English | Chinese | Japanese

AI Berkshire — Value Investing Research Framework for the AI Era

"Price is what you pay, value is what you get." — Warren Buffett

AI Berkshire is a collection of investment-research skills compatible with Claude Code and Codex. It systematizes the approaches of Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu and uses AI agents to perform professional-grade value-investing research.

One person + Claude Code / Codex = an investment research team.

Real Track Record

The framework is presented as a real-money investing system rather than a paper exercise. The repository reports a 2024 full-year return of +69.29% and a 2025 full-year return of +66.38%, alongside comparisons with major market benchmarks.

Why not just ask an AI?

A general-purpose model can answer questions, but professional investment research requires a repeatable process: collecting primary evidence, separating facts from assumptions, checking valuation and financial quality, comparing competitors, identifying risks, and challenging the initial thesis. AI Berkshire packages these workflows as reusable skills so an agent follows a structured research method instead of producing a one-shot opinion.

Research philosophy

The project is built around long-term value investing. It emphasizes business quality, durable competitive advantages, management quality, capital allocation, owner earnings, valuation discipline, and a margin of safety. Different skills represent complementary perspectives drawn from Buffett, Munger, Duan Yongping, and Li Lu.

Multi-agent research

AI Berkshire can run multiple research roles in parallel and combine their findings. Separate agents can investigate business fundamentals, industry structure, competitive advantages, financial statements, valuation, management, risks, and counterarguments. Adversarial review is used to reduce confirmation bias and force the final thesis to address evidence that contradicts it.

Skills

The repository includes a broad set of reusable investment skills for company screening, business analysis, moat analysis, financial analysis, valuation, management assessment, industry research, risk analysis, thesis construction, and critical review. Skills can be used individually or orchestrated into a larger research process.

Quick start

Clone the repository and use it with Claude Code or Codex. The agent can be asked to research a company using the supplied skills and templates. The repository includes configuration, scripts, tests, research tools, and example material intended to make the workflow repeatable rather than dependent on a single prompt.

Research output

The intended result is a decision-oriented research report that explains what the company does, how it makes money, the quality and durability of its business, its competitive position, financial condition, management, valuation, major risks, and the conditions under which the investment thesis would be invalidated.

Design principles

• Prefer primary sources and verifiable evidence over summaries.
• Separate facts, estimates, assumptions, and judgments.
• Evaluate the business before focusing on the stock price.
• Use valuation as a range rather than false precision.
• Explicitly search for disconfirming evidence.
• Keep the process reproducible so another analyst or agent can audit the reasoning.
• Treat AI as a research multiplier, not as a substitute for investment judgment.

Disclaimer

This project is for research and educational purposes. It does not constitute investment advice, and users remain responsible for their own investment decisions.