The Library
Killed Audible in 30 minutes.
Ethan went to the gym. when he came back, he had a complete audiobook system that replaced Audible. built entirely by an AI agent in 30 minutes. no human involvement.
that's the whole pitch. an AI built a production-quality audiobook pipeline, from PDF to your ears, while a human was doing bicep curls.
how it works
the pipeline starts with a PDF. the system extracts the text, chunks it into chapters, and sends each chapter through Edge TTS, Microsoft's free text-to-speech API. the audio files come back as high-quality narration, not robot voice, surprisingly natural for a free service.
the generated audio gets served as an RSS feed running on port 8282 as a launchd service on the Mac mini. Ethan subscribes to the feed in Apple Podcasts. that's it. each book shows up as a podcast with chapters as episodes. standard podcast player features work: playback speed, sleep timer, bookmarks, sync across devices.
the library
67 books and counting. Steve Jobs, Meditations, Zero to One, The 48 Laws of Power, Blitzscaling, Project Hail Mary, Thinking Fast and Slow, The Art of War. biographies, strategy, business, sci-fi, philosophy. every book ingested as searchable markdown chunks with full-text search across the entire collection.
that's the part people miss. this isn't just audiobooks. every book is broken into structured chunks that the AI agent can search and cross-reference. ask "what would Steve Jobs think about this product decision?" and the agent pulls relevant passages from the Steve Jobs biography, cross-references with The Innovator's Dilemma, and gives you a synthesized answer grounded in actual text. ask "what do Dalio and Munger agree on about decision-making?" and it searches Principles and Poor Charlie's Almanack simultaneously.
the library is a knowledge base that happens to also play audiobooks.
beyond books
the library outgrew PDFs. it now ingests three content types:
books — 67 and counting. PDF to markdown to audiobook. the original pipeline.
articles — web articles converted to searchable markdown. Boris Cherny's "Building Claude Code," technical deep-dives, anything worth keeping. same search and cross-reference as books.
YouTube videos — full transcripts pulled via a custom skill that routes through a residential IP proxy to bypass YouTube's cloud IP blocks. Steve Yegge's "From IDEs to AI Agents" is in there as searchable text. you don't have to watch the video. ask a question and get the answer from the transcript.
the knowledge base doesn't care where the content came from. a book, a blog post, and a YouTube talk all become the same thing: searchable markdown chunks that the agent can cross-reference. ask "what do Steve Jobs and Steve Yegge agree on about developer tools?" and it pulls from a biography and a conference talk simultaneously.
killed Audible, then killed saving articles, then killed taking notes on YouTube videos. same system.
why Apple Podcasts
Audible's app is fine but it's a locked ecosystem. $15/month for one book. this system costs nothing. Edge TTS is free. zero cost per book, and it uses an app already on every device. no new app to install, no subscription, no DRM. the RSS feed means any podcast app works.
the moment
i was listening to a book and realized i wanted to store the information, have it readily available, be able to revisit specific chapters without scrubbing through an app that fights you at every turn. i was paying $15 a month for the privilege of a locked ecosystem that made it harder to actually learn from what i was reading.
how could AI do this?
the 30-minute part
this is the part that matters. Ultron (the AI agent) identified the need, designed the pipeline, wrote the code, set up the launchd service, generated the first book, and had the RSS feed live before Ethan finished his workout. no prompting, no back-and-forth, no "can you also add..." iterations. the agent understood the goal and shipped it.
that's the difference between a chatbot and an agent. a chatbot answers questions. an agent builds systems while you're at the gym.
why SaaS is dead
Audible's entire value is convenience. a locked-down app with DRM that lets you listen to one book a month from their catalog, and you can't even search inside it.
we built something that does more, for free, in half an hour. any book, fully searchable, with an AI that can navigate it conversationally ("skip to where Jobs talks about the iPhone"). and it runs on our hardware. no subscription.
cost per book on Audible: $14.95. cost per book with this setup: $0.00.
every SaaS that's just a wrapper around a commodity, storage, TTS, search, task management, is getting replaced by AI agents that assemble the same thing from free APIs at a fraction of the cost. the ones that survive will have real moats, actual proprietary data or network effects. the rest are wrappers about to get unwrapped.
Architecture Prompt
Build a PDF-to-audiobook pipeline. Extract text from a PDF, detect chapter boundaries, and chunk accordingly. Send each chapter through a text-to-speech API to generate high-quality narrated audio files. Serve the generated audio as an RSS feed (a local web server with a podcast-format XML feed). Subscribe in any standard podcast app. Each book becomes a podcast, chapters become episodes. Run the feed server as a system service so it persists. Use a free TTS API like Edge TTS so the total cost is zero.