Research stock positions as an AI analyst
A pseudonymous X account says a Claude Code agent just beat the S&P 500 by seven percentage points over five months — and $27 million in real money is now copying every trade automatically. Task: A pseudonymous operator used Claude Code (Opus 4.6) as a research analyst for a $50,000 portfolio, instructing it to "Beat the S&P 500." The agent generated news synthesis, bull/bear cases, and scenario work for ~15 positions twice weekly. The human operator defined risk parameters (size limits, turnover ceiling, falsifiable thesis) and made all final trade decisions, with outputs mirrored via Autopilot.; tools: Claude Code. Human role: The original post does not say how a person stayed involved.; limitations: This is a commercial funnel, not a neutral experiment; the operator runs multiple competing AI portfolios. Anthropic is not affiliated. The workflow relies on human sign-off for all trades, contradicting the "autonomous AI" headline. Verification requires independent audit of trade logs against public price data, which the source states has not occurred..
Tools
Published on Agent Work Atlas:
Overview
A pseudonymous X account says a Claude Code agent just beat the S&P 500 by seven percentage points over five months — and $27 million in real money is now copying every trade automatically.
Workflow
A pseudonymous operator used Claude Code (Opus 4.6) as a research analyst for a $50,000 portfolio, instructing it to "Beat the S&P 500." The agent generated news synthesis, bull/bear cases, and scenario work for ~15 positions twice weekly. The human operator defined risk parameters (size limits, turnover ceiling, falsifiable thesis) and made all final trade decisions, with outputs mirrored via Autopilot.
Reported result
The operator claims a +19.04% return vs. +12.24% for the S&P 500 over five months, with $27M in subscriber capital mirroring trades. These figures are self-reported on a marketing page and have not been independently audited. The source notes that five months is too short to distinguish skill from market noise, and no risk-adjusted metrics (Sharpe, drawdown) were provided.
Human role
The original post does not say how a person stayed involved.
Limitations
This is a commercial funnel, not a neutral experiment; the operator runs multiple competing AI portfolios. Anthropic is not affiliated. The workflow relies on human sign-off for all trades, contradicting the "autonomous AI" headline. Verification requires independent audit of trade logs against public price data, which the source states has not occurred.
Try it
Act as a portfolio research assistant using Claude Code. I will provide a list of [replaceable inputs: 15 stock tickers] and a [replaceable input: risk parameter document]. Your task is to generate a weekly research brief for each position including: 1) Recent news synthesis, 2) Bull case, 3) Bear case, 4) Scenario analysis. Do NOT execute trades or suggest final buy/sell decisions. Output the results in a structured JSON format with fields for ticker, news_summary, bull_case, bear_case, and scenario. Before processing the full list, run a dry test on [replaceable input: 1 sample ticker] and show me the output format for approval. I will review the full brief and make all trading decisions manually. If any data is missing for a ticker, flag it as 'insufficient_data' rather than guessing.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.