural language queries, but does not yet execute trades. The article highlights that AI agents should operate within defined limits and that human oversight remains crucial.
- Abbreviations
- MCP = Model Context Protocol — протокол контекста модели
Source: Habr — хаб ИИ —
original