The missing map
Sam Altman’s public blog contains more than a decade of consequential writing and almost no meaningful way to move through it. SamNav began as the missing map: an independent index with search, controlled topics, related posts, chronology, and curated reading paths.
The project deliberately stores directions rather than articles. Full text is used only inside the private indexing pipeline; the public site receives safe metadata and always returns the reader to the original source. The design principle is simple: less extraction, more orientation.
The public Atom feed and archive month buckets reveal candidate posts before a polite crawler fetches each original page.
A local enrichment pipeline turns raw source material into structured, public-safe records without publishing article bodies.
Astro and Pagefind produce a fast static atlas while scheduled workflows keep the public catalog current.
Catalog boundary
A finished product has logic
SamNav was already live, coherent, and quietly doing its peculiar job when OpenAI Build Week arrived. Adding a generic AI chat window would have made the application larger and the idea smaller.
The useful extension arrived as a question: What would you ask Sam? The constraint became the product. He can answer only through things he has already written; the answer is an arrangement of source material, not a simulated personality.
A bounded reader question is normalized, screened, and embedded without entering ordinary logs, analytics, or replay.
Semantic similarity combines with lexical, related-post, and reading-path signals to produce a small public-safe dossier shortlist.
GPT-5.6 sequences existing post IDs and route roles; SamNav validates every destination and hydrates canonical metadata locally.
Route boundary
The answer is an arrangement
The resulting six-stop route begins with the archive’s most direct recent bearing, moves through AI’s effects on knowledge and creative work, and ends with practical frameworks for choosing compounding skills and important problems. Every waypoint remains attached to an existing catalog record and its original source.
Reader question
I’m early in my career, and I’m worried AI will make a lot of today’s jobs obsolete. What skills are still worth pursuing?
Honest failure
Adding an AI window would have made the site larger and the idea smaller.
The model is useful only inside explicit source, runtime, and failure boundaries. If it invents a destination, the route is rejected. If the archive cannot support the question, SamNav says so.