Anthropic is building AI agents for traders, but they're not ready to replace them

Anthropic has developed a new series of AI agents built specifically for financial markets. The company wants to win over traders and investment funds on Wall Street, but recent experiments show that today's language models aren't yet ready to operate fully independently in real-world trading.

Ambitions in the financial market

The financial sector is one of the most attractive markets for B2B AI. Every day, traders handle complex tasks involving huge volumes of data, with billions of dollars at stake. If AI can reliably help with market analysis, spotting opportunities, and decision-making, it could give any fund a massive competitive edge.

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 risk of costly human errors.

Where the experiments fell short

But testing exposed serious problems. It turns out that LLMs, for all their strength on text tasks, are fundamentally ill-suited to the demands of financial trading:

• Execution speed — in trading, microseconds are everything. LLMs operate with latencies the market won't tolerate.
• Hallucinated numbers — models can misread quotes or "make up" figures, and in finance that means losses.
• Shallow grasp of context — LLMs miss political signals and macroeconomic news that move markets by billions.
• No accountability — huge sums require guarantees and a clear chain of decisions, which AI cannot provide.

Experienced traders aren't just people doing routine work. They are professionals who combine intuition, decades of experience, instant reactions to market moves, and skillful risk management. Replacing all of that with a single model proved far harder than expected.

The path to hybrid systems

The setback in Anthropic's experiments isn't a failure of AI but an entirely expected stage of development. The company has gathered valuable data, learned exactly where its models fall behind, and can work on improvements.

The realistic path forward appears to be hybrid systems rather than full automation. AI handles the analysis: it works through hundreds of reports, picks out anomalies, monitors markets in real time, and issues recommendations. The trader reviews the analysis, adds context, and executes the decision. Human judgment stays at the final stage, where the stakes are high.

What this means

Traders can breathe easier — AI won't take their jobs tomorrow. Robots won't take over Wall Street. Instead, we'll see gradual, practical automation: report analysis, anomaly detection, position monitoring. Complex strategic decisions that demand experience and accountability will remain in human hands for a long time to come.