# How AI Berkshire Incorporates Charlie Munger's Inverse Thinking into Investment Analysis > AI Berkshire applies Charlie Munger's inverse thinking to investment analysis by asking "How could this company die?" Discover how this AI flags weak moats and avoids poor investments. - Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire) - Tags: how-to-guide - Published: 2026-07-05 --- **AI Berkshire automates Charlie Munger's "inversion" mental model by forcing its LLM agents to ask "How could this company die?" before rendering any investment verdict, effectively flagging shallow economic moats when users and suppliers could rapidly find substitutes.** AI Berkshire is an open-source investment research framework that codifies the four-master methodology of Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu. The system implements Charlie Munger's inverse thinking as a mandatory risk-analysis stage that runs parallel to traditional valuation methods, ensuring that catastrophic failure modes are surfaced before capital is committed. ## The Munger Inversion Module in the Research Pipeline The investment research workflow explicitly defines inversion as a core analytical phase. In [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md), the research flow lists **"Inversion (Munger)"** as the fourth sequential module, positioned after data collection, business-essence analysis (Duan), and moat assessment (Buffett) γ€πŸ“„ Lines 28-34】. This stage forces the system to invert the standard bullish analysis. Instead of asking why a company will succeed, the agent must answer: *"If this company disappeared tomorrow, how would its users and suppliers react?"* If the answer indicates that stakeholders would quickly find substitutes, the system classifies the moat as shallow and flags the investment for further scrutiny. ## The Inverse Thinking Prompt: "How Could This Company Die?" Within the same skill definition, AI Berkshire generates a concrete Munger-style test prompt for the LLM. The specific prompt **"How could this company die?"** is embedded in the research logic to capture catastrophic risk scenarios γ€πŸ“„ Lines 94-100 in README_EN.md】. The agent collects a list of plausible failure paths, including: - Regulatory crackdowns or policy shifts - Competitive displacement by agile entrants - Technology obsolescence rendering core products irrelevant - Supply chain collapse or key supplier defection Each failure path is documented as a **"Top Risk (Munger)"** entry, preserving Munger's directive to *"Invert, always invert"* when evaluating business durability. ## Risk Table Generation and Structured Output The inversion analysis feeds directly into a structured risk table in the final investment memo. According to [`README_EN.md`](https://github.com/xbtlin/ai-berkshire/blob/main/README_EN.md), the output template dedicates a specific row titled **Top Risk (Munger)** that displays scenarios, probability assessments, and impact weights derived from the inversion exercise γ€πŸ“„ Lines 394-401】. The resulting markdown table follows this format: ```markdown | Top Risk (Munger) | Regulatory policy uncertainty; new business losses dragging overall profits | β˜…β˜…β˜…β˜†β˜† | | **Munger**: "Invert, always invert β€” if this company vanished tomorrow, what would users and merchants do? If the answer is 'quickly find a substitute,' the moat isn't deep enough." | ``` This structured output ensures that subjective risk assessments are captured alongside quantitative metrics, with [`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py) verifying that any quantifiable risks (e.g., potential market-cap loss) cited in the Munger analysis are mathematically sound. ## Parallel Execution in the Investment Team Workflow AI Berkshire implements the inversion step as a dedicated agent within its multi-agent architecture. In the `/investment-team` workflow, the Munger inversion agent operates **in parallel** with the other three mastersβ€”Buffett's moat analysis, Duan's business-essence evaluation, and Li Lu's civilizational trend assessment γ€πŸ“„ Lines 4-6 in README_EN.md】. This parallel execution ensures that contradictory viewpoints surface naturally before the final synthesis. The Munger agent produces its own independent risk list, which appears in the combined risk table under **"Industry & Competition | Munger"**, preventing confirmation bias that might arise from sequential analysis. ## Practical Usage Examples To execute a full deep-research cycle including the Munger inversion stage, invoke the research skill with a ticker symbol: ```bash # Run a full deep-research on a ticker; the Munger inversion is part of the flow /investment-research Tencent ``` For the multi-agent team approach where the Munger analysis runs in parallel with other masters: ```bash # Four agents run in parallel; the Munger agent will produce its own risk list /investment-team Meituan ``` The second command triggers four specialized agents simultaneously, with the Munger agent specifically tasked with stress-testing the investment thesis through inverse thinking. ## Summary - **Dedicated Module**: The `investment-research` skill explicitly lists "Inversion (Munger)" as the fourth analytical stage in [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) γ€πŸ“„ Lines 28-34】. - **Specific Prompt**: The system generates the concrete question "How could this company die?" to force catastrophic risk identification γ€πŸ“„ Lines 94-100】. - **Structured Output**: Results populate a **Top Risk (Munger)** table row with probability and impact assessments in the final memo γ€πŸ“„ Lines 394-401】. - **Parallel Validation**: The `/investment-team` workflow runs the Munger agent concurrently with Buffett, Duan, and Li Lu agents to surface contradictions before synthesis γ€πŸ“„ Lines 4-6】. - **Mathematical Rigor**: [`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py) verifies quantifiable risks identified during inversion analysis. ## Frequently Asked Questions ### How does AI Berkshire's inversion analysis differ from standard risk assessment? Standard risk assessment typically evaluates known vulnerabilities within a company's current trajectory. AI Berkshire's Munger-style inversion instead forces the system to assume total business failure and work backward, asking specifically *"If this company disappeared tomorrow, how would its users and suppliers react?"* This approach, implemented in [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md), identifies moat fragility that forward-looking analysis often misses. ### Can the Munger inversion agent override a positive investment thesis? Yes. The parallel architecture in `/investment-team` allows the Munger agent to flag "Pass/Fail/Gray-Zone" recommendations independently. If the inversion analysis reveals that customers could easily substitute the company's product (indicating a shallow moat), this risk assessment feeds directly into the final decision memo, potentially overturning bullish signals from other agents. ### What specific output format does the Munger inversion generate? The agent produces a structured risk table entry labeled **Top Risk (Munger)** containing failure scenarios, subjective probability ratings (e.g., β˜…β˜…β˜…β˜†β˜†), and impact assessments. This output appears in the final investment memo according to the template defined in [`README_EN.md`](https://github.com/xbtlin/ai-berkshire/blob/main/README_EN.md) γ€πŸ“„ Lines 394-401】, alongside a direct quote reminding analysts: *"Invert, always invert β€” if this company vanished tomorrow, what would users and merchants do?"* ### Is the inversion analysis mathematically verified? While the initial inversion step relies on LLM reasoning to identify failure modes, any quantifiable financial impacts (such as projected market-cap loss or revenue decline) cited in the **Top Risk (Munger)** section are validated by [`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py). This ensures that subjective risk assessments are grounded in mathematically verifiable data before influencing the final investment recommendation.