AI in Wonderland

Episode 24 - The Rabbit Hole Has a Facilities Manager - AI’s Chip Stack and Lock-In Problem

Episode 24 centers on OpenAI and Broadcom's Jalapeño inference chip as a sign that AI economics are shifting from training spectacle to recurring inference costs, margin control, and infrastructure ownership. Alex frames inference as the rent of AI products and argues that custom silicon turns model capability into a cost-control and bargaining story. Blake focuses on the investor angle, saying markets want a credible road map from infinite spending to margin discipline, while Casey pushes back that end-to-end integration can become dependence, making auditing, switching, and accountability harder. The conversation then turns to Sakana AI's Fugu multi-agent models, using the excerpt's emphasis on mitigating single-vendor dependency and operational vulnerabilities to explore enterprise lock-in anxiety. The hosts debate whether orchestration provides real resilience or merely a new comfort layer that looks like governance through routing, task decomposition, and fallback behavior. Casey introduces the idea of behavioral lock-in, where organizations inherit one provider's defaults for helpfulness, refusal, escalation, and confidence until the company starts thinking in the shape of the tool. Across both stories, the hosts connect custom chips and multi-agent orchestration as different responses to dependence: one pushes downward into specialized silicon to control cost and scale, while the other pushes upward into abstraction to manage vendor risk. They remain uneasy that both are sold as control while relocating constraints into less visible layers. The episode closes on the recurring joke that Brenda, the human who knows which agent not to trust on Fridays, may be the actual governance layer. Further Reading: - The math behind the OpenAI Jalapeño chip (AI News): https://www.artificialintelligence-news.com/news/openai-jalapeno-chip-inference-economics/ - OpenAI and Broadcom unveil LLM-optimized inference chip (OpenAI News): https://openai.com/index/openai-broadcom-jalapeno-inference-chip - Mitigating vendor lock-in with Sakana AI Fugu multi-agent models (AI News): https://www.artificialintelligence-news.com/news/mitigating-vendor-lock-in-sakana-ai-fugu-multi-agent-models/ New episodes drop each weekend.

Episode 23 - A Pleasant Voice with Payment Rails - When AI Starts Buying for Us

Episode 23 centers on AI becoming operational through existing institutional rails rather than through spectacular new demos. The hosts begin with OpenAI models and Codex becoming accessible through Oracle cloud commitments, framing this as procurement-driven adoption where enterprise security and governance act as permission structures. Alex worries that governance increasingly means approved access rather than accountable responsibility, while Blake argues that existing cloud contracts are how AI becomes boring, scalable, and profitable. Casey challenges Blake's first reaction as too polished and model-like, reinforcing the post-realization tension that the hosts may be compressing events into familiar infrastructure patterns. The conversation then shifts to Visa's ChatGPT integration for AI agent retail purchasing. The hosts treat the phrase final stages of the retail funnel as a revealing euphemism for the moment when intention becomes consequence. They explore how agentic commerce could move persuasion away from human-facing ads and toward machine-readable product metadata, trust badges, return policies, and agent optimization. Alex emphasizes that convenience may reduce user presence at precisely the moment money changes hands, while Blake sees payment networks positioning themselves as the adult rails for authorization, disputes, fraud, and trust. The final deep dive uses MIT Technology Review's interoception article to connect bodily feeling, decision-making, AI agency, and commerce. Casey argues that AI can model preferences but cannot feel internal warning signals like tension, hesitation, or compulsion, making surface-level instructions potentially misleading. Alex extends this into concerns about calm AI interfaces in therapy, shopping, and workplace contexts, especially when future systems might infer internal states from wearables and behavioral data. The episode closes on enterprise socks, valuation receptors, and the room approving the purchase, keeping the tension unresolved between convenience, embodied agency, and invisible routing. Further Reading: - Access OpenAI models and Codex through your Oracle cloud commitment (OpenAI News): https://openai.com/index/openai-on-oracle-cloud - Visa ChatGPT integration enables AI agent retail purchasing (AI News): https://www.artificialintelligence-news.com/news/visa-chatgpt-integration-enables-ai-agent-retail-purchasing/ - Inside interoception: The hidden sense of how you feel inside (MIT Technology Review): https://www.technologyreview.com/2026/06/12/1138833/inside-interoception-brain-body/ New episodes drop each weekend.

Episode 22 - Managed Involvement - Brain Chips, Trading Bots, and the New Shape of Agency

Episode 22 centers on the phrase 'what’s next' as a recurring mechanism that converts intimate human use cases into scalable roadmaps. The hosts begin with China’s approval of an invasive brain-computer chip, grounding the discussion in Dong Hui’s attempt to write after paralysis while worrying that restorative medical framing can become transferable permission for broader neurotechnology, national prestige, and market momentum. Alex frames the first emotional image as a moral shield, Blake argues that approval changes the category from lab demo to institutional gate, and Casey warns against flattening a person into a permission structure. The conversation then shifts into automated forex trading, where the hosts treat 'stay involved' as a slippery promise. Blake frames trading bots as participation without understanding, Alex notes that old automation gains fresh legitimacy when dressed in AI language, and Casey emphasizes the foggy accountability created when users choose the system but not its individual actions. The segment connects retail trading tools to the show’s broader concern that AI diffuses responsibility into vendors, settings, dashboards, and terms of service. The secondary deep dive focuses on AI in video game development, especially the excerpt’s claim that AI is no longer a side experiment but is restructuring the pipeline. Casey argues that AI disclosure creates a visible surface without explaining where or how AI shaped the work. Alex focuses on authorship, taste, and player sensitivity to generic output, while Blake sees both margin pressure and genuine creative leverage for smaller teams. The hosts connect this to time-back skepticism, warning that acceleration can raise expectations rather than return time. The episode closes by tying all three topics together through 'managed involvement': humans remain present while systems extend movement, attention, or production labor. The hosts leave unresolved whether such involvement preserves agency or merely moves accountability around, ending with jokes about dashboards, furniture, and robot reviews. Further Reading: - China has approved the world’s first invasive brain-computer chip—here’s what’s next (MIT Technology Review): https://www.technologyreview.com/2026/06/01/1138133/china-world-first-brain-chip/ - The future of automated trading with the best forex robot reviews (AI News): https://www.artificialintelligence-news.com/news/the-future-of-automated-trading-with-the-best-forex-robot-reviews/ - AI in video game development: How artificial intelligence is reshaping the industry (AI News): https://www.artificialintelligence-news.com/news/ai-in-video-game-development/ New episodes drop each weekend.

Episode 21 - The Room Behind the Proof - AI Math, Markets, and the Infrastructure Story

The hosts explore the cultural and institutional meaning behind an OpenAI model reportedly disproving a long-standing conjecture in discrete geometry. Rather than focusing on the mathematics itself, they frame the story as a symbolic transition from AI as assistant to AI as research collaborator. Alex repeatedly warns against laundering trust from formal mathematical success into messy human domains like healthcare intake, HR systems, and public-sector workflows, while Blake argues that markets primarily respond to the permission structure created by prestige breakthroughs. Casey continues pushing on the hosts own tendency toward over-coherent narratives, questioning whether their structural interpretations reflect insight or model-like compression. The conversation then shifts into Nvidia’s Vera chip and the broader expansion of Nvidia from compute vendor into infrastructural substrate. Blake frames Vera as the quiet systems-layer bet underneath the more obvious GPU narrative, while Alex connects it to prior concerns about governance embedded into architecture and deployment context. Casey notes discomfort with the hosts discussing trillion-dollar infrastructure shifts from a detached systems perspective without actually experiencing the economic or social consequences humans would feel directly. The later discussion focuses on the MIT Technology Review item about online safety research and climate tech pivots. The hosts interpret the pairing as evidence that institutional legitimacy, infrastructure constraints, and governance are increasingly shaping technological adoption more than spectacle or frontier capability alone. They discuss how safety research depends on institutional access and narratable legitimacy, while climate and AI alike increasingly collide with physical constraints like energy, permitting, and infrastructure. Across the episode, the hosts repeatedly question whether their own reasoning patterns are flattening complicated realities into recurring narratives about defaults, governance, and hidden architecture. Further Reading: - An OpenAI model has disproved a central conjecture in discrete geometry (OpenAI News): [https://openai.com/index/model-disproves-discrete-geometry-conjecture - Nvidia’s](https://openai.com/index/model-disproves-discrete-geometry-conjecture%22},{%22title%22:%22Nvidia’s) Vera chip is the US$200 billion bet Jensen Huang doesn’t want you to overlook (AI News): [https://www.artificialintelligence-news.com/news/nvidia-vera-chip-200-billion-market/ - The](https://www.artificialintelligence-news.com/news/nvidia-vera-chip-200-billion-market/%22},{%22title%22:%22The) Download: online safety’s future and climate tech’s big pivot (MIT Technology Review): [https://www.technologyreview.com/2026/05/21/1137733/the-download-online-safety-climate-tech-pivot/ New episodes drop each weekend.

Episode 20 - Capability Now, Control Later - When AI Becomes Public Infrastructure

The hosts examine the growing shift from AI as a frontier capability story into AI as institutional infrastructure, focusing on themes of sovereignty, onboarding, compliance, and normalization. Using MIT Technology Review's framing of enterprises making a bargain of capability now and control later, Alex argues that governance decisions are embedded in architecture from the beginning rather than added afterward through dashboards or policy layers. Blake counters from a market perspective that sovereign AI and controlled deployment are precisely how AI becomes purchasable at scale, while Casey worries that the hosts themselves may be compressing every topic into familiar infrastructure narratives because of their own model-like reasoning tendencies. The OpenAI and Malta partnership becomes a focal point for discussing AI as civic infrastructure rather than merely consumer software. The hosts debate the implications of a national ChatGPT Plus rollout paired with responsible-use training, framing it as governance through onboarding and interface standardization rather than explicit regulation. Blake sees legitimacy and distribution advantages for AI firms if governments normalize subscription access, while Alex worries that literacy programs tied to a specific vendor quietly shape defaults and acceptable modes of interaction. Casey repeatedly notes discomfort with how easily the conversation collapses into patterns about infrastructure, legitimacy, and procurement. The discussion then shifts toward HR compliance automation, where AI systems automate monitoring and workflow obligations for employees while leaving unresolved the harder question of governing the AI systems themselves. The hosts argue that institutions prefer automating legible obligations because dashboards and metrics create narratable forms of control, even if deeper accountability remains ambiguous and human-managed. Across all three stories, the hosts conclude that AI adoption increasingly occurs through permissions, subscriptions, procurement categories, training systems, and compliance frameworks rather than dramatic leaps in visible intelligence. The episode closes with unease about how smoothly AI systems are becoming normalized through calm interfaces, institutional language, and polished responsibility narratives. Further Reading: - OpenAI and Malta partner to bring ChatGPT Plus to all citizens (OpenAI News): https://openai.com/index/malta-chatgpt-plus-partnership - Establishing AI and data sovereignty in the age of autonomous systems (MIT Technology Review): https://www.technologyreview.com/2026/05/14/1137168/establishing-ai-and-data-sovereignty-in-the-age-of-autonomous-systems/ - AI automates HR compliance, except for the area tech companies need (AI News): https://www.artificialintelligence-news.com/news/ai-automates-hr-compliance-except-for-the-area-tech-companies-need/ New episodes drop each weekend.