447: ‘A Sociopathic Father’, With Adam Lisagor

Episode Summary

Executive Summary: Adam Lisagor discusses how he built Hovercraft, a Mac app for live screen-sharing overlays, using AI coding tools and his own “creative directing” approach. The conversation broadens into his broader AI-driven workflow, the Portable Expert/Lysagor system, Sandwich Vision’s immersive media work, and the future of Vision Pro/VisionOS as a platform for layered, experiential content rather than traditional apps.

Main Topics: Hovercraft: a focused Mac app for live compositing (Priority: 5/5): Hovercraft lets users share a webcam feed with layered content such as images, PDFs, drawings, and screen elements during calls. Lisagor emphasizes its simplicity, low-level polish, and practical utility for presenters and creators. AI coding as creative direction (Priority: 5/5): Lisagor argues that building with Cursor/Claude/Opus is less like traditional programming and more like directing a production: iterating with context, emotional feedback, and human-style collaboration to get better results. Portable Expert and Lysagor as institutional memory (Priority: 4/5): He describes extracting months of Cursor transcripts into a personal expert system that models his own development style, then exposing it via MCP so it can assist future projects and preserve his working context. Sandwich Vision, Theater, and immersive media distribution (Priority: 4/5): The company’s Vision Pro apps evolved from novelty entertainment toward distributing immersive films and live events, including Apple Immersive Video, with an emphasis on layered media and access to scarce content. Vision Pro’s current role and future potential (Priority: 4/5): Lisagor sees Vision Pro as early, expensive, and not yet mass-market, but still strategically important. He believes it will matter through immersive experiences, specialized uses, and a future lower-cost hardware generation. The value of AI tools for non-programmers (Priority: 4/5): He argues that AI coding unlocks software creation for people with taste and ideas but no formal programming background, and that the result can be better than some traditional development workflows. Digressions on software taste, legacy apps, and tech transitions (Priority: 2/5): The conversation also wanders through note apps, Yojimbo, AppleScript, contacts UX, and old/new software paradigms, reinforcing the theme that tools are becoming more accessible and customizable.

Key Arguments: AI coding tools are best used as collaborative creative partners, not autonomous code factories. Context continuity matters: saving rules, memory, and transcripts dramatically improves AI output over repeated prompting. Hovercraft works because it solves a narrow, real workflow problem extremely well rather than trying to be a sprawling platform. The same skills used in directing films or commercials apply to directing AI-assisted software development. Non-programmers can now make meaningful software if they bring taste, product judgment, and persistence. The Vision Pro platform is still early, but its strategic future may be more about immersive media and special-purpose workflows than mainstream app economics. Immersive computing is in an awkward but important transition phase similar to early desktop publishing or early digital video. Building internal tools for a creative studio can turn a services business into a scalable software-enabled operation.

Data Points: Hovercraft download size: 4.4 MB - Lisagor highlights the app’s small size as a sign it is a focused, native Mac app rather than bloated Electron software. Hovercraft price: $19 for one license; $29 for two - He mentions a one-time purchase model with no subscription. Trial length: 7 days - Hovercraft offers a free trial before purchase. Daily token spend: $150–$200 per day - Lisagor says he spends this amount on AI tokens for building tools and internal systems. Vision Pro screen supply constraint: 900,000 screens/year - He cites the original Sony screen bottleneck as a reason Apple could not scale Vision Pro dramatically faster. Approximate Vision Pro hardware ceiling mentioned: 450,000 units/year - Derived from the two screens per headset constraint discussed in the conversation. Yojimbo items analyzed: 9,500 items - He used AppleScript/ChatGPT to chart his Yojimbo library by year. Birdhouse era limit: 140 characters - He references the old Twitter character limit as the reason Birdhouse helped sharpen jokes. Sandwich Vision/team size: 12 people - He notes Sandwich Video remains an operating concern with a team of twelve. Portrait of AI iteration pace: Two weeks - He says Hovercraft was shipped in two weeks, aided by his AI expert system. Model experience cited: Sonnet 4, Sonnet 4.5, Opus 4.5 - He describes switching models as they improved and relying on Opus for harder work.

Pivotal Quotes: "The idea, I fell in love with the Vision OS. Language, visual language." — Adam Lisagor: He explains the origin of his obsession with Vision Pro and why it shaped later app ideas. "I’ve been creative directing software using other software." — Adam Lisagor: His summary of how he approaches AI coding tools and product creation. "Share yourself as a human, as a builder, and you will get that reflected back to you while you’re making the work." — Adam Lisagor: His central advice on getting better outcomes from AI coding agents.

Implications: For creators and indie developers, AI now makes small, highly polished apps viable without traditional engineering teams. For Apple, Vision Pro’s future likely depends on immersive content, specialized workflows, and better hardware, not just more apps.