## Sources

1. [Choosing the Right Graph](https://jessicatalisman.substack.com/p/choosing-the-right-graph)
2. [AI Won’t Save the Kingdoms We Built](https://cutlefish.substack.com/p/ai-wont-save-the-kingdoms-we-built)
3. [The Insanity of Data Education](https://joereis.substack.com/p/the-insanity-of-data-education)
4. [You can't just blame data producers for poor data quality](https://andrewrjones.substack.com/p/you-cant-just-blame-data-producers)
5. [TBM 405: Hope, Context, and Control](https://cutlefish.substack.com/p/tbm-405-hope-context-and-control)
6. [Live with Joe Reis - February AMA. Random Rants About Random Questions.](https://joereis.substack.com/p/live-with-joe-reis-february-ama-random)

---

### **AI Won't Save the Kingdoms We Built - John Cutler**

*   **Main Arguments**:
    *   Many leaders who benefited from the "ZIRP" (Zero Interest Rate Policy) era and built complex organizational "fiefdoms" are now looking to **AI to solve the very bureaucracy and coordination problems they helped create** [1].
    *   The "General Manager" (GM) fiefdom model was based on the assumptions that **autonomy creates speed** and business accountability creates focus, but it led to a "geometric problem" where coordination burdens eventually collapsed the system [2, 3].
    *   While AI can reduce information friction by summarizing and retrieving context, it **cannot fix underlying organizational issues** such as incentive misalignment, contested ownership, or political maneuvers [4].

*   **Key Takeaways**:
    *   AI can either be a "cheap context layer" that aids a coherent operating model or it can become **"coordination theater,"** creating the appearance of working together while allowing individuals to avoid actual collaboration through automated agents [5, 6].
    *   The real opportunity for organizations is to make the **hard structural choices** they previously avoided: establishing clear end-to-end ownership, matching decision rights to responsibility, and prioritizing tradeoffs [5].
    *   AI acts as an amplifier; it can supercharge avoidance and fiefdom-preservation or supercharge adaptation to a flatter, clearer reality, depending on the organization's honesty about its problems [7].

*   **Important Details**:
    *   The "coordination tax" is currently being pushed to the front lines by **"async abundance"**—a deluge of AI-generated documentation and messages that eat away at employee focus [8].
    *   Good collaboration is identified as one of the highest-leverage activities, potentially saving millions of dollars in wasted effort [7].

---

### **Choosing the Right Graph - Jessica Talisman, MLS**

*   **Main Arguments**:
    *   The knowledge graph category consists of two distinct traditions: **RDF/OWL** (descending from formal logic and library science) and **Labeled Property Graphs (LPGs)** (descending from graph theory and operational data needs) [9].
    *   Choosing the wrong model is a **costly architectural mistake** because their data models, semantics, and engineering economics differ significantly [10].

*   **Key Takeaways**:
    *   **Use RDF/OWL** when the dominant problems are **meaning, data integration across organizational boundaries, and formal reasoning** [11]. It is the standard for FAIR (Findable, Accessible, Interoperable, Reusable) data publishing [12, 13].
    *   **Use LPGs** when the focus is on **operational performance, deep real-time traversals** (e.g., fraud detection, recommendations), and developer ease of adoption [11, 14, 15].
    *   The release of **RDF 1.2 in 2026** is a major milestone that allows for native edge annotations, narrowing the functional gap between RDF and LPGs regarding edge attributes [10, 16, 17].

*   **Important Details**:
    *   **RDF 1.2** introduces the "triple term," allowing a triple to be the object of another triple, which facilitates statement-level metadata (provenance, confidence) without complex workarounds [16, 18].
    *   **ISO/IEC 39075:2024 (GQL)** is the first new ISO database query language standard since 1987, providing a standardized language for property graphs [19, 20].
    *   Native LPG engines like Neo4j utilize **"index-free adjacency,"** which allows deep traversals to be significantly faster than relational joins or triple-store queries [21, 22].

---

### **Live with Joe Reis - February AMA - Joe Reis**

*   **Main Arguments**:
    *   Findings from the "2026 State of Data Engineering Survey" indicate a widespread **organizational crisis in data modeling**, with **89% of respondents** reporting at least one major pain point [23].

*   **Key Takeaways**:
    *   The **pressure to move fast** and a **lack of clear ownership** are the two primary causes of failure in data modeling practices [23, 24].
    *   The high failure rate suggests that the current industry approach to implementing and teaching data modeling is fundamentally struggling to meet the needs of engineers [23].

---

### **TBM 405: Hope, Context, and Control - John Cutler**

*   **Main Arguments**:
    *   There is a persistent tension in product development between **"Legibility"** (simplifying reality for institutional control) and **"Mētis"** (locally grounded, tacit knowledge used to navigate complex situations) [25].
    *   "Systems of record" often provide legibility at the cost of losing the real-world context where the most important decision-making resides [25, 26].
    *   Leadership's fixation on **"Rollup Systems"**—neat, tidy reports where everything adds up—often ignores the emergent, complex nature of sociotechnical systems [27].

*   **Key Takeaways**:
    *   AI could potentially be a **catalyst for humanistic collaboration**, expanding the boundaries of "mētis" by helping teams share context and find expertise more efficiently [28, 29].
    *   Pessimistically, AI could be used to **supercharge techno-authoritarian control**, integrating the "hive mind" of creative workers into a centralized management "brain" to maximize legibility and top-down management [30, 31].

*   **Important Details**:
    *   A cited study found that AI tools actually **increased human interaction**, with employees gaining significantly more collaboration and knowledge-sharing ties [28].
    *   In the AI-augmented environment, **specialists became "knowledge magnets,"** as the tool helped others find and access their deep expertise faster [28].

---

### **The Insanity of Data Education - Joe Reis**

*   **Main Arguments**:
    *   The data industry is in a state of **"insanity,"** repeating the same educational and management mistakes for 40 years while practitioners continue to struggle [32, 33].
    *   Data education is often **too dry, preachy, and detached** from the daily dysfunctions (like extreme time pressure) that engineers face [34].
    *   Practitioners are not lazy or unintelligent; they simply **lack the agency** and time to implement "best practices" defined in academic textbooks [33, 35].

*   **Key Takeaways**:
    *   The industry needs a "massive reset": **teach pragmatic "building blocks"** instead of dogmatic "religion" [36, 37].
    *   **Education is useless without ownership.** Leaders must establish clear boundaries and give teams the "air cover" to pause and build things correctly [37].
    *   Material must be made **"absurdly digestible"** to compete for the limited attention of busy professionals [36, 37].

*   **Important Details**:
    *   The survey cited (1,100+ professionals) shows **59% suffer from pressure to move fast** and 51% suffer from a lack of clear ownership [38, 39].
    *   Reis advocates for "Mixed Model Arts," a more flexible, pragmatic approach to data modeling that meets practitioners where they are [40, 41].

---

### **You can't just blame data producers for poor data quality - Andrew Jones**

*   **Main Arguments**:
    *   Data teams frequently blame "upstream" data producers (software engineers) for poor data quality, but these producers are often **unaware of what "good data" looks like** for downstream analytical needs [42, 43].
    *   Software engineers have their own deadlines and shouldn't be expected to guess the reporting requirements of the data team [43].

*   **Key Takeaways**:
    *   Data teams must transition from "blaming" to **"supporting"** producers by providing the tools and guidance necessary to reduce their cognitive load [44, 45].
    *   Effective support includes: **clear guidelines** for measurement, **shared libraries** for data collection with local validation, **sample schemas**, and **AI "agent skills"** to help design data for specific features [45].
    *   Providing this support at the exact time it is needed allows software engineers to easily publish high-quality data that the rest of the organization can depend on [46].