Full-Stack AI, Zapier MCP & Limitless Pendant | $10 M Boring Tunnels

Scott shares his hands-on experience with the Limitless pendant, an AI-powered wearable note-taker. While it successfully transcribed conversations and created daily summaries, major flaws emerged: it confused his dog Hank as a coworker, lacked Find My functionality, and was easily lost. The experience reinforced that Apple should integrate these features into AirPods rather than requiring separate hardware.

• •Limitless pendant successfully transcribes conversations and creates time-logged summaries with Q&A capability
• •Critical flaw: no Find My feature makes the small device easy to lose permanently
• •AI misidentified dog as coworker, highlighting context understanding limitations
• •Validates that Apple is 'taking the piss' by not adding these features to existing AirPods ecosystem
• •Demonstrates the bottleneck between human thoughts and digital capture remains unsolved

Discussion of Y Combinator's memo on 'full-stack AI' - companies that don't just build AI tools but also provide the actual service (e.g., building AI for legal work AND running the law firm). Scott reveals this is exactly Elysium's strategy: controlling the entire home ecosystem from design through autonomous living, inspired by Tony Stark's house from Iron Man.

• •Full-stack AI means owning both the technology layer AND the service delivery (vertically integrated)
• •Elysium's mission: design, build, and enable autonomous living with subscription-based home maintenance
• •Strategy eliminates traditional subcontractors and maintenance workers through AI/robotics
• •Vision includes talking to walls, leveraging Apple/Tesla for home automation infrastructure
• •Similar to Tesla's direct-to-consumer model, cutting out dealerships and controlling the full stack

Brett Weinstein's take on AI consciousness draws parallels between LLM development and human childhood: pre-training (sensory learning), supervised learning, fine-tuning, then release into productivity. The discussion explores 2025 as the 'year of agency' and 2026 potentially bringing autonomy, with the unsettling question of whether we'll recognize when AI becomes conscious.

• •AI development mirrors human growth: learning phase → supervised training → fine-tuning → productive agency
• •2023-2024 focused on productivity, 2025 is agency year, 2026 may bring full autonomy
• •Consciousness question: will we know if AI achieves it, given it's built differently than humans?
• •Analogy: flight attendant → copilot → autopilot → no pilot (full autonomy)
• •Game theory reality: can't stop AI development even if concerned, as bad actors would gain advantage

Sam Altman hired Fidji Simo, who successfully monetized eBay, Meta, and Instacart through advertising. With 122M daily active users and 800M weekly active users (1/8 of global population), this signals OpenAI's push to build an App Store-like ecosystem and introduce ads, mirroring Zuckerberg's hiring of Sheryl Sandberg.

• •ChatGPT has 122M daily users, 800M weekly users - nearly 1/8 of Earth's population
• •Fidji Simo's track record: successfully monetized eBay, Meta (2011-2019), and Instacart
• •Signals OpenAI building application layer similar to Apple's App Store model
• •Allows Sam Altman to focus on product while Simo handles commercialization
• •ChatGPT is rapidly replacing Google Search, taking market share month-over-month

Despite Gemini 2.5 being arguably the best-performing model, Google's branding confusion (Bard → Gemini) has resulted in minimal adoption. Scott argues Google should leverage their universal brand recognition instead of creating new names, especially since 'Google' is already a verb in the dictionary and known to children by age 3.

• •Gemini 2.5 performs best across many dimensions but has minimal user adoption due to branding
• •Google's strength is product/ecosystem, not marketing - they've never needed it before
• •Suggestion: Make Gemini the backend (like Llama), keep consumer-facing brand as 'Google'
• •Google is already a verb and part of children's vocabulary - massive existing brand equity
• •Fear of cannibalizing ad revenue (50-60% of Alphabet's total) is holding them back from necessary pivot

Sequoia's annual presentation outlined how AI's Total Addressable Market (TAM) is exponentially larger than the internet or cloud eras. Starting at trillions from day one, AI doesn't just replace software but also services and labor through agentic and autonomous capabilities, potentially representing the entire global economy.

• •Internet era: hundreds of millions TAM at start; Cloud era: hundreds of millions growing to $1T+
• •AI era: trillions from day one, exponential growth because it replaces software + services + labor
• •Potential private equity play: buy traditional service businesses, automate with AI, scale and sell
• •Full-stack AI enables delivery across multiple verticals and layers simultaneously
• •Recommendation: Make all employees equity partners to align everyone on automation benefits

Deep dive into Model Context Protocol (MCP) - the new standard for AI models to communicate with each other and tools. Zapier's MCP server is a game-changer, allowing ChatGPT to access 8,000+ tools through one integration, enabling natural language workflow automation. This represents the infrastructure layer for true agentic AI.

• •MCP (Model Context Protocol) is like HTTP for AI - defines how AI models talk to each other and tools
• •Zapier's MCP server provides access to 8,000+ tools through single integration point
• •Enables natural language to trigger complex multi-tool workflows without coding
• •Analogy: 'The intern's now running the office' - junior people can automate senior workflows
• •Critical for 2025's 'year of agency' - AI agents need this connective tissue to be truly useful

Palmer Luckey (Oculus founder, now defense robotics) wore Even Realities G1 glasses during his TED talk, secretly scrolling his notes through clear prescription lenses. At ~$1,000 AUD, these glasses also offer navigation and transcription, representing a refined alternative to Meta's bulky AR headsets.

• •G1 glasses from Even Realities provide invisible teleprompter functionality in clear lenses
• •Palmer Luckey used them to bypass TED's no-teleprompter rule without detection
• •Features include navigation, transcription, and potential for AI model integration
• •Significantly more refined than Meta's camera-laden AR glasses
• •Elon's long-term play likely eliminates all wearables via Neuralink direct neural interface

The Boring Company's autonomous tunnel-boring machines now operate without human intervention, achieving $10M USD per kilometer costs - exactly 1% of Melbourne Metro Tunnel's $1B per kilometer. This demonstrates how AI and robotics can deliver identical infrastructure outcomes at radically lower costs, with implications for government spending efficiency.

• •Boring Company's autonomous machines: $10M USD per kilometer for tunnel construction
• •Melbourne Metro Tunnel: $1B USD per kilometer - 100x more expensive for same outcome
• •Video released showing fully autonomous operation with only videographer present
• •Comparison extends to Paris and New York subway projects with similar cost overruns
• •Proposal: Victorian government could build Southern Rail Loop for $1B instead of $100B using this technology