# How AI Berkshire Conducts Deep Dives into Management Teams: A Parallel-Agent Architecture > Discover how AI Berkshire analyzes management teams using parallel agents. Learn its approach to assessing CEO ability, capital allocation, governance, and reputation for a comprehensive score. - Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire) - Tags: how-to-guide - Published: 2026-07-06 --- **AI Berkshire conducts management deep dives through a dedicated `/management-deep-dive` skill that orchestrates four parallel agents to simultaneously assess CEO ability, capital allocation decisions, governance structures, and side-channel reputation, ultimately generating a weighted composite score based on integrity (35%), ability (25%), capital allocation (25%), and governance (15%).** The **xbtlin/ai-berkshire** repository implements management evaluation as a reproducible, code-driven workflow rather than subjective research. At its core, the system treats leadership assessment as a structured data pipeline, leveraging a **three-layer design philosophy** (Skill → Agent → Tool) defined in `assets/architecture.mmd` to deliver evidence-based analysis aligned with Buffett’s "integrity, wisdom, energy" criteria. ## The `/management-deep-dive` Skill Architecture ### Skill Definition and Repository Structure The management deep dive is encapsulated as a discrete **Skill** within the AI Berkshire ecosystem. The human-readable workflow definition resides in [`skills/management-deep-dive.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/management-deep-dive.md), which specifies the orchestration logic, agent prompts, and scoring rubrics. This source file is auto-synchronized to [`codex-skills/management-deep-dive/SKILL.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-skills/management-deep-dive/SKILL.md) to ensure compatibility with Claude Code and Codex environments, maintaining parity between local and cloud-based execution. The skill follows the repository’s **three-layer architecture**: the Skill layer defines the command interface (`/management-deep-dive`), the Agent layer executes background data collection, and the Tool layer provides precise financial calculations through utilities like [`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py). ## Parallel Agent Execution for Data Collection ### Identification and Enumeration The workflow begins with a web-search step to identify key executives—including the CEO, CFO, founder, controlling shareholder, and other senior managers—before launching specialized agents. This identification phase is documented in [`skills/management-deep-dive.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/management-deep-dive.md) lines 25‑35. ### The Four Concurrent Background Agents Once key personnel are identified, the skill spawns **four parallel agents** to gather distinct data streams simultaneously (lines 53‑58): - **Agent 1** collects CEO public statements, shareholder letters, earnings calls, interviews, and social media activity to assess strategic communication. - **Agent 2** tracks capital-allocation actions, including M&A history, buy-back programs, dividend policies, and new-business investments. - **Agent 3** analyzes governance and compensation structures, examining equity structures (AB-shares, super-voting rights, VIE arrangements), executive pay rationality, and related-party transactions. - **Agent 4** performs side-channel validation by scraping employee reviews (Glassdoor), customer feedback (App Store, Zhihu), and industry reputation data to triangulate management quality. ## CEO Ability-Circle and Integrity Evaluation ### Strategic Foresight and Execution Capability The system evaluates CEO performance through an **ability-circle assessment** (sections 2.1 & 2.2). This examines **strategic foresight**—measured by the accuracy of past market predictions and forward guidance—alongside **execution capability**, which tracks delivery on promises, talent attraction metrics, and crisis-response effectiveness. ### Commitment Fulfillment Tracking For integrity evaluation, the skill calculates a **commitment fulfillment rate** by comparing public commitments made over the past three years against actual delivery (sections 3.1 & 3.2). The analysis quantifies the gap between promises and outcomes, weighting leadership behavior during crises as a critical modifier to the integrity score. ## Capital Allocation and Governance Assessment ### Financial Rigor Verification Every major capital-allocation decision is logged and scored against financial fundamentals. The skill invokes **`tools/financial_rigor.py verify-valuation`** to cross-check M&A premiums, buy-back timing relative to intrinsic value, and dividend sustainability (sections 4.1 & 4.2). This ensures capital deployment aligns with shareholder value creation rather than empire-building. ### Equity Structure and Compensation Analysis The governance review examines equity structures for control mechanisms like AB-shares, super-voting rights, or VIE arrangements that might misalign incentives (sections 5.1‑5.3). The system flags excessive related-party transactions and evaluates whether executive compensation rationally correlates with long-term performance metrics. ## Side-Channel Validation and Succession Planning ### External Sentiment Analysis The **side-channel validation** layer aggregates publicly available sentiment from employee review platforms, app store ratings, and industry forums (sections 6.1‑6.3). This data serves as a reality check against polished public communications, identifying cultural red flags or operational excellence not visible in SEC filings. ### CEO Succession Scenario Testing The skill explicitly tests organizational resilience by asking: "What happens if the CEO leaves tomorrow?" (section 7). This analysis evaluates team depth, documented succession plans, and whether competitive advantage resides in the individual leader or the institutional organization. ## Report Generation and Scoring Methodology ### Structured Output Format All findings compile into a structured markdown report at `reports/{公司名}-management-{YYYYMMDD}.md` (sections 8‑9). The report contains dedicated sections for key personnel profiles, integrity analysis, ability assessment, governance review, side-validation summary, and the final composite calculation. ### Weighted Composite Scoring The final management quality score follows a fixed weighting scheme: - **Integrity**: 35% - **Ability**: 25% - **Capital-allocation**: 25% - **Governance**: 15% This quantified output enables direct comparison across portfolio candidates and provides an audit trail for investment decisions. ## Invoking the Management Deep Dive Users trigger the analysis through the command interface: ```text # Analyze a listed company /management-deep-dive 美团 # Analyze a specific executive within a company /management-deep-dive 王兴 美团 # Analyze an unlisted private firm /management-deep-dive SpaceX ``` In Codex environments, the equivalent syntax uses Chinese keywords: ```text 使用 management-deep-dive 研究 美团 使用 management-deep-dive “王兴 美团” 使用 management-deep-dive SpaceX ``` ## Summary - **AI Berkshire** implements management analysis as a codified skill (`/management-deep-dive`) within the `xbtlin/ai-berkshire` repository, defined in [`skills/management-deep-dive.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/management-deep-dive.md) and auto-generated for Codex at [`codex-skills/management-deep-dive/SKILL.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-skills/management-deep-dive/SKILL.md). - The architecture employs **four parallel agents** to simultaneously gather data on CEO statements, capital allocation, governance, and side-channel reputation. - **Integrity scoring** relies on a three-year commitment fulfillment rate, while **ability scoring** measures strategic foresight against execution track records. - **Capital allocation** decisions are validated using `tools/financial_rigor.py verify-valuation` to ensure financial discipline. - Output generates as a structured markdown report with a weighted composite score (Integrity 35%, Ability 25%, Capital-allocation 25%, Governance 15%). ## Frequently Asked Questions ### How do I invoke the management deep dive skill in AI Berkshire? Execute `/management-deep-dive` followed by the company name or specific executive identifier. For example, `/management-deep-dive 美团` initiates a full analysis of Meituan’s management team, while `/management-deep-dive 王兴 美团` focuses specifically on CEO Wang Xing within the Meituan context. In Codex environments, use the syntax `使用 management-deep-dive 研究 [company]`. ### What weighting does AI Berkshire use for its management composite score? The final score applies a fixed weighting scheme: **Integrity** contributes 35%, **Ability** contributes 25%, **Capital-allocation** contributes 25%, and **Governance** contributes 15%. These weights reflect the Buffett-Munger emphasis on character and capital discipline over raw operational competence. ### Which tool validates capital allocation decisions in AI Berkshire? The system invokes **[`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py)** with the `verify-valuation` function to cross-check M&A premiums, buy-back timing, and dividend sustainability against intrinsic value calculations. This ensures capital deployment decisions meet quantitative financial rigor standards before receiving positive scoring. ### How does AI Berkshire verify management integrity beyond public statements? Beyond formal disclosures, the skill employs **Agent 4** to conduct side-channel validation through employee review platforms (Glassdoor), customer feedback channels (App Store, Zhihu), and industry reputation data. This triangulates public claims against ground-level sentiment, while the commitment fulfillment rate quantifies the gap between three years of public promises and actual operational delivery.