## Sources

1. [AI Sovereignty and the Architecture of Participation](https://www.oreilly.com/radar/ai-sovereignty-and-the-architecture-of-participation/)
2. [SaaS Is Not Dead Yet](https://www.oreilly.com/radar/saas-is-not-dead-yet/)

---

### **AI Sovereignty and the Architecture of Participation – Tim O’Reilly**

**Main Arguments**
*   **The Rise of AI Sovereignty:** Similar to how nations like Brazil seek "medical sovereignty" to avoid dependence on foreign vaccine supply chains, countries are increasingly pursuing **sovereign AI** [1, 2]. This movement is driven by a desire to avoid relying on a handful of dominant American or Chinese companies for a foundational technology [2].
*   **The Failure of Free Trade’s Original Vision:** The original "architecture of participation" for free trade—which promised mutual benefit and shared exchange—has instead devolved into a system of **platform dominance** [3, 4]. Value is often extracted to central hubs, leaving smaller players in a position similar to small businesses on Amazon or developers in an app store [4, 5].
*   **Federation Over Decoupling:** Seeking sovereignty is not a retreat from the world or a simple "decoupling" [6]. Instead, it is a push for **federation**, where nodes remain connected and interoperable but are not wholly at the mercy of any single supplier [6].
*   **The Critical Role of Infrastructure:** True sovereignty cannot exist at the model layer alone [7]. Because AI is physically place-bound, it requires **massive physical infrastructure**, including data centers, chips, and electrical grids [8, 9]. Without controlling this infrastructure, any gains made through open-source models will eventually be recaptured by the few companies that own the servers [7].

**Key Takeaways**
*   **Open weights are not enough:** While open-source models (like Llama or Mistral) are essential "tools of sovereignty," they are only part of the solution; running them at scale requires physical and institutional capacity [9-11].
*   **Industrial policy is necessary:** AI sovereignty requires public procurement and capacity-building, as the market alone will likely prioritize centralizing value [7].
*   **The need for a new economic pattern:** The choices made today regarding AI organization at the model, protocol, and infrastructure layers will define economic activity for a generation [12]. A federated approach could provide a better pattern for global trade [12, 13].

**Important Details**
*   **Enshittification and Capture:** The source notes a recurring cycle where open architectures foster innovation, only for winners to emerge, consolidate power, and eventually degrade the system for profit (a process dubbed "enshittification" by Cory Doctorow) [14].
*   **Infrastructure Initiatives:** Several regions are already making "infrastructure plays," such as the **EU’s AI Gigafactories program**, India’s IndiaAI mission, and compute buildouts in the Gulf [9].
*   **The Intelligence Grid:** O'Reilly proposes an **interoperable intelligence grid** that seamlessly manages frontier models in large data centers alongside local models, balancing cost, privacy, and specialized knowledge [15].

***

### **SaaS Is Not Dead Yet – Mike Loukides**

**Main Arguments**
*   **The Limits of Individualized AI:** While AI agents allow individuals to "vibe-code" custom tools using natural language, this approach currently fails to address the **team-based nature of work** [16, 17]. When every employee creates their own siloed version of a tool (e.g., a personalized CRM), data interoperability and company-wide reporting become impossible [17, 18].
*   **SaaS as a Collaborative Resource:** The core value of Software as a Service (SaaS) lies in its ability to facilitate **sharing and collaboration** across departments and corporations [19]. It acts as a "system of record" that ensures everyone in an organization is working with the same metrics and data schemas [18, 20].
*   **Shifting Customers from Humans to Agents:** The threat to SaaS is not that it will disappear, but that it must radically adapt [21]. Instead of focusing primarily on human-centric dashboards and UIs, SaaS companies must pivot to providing **reliable APIs designed for agents** [20, 22].

**Key Takeaways**
*   **From Dashboards to Raw Data:** Humans need the data compression of a dashboard to understand information, but **agents require raw, structured data**, relationship graphs, and machine-readable playbooks to be effective [17, 20].
*   **The Value of the Bundle:** SaaS provides a "bundle" of features that users may not realize they need until a specific problem arises, offering utility beyond immediate individual requirements [18].
*   **Adaptation or Extinction:** Established companies like Salesforce and Google have momentum, but they risk being blindsided if they do not move quickly to build the infrastructure needed for next-generation, agent-centric SaaS [21, 22].

**Important Details**
*   **Requirements for Corporate AI Skills:** For AI "skills" to be useful across a company, several software engineering disciplines must be applied to them, including **versioning, testing, security, and requirement resolution** [23]. 
*   **The Friction of "Easy" Solutions:** Proponents of agentic programming often underestimate the friction involved in non-technical users managing tools via Git or other developer-centric workflows [19].
*   **Structured State for Agents:** Future SaaS APIs will need to provide "structured state, task objectives, relationship graphs, permissioned memory," and other elements that allow agents to update intent accurately [20].