From useful reading to maintained knowledge.

Now — a public knowledge layer

Browse and search released knowledge, inspect sources and limits, follow new reading through RSS, and download public collections as text or JSON for use with an AI assistant. The website and machine-readable files use the same selected knowledge edition.

Next — turn reading into work

• Build task-oriented Kits that start from a real situation and end in a usable artifact such as a decision brief, test plan or review checklist.
• Shape compact cards for the moment after an explanation: one useful distinction, one practical move, one check and one important limit.
• Publish worked cases that show the full path from messy input to method, output and check.
• Connect longreads to relevant methods and collections so discovery ends in something the reader can use.
• Make meaningful changes easier to follow: new limits, corrected claims, stronger evidence and useful worked cases rather than artificial weekly novelty.

Agent layer — current knowledge on demand

• Prototype a read-only MCP or equivalent interface only after the static AI workflow proves useful.
• Expose stable IDs, editions, applicability, limits and source links so an agent can retrieve a method without stripping away its boundaries.
• Add freshness and change checks so a cached method can be compared with the current public edition before consequential reuse.
• Explore adapters for Telegram, WhatsApp and other assistant surfaces only when delivery, privacy and user-control boundaries are clear.
• Evaluate selective activation: whether giving an agent the right maintained method improves a real task compared with a generic assistant using the same base context.

Later — only if real use proves the need

• Private team methods, templates and organization-specific overlays with clear ownership and review boundaries.
• Maintained exports and change notifications for products or agent workflows that depend on version-pinned knowledge.
• More languages and localized collections when the editorial process can preserve meaning, terminology and evidence.
• Commercial services around maintenance and delivery, rather than a paywall around the same public cards.

What is deliberately not on the roadmap

No race to publish thousands of weak pages. No forced daily streak just to increase engagement. No claim that an AI agent replaces expertise. No silent access to accounts or autonomous actions simply because knowledge can be machine-readable. And no promise that a protocol, API or paid feature ships before a real use case justifies the maintenance cost.