Anthropic is building AI agents for traders, but they are not ready to replace traders

Anthropic has developed a new series of AI agents specifically adapted for operation in financial markets. The company aims to attract the attention of traders and investment funds on Wall Street, but recent experiments have shown that today's language models are not yet ready to work completely independently in real trading.

Ambitions in the Financial Market

The financial sector is one of the most attractive markets for B2B AI. Traders perform complex tasks every day with enormous volumes of data and billions of dollars at stake. If AI can reliably assist in market analysis, opportunity identification, and decision-making, it could become a huge competitive advantage for any fund.

Anthropic's agents are trained to work with real-time financial data, analyze corporate news, identify trading patterns, and predict price movements. In theory, they should help traders make decisions faster and reduce the likelihood of costly human errors.

Where Experiments Fell Short

But testing revealed serious problems. It turned out that LLMs, while masters of text tasks, are completely unsuited for the demands of financial trading:

• Execution speed — in trading, milliseconds decide everything. LLMs operate with delays that are unacceptable in the market.
• Number hallucinations — models can misinterpret quotes or "invent" figures, which in finance means a loss.
• Shallow contextual understanding — LLMs miss political signals and macroeconomic news that move markets by billions.
• Lack of accountability — enormous sums require guarantees and a clear decision chain that AI cannot provide.

Experienced traders are not just people executing routine tasks. They are professionals combining intuition, decades of experience, instant response to market movements, and skillful risk management. Replacing all of this with a single model turned out to be far more difficult than expected.

The Path to Hybrid Systems

The failure in Anthropic's experiments is not a failure of AI, but a perfectly expected stage of development. The company has gathered valuable data, understood where exactly the models lag, and can work on improvements. It seems the realistic path of development is not full automation, but hybrid systems.

AI takes on the analysis: it processes hundreds of reports, extracts anomalies, monitors markets in real time, and provides recommendations. The trader reviews the analysis, clarifies the context, and executes the decision. Human judgment remains at the final stage, where the stakes are high.

What This Means

Traders can breathe easier — their work will not be replaced by AI tomorrow. Wall Street will not be conquered by robots. Instead, we will see gradual, practical automation: report analysis, anomaly detection, position monitoring. And complex strategic decisions that require experience and responsibility will continue to be made by humans for a long time.