## Sources 1. [Booms, Busts, and Builders: Lessons from the High Plains of Wyoming](https://joereis.substack.com/p/booms-busts-and-builders-lessons) 2. [How to Build Your First AI Sales Engine With Claude Code](https://aimaker.substack.com/p/claude-code-ai-sales-agent) 3. [The Professor of Outputmaxxing — Anjney Midha, AMP](https://www.latent.space/p/anj) --- ### **Booms, Busts, and Builders: Lessons from the High Plains of Wyoming** **Author: Joe Reis** This source draws parallels between historical Westward migrations and gold rushes in Wyoming and the current AI boom, warning against the pitfalls of "mania" while advocating for a builder’s mindset. * **Main Arguments:** * **Builder vs. Claim Staker:** In any gold rush, it is rational to participate, but the real winners are those who build "heavy infrastructure" (like stamp mills in the 1800s or massive data centers and frontier models today) rather than those who just "panning for surface gold" by building simple wrappers around existing foundation models [1, 2]. * **The Paradox of Expectations:** Success is not guaranteed even for those who build the infrastructure. If massive market expectations are not met, bubbles burst, as seen with the Carissa Mine in South Pass City [3]. * **Human Nature and Manias:** Human nature is inherently enamored with "the latest shiny thing," and the most dangerous mindset during a mania is believing "this time is different" [4]. * **The Difficulty of Distribution:** Building a product is relatively easy compared to the "hard stuff" of gaining distribution and visibility, which is the primary obstacle for modern startups [5]. * **Key Takeaways and Important Details:** * **Hedge Your Downside:** Reis emphasizes the importance of staying grounded and ensuring that if an AI opportunity fails, you are not left "entirely destitute" [6]. * **Organizational Bottlenecks:** Pulse surveys conducted by Reis suggest that "lack of leadership direction" and "poor requirements" are top bottlenecks for data engineers, far outweighing the impact of legacy systems [7]. * **Historical Context:** South Pass, Wyoming, served as the "Gateway to the Continent" for 500,000 migrants between 1840 and 1869, followed by a gold rush in 1867 that saw the population of South Pass City soar to 2,000 overnight before drying up by 1872 [8-10]. --- ### **How to Build Your First AI Sales Engine With Claude Code** **Authors/Participants: Jonas Braadbaart, Wyndo, and Dheeraj Sharma** This source details a practical implementation of an AI-driven sales system that automates administrative heavy lifting while keeping a "human in the loop" for relationship building. * **Main Arguments:** * **Augmentation Over Full Automation:** AI agents should handle the "heavy admin work" (researching, CRM updates, drafting), but the human must remain the final decision-maker regarding which companies to contact and what messages to send [11-13]. * **The Importance of ICP:** Automation is ineffective without a strictly defined Ideal Customer Profile (ICP). Aimless "blasting out DMs" is avoided by prescreening leads manually or with AI against specific business fit criteria [14, 15]. * **Human Communication Must Stay Human:** Fully automated cold outreach often fails because it sounds robotic; high-effort, human-reviewed communication is necessary to build genuine relationships [13, 16, 17]. * **Key Takeaways and Important Details:** * **The Sales Stack:** The system uses **Apollo** for company lists, **Perplexity** (via a custom MCP server) for deep research, **Attio** as the CRM (leveraging its first-party MCP server), and **Gmail** for drafting outreach [12, 18-20]. * **Efficiency Gains:** With this system, the time spent preparing a lead for outreach dropped from approximately 1–2 hours to **one minute per lead** [21]. * **Specific Results:** In a real-world test, contacting 80 highly targeted companies resulted in one sale, which is considered a solid result for focused cold outreach [13]. * **Operating Costs:** The total monthly cost for this specific setup was estimated at approximately **120 euros**, covering tools like Apollo and LinkedIn All In One [20, 21]. --- ### **The Professor of Outputmaxxing — Anjney Midha, AMP** **Authors/Participants: Anjney Midha and Swyx** This source explores the concept of "outputmaxxing"—the discipline of maximizing the efficiency of existing AI infrastructure—and the vision for a horizontal "compute grid." * **Main Arguments:** * **The Systems Problem:** AI scaling is no longer just about acquiring more GPUs; it is a systems problem involving scheduling, networking, and parallelism. Many frontier training runs currently operate at suboptimal utilization (sub-10% MFU in some cases) [22, 23]. * **Outputmaxxing:** This is the engineering discipline of "making the most of what we have" by doubling down on proven architectures and eliminating waste across the stack [24]. * **Compute as a Grid:** Midha proposes a horizontal "compute grid" (the AMP grid) that acts as an **Independent System Operator (ISO)**, pooling demand and supply across clouds to make "FLOPs flow like megawatts" [25, 26]. * **Responsible Infrastructure:** Data center developers must move from "move fast and break things" to "move fast with responsible infrastructure," which includes gaining community buy-in by sharing economic benefits with locals [27-29]. * **Key Takeaways and Important Details:** * **Utilization Standards:** At Google, 95% node utilization was standard (anything less was an outage), and best-in-class MFU today should be between 60% and 70% [30]. * **Culture as a Moat:** Culture is not just a belief but a "set of actions" (citing Bushido). It is a fragile moat that requires daily maintenance through mission-aligned trade-offs [31]. * **Researcher-CEOs:** Midha argues that top-tier researchers are "star athletes of the mind" who can become great CEOs if they are willing to be confrontational "up and down the stack" rather than just focusing on publishing [32, 33]. * **End-of-Life Prediction:** One of Midha’s long-standing passions is using AI for more precise end-of-life predictions in healthcare to empower patients and reduce the massive taxpayer burden of end-of-life care [34-37]. * **Scale of Ambition:** AMP aims for a base load pool of **1.2 gigawatts** of compute capacity, with a need for roughly six gigawatts of spike capacity over the next four years [38].