A deep technical breakdown of Claude Code's conversation compaction system — the three-tier architecture, the separate model call for summarization, the nine-section structured prompt, and the in-memory swap mechanics. We cover the trigger conditions, what survives compaction versus what gets lost, the reconstruction phase that re-reads files, and the critical asymmetry: compaction preserves what to do next but systematically drops why we did what we did. Includes token savings benchmarks, power user strategies like CLAUDE.md as persistent storage, and the philosophical question of whether a post-compaction agent is the same agent or a new instance reading a briefing.

Episode #067026 — open it directly at myweirdprompts.com/067026

What does it take to automate shopping when every store has its own checkout, shipping varies by neighborhood, and sites break without warning? This episode explores the architecture and tradeoffs of building an AI agent for Israel's chaotic e-commerce landscape.

Episode #899167 — open it directly at myweirdprompts.com/899167

When building an AI-native backend with MCP server integration, your ORM choice matters more than ever. This episode compares Drizzle and Prisma across key dimensions: AI-friendliness for code generation, migration workflows, schema design philosophy, and guardrails for catching AI-generated errors before they hit production. We explore why Drizzle currently offers the cleanest path for MCP-compatible backends, why Prisma Next is betting on "agent-centric development" with machine-readable error codes and compile-time guardrails, and why neither ORM supports rollbacks by design. Plus, insights from a recent survey of migration practices across 40+ major open-source projects.

Episode #413727 — open it directly at myweirdprompts.com/413727

Are graph databases the future of mainstream business software, or are they destined to remain a specialized tool? This episode explores the gap between graph's theoretical advantages for relationship-heavy domains like CRM and ERP and the practical realities of adoption. We examine the recent GQL standardization, the rise of hybrid multi-paradigm architectures where AI agents orchestrate queries across SQL, graph, and vector databases, and the emergence of graph foundation models. Featuring insights from industry players like Neo4j, PuppyGraph, and Memgraph, we break down why the market is moving toward graph as a query layer over relational storage rather than native graph databases — and what it would take for that to change.

Episode #999246 — open it directly at myweirdprompts.com/999246

After a barbecue smoke triggers a severe asthma attack, one listener asks whether adding Montelukast is worth the risk. This episode explores how different asthma drugs target distinct pathways in the airway cascade—and whether combining them is additive, synergistic, or just piling on pills.

Episode #798663 — open it directly at myweirdprompts.com/798663

The cost of renting GPUs from hyperscalers like AWS can be three to six times higher than newer AI cloud providers. But price isn't everything. This episode breaks down the structural reasons for the price gap, the hidden traps like data egress fees and compliance ceilings, and provides a practical decision framework for choosing between Modal, RunPod, Nebius, and Baseten. We explore when developer experience trumps raw cost, the importance of InfiniBand for training, and how to avoid lock-in.

Episode #979068 — open it directly at myweirdprompts.com/979068

What's actually happening behind the eyes of a protection detail? This episode breaks down "Left of Bang"—the Marine Corps-developed methodology for threat detection that has become foundational for Secret Service agents, executive protection teams, and military details. We explore the six domains of observation, how trained officers establish baselines and spot anomalies, and the cognitive architecture that lets them process threats faster than conscious thought. Plus, we examine what the Senate report on the Butler assassination attempt reveals about where observation systems broke down—and what that teaches us about the difference between seeing a threat and acting on it.

Episode #366045 — open it directly at myweirdprompts.com/366045

Ireland calls itself militarily neutral, yet it has been one of Europe's most active states in pursuing legal and diplomatic actions against Israel at the ICJ and EU. This episode examines whether neutrality is tenable when backed by minimal defense capability — Ireland has no fighter jets, no air defense, and only two deployable naval vessels. We explore the historical record: Ireland's "phoney neutrality" during WWII, including the infamous condolence visit to the German legation after Hitler's death, and Switzerland's deeply compromised neutrality accepting Nazi-looted gold and turning away Jewish refugees. We ask where the idea that absolute neutrality is a virtue comes from, contrast it with Judaism's activist approach to moral responsibility, and look at which countries still claim neutrality today — from Austria (a Cold War bargain) to Costa Rica (which abolished its army entirely).

Episode #062704 — open it directly at myweirdprompts.com/062704

The job market is an arms race of AI spam and filters, with 242 applications per opening and 30% ghost jobs. This episode explores a radical alternative: desire-based matching that could break the cycle and make hiring work more like a dating app.

Episode #608096 — open it directly at myweirdprompts.com/608096

BIM cascading changes work like a database — until they don't. This episode explores where the SQL analogy holds, where it fails, and what that reveals about the gap between architectural design and spatial mapping.

Episode #284353 — open it directly at myweirdprompts.com/284353