Field guide / sub-domain
33
sources in this field
Updated September 5, 2026
Current thesis
The shortest path to orientation.
This synthesis records claims and practices from the cited sources; reported outcomes and product capabilities have not been independently verified.
The 2026 dominant second-brain architecture pairs vault-as-foundation (plaintext markdown) with Claude Code—repo-as-database means knowledge organization directly impacts reasoning quality. Vaults split into raw/ (unmodified sources: articles, transcripts, journals) and wiki/ (agent-maintained layer: cross-linked pages, one concept per file, frontmatter with type/sources/dates). Retrieval avoids embeddings: an 18-line root index enables drill-down like code browsing. Pre-commit hooks lint wikilinks, frontmatter, and orphaned pages; weekly agent passes check contradictions and stale status lines. Volatile facts live on exactly one page; all others link there, preventing duplicates. Agents write and update their own memory autonomously rather than relying on static files. Four independent implementations (DoorDash, Pendo, Google, solo builders) converged on three-layer architecture addressing institutional knowledge loss—47% of companies cite offboarding as their top challenge. Company knowledge layer (/hwiki) applies this architecture at org scale: auto-updating markdown documenting every client, SOP, transcript, and campaign. Have a frontier model audit your 'second brain' by inspecting folder structure, duplicate knowledge, retrieval gaps, ingestion crons, and unused files. Data compounding, not technology, is the moat.
Evidence board
Claims worth carrying forward01GPT-6 Astra is claimed to score 32% higher than 'Fable 5.1' on an unspecified benchmark, positioned by the author as making it 'officially the best model out there' — this is an unverified marketing claim with no benchmark name or methodology given.
02Recommended workflow: feed a new frontier model your full session/project history across multiple agents (Claude, Codex, Hermes) and ask it to surface repeated prompts, manual steps that should become scripts/integrations, repeatable workflows that should become skills, corrections that belong in project instructions, and candidates for cron jobs.
03Recommended workflow: have a frontier model audit your 'second brain' by inspecting folder structure/data schemas, duplicate or conflicting knowledge, retrieval gaps, ingestion cron jobs, how sessions/research get saved, and files/data agents never use.
04For stalled 'evergreen' projects that earlier models couldn't complete, the suggested pattern is: send the frontier model project files, have it run existing tests and trace failed workflows, produce a prioritized change plan with per-change pass/fail checks, then hand the defined execution plan to cheaper models for implementation.
05A 20-person AI-native GTM company reports Claude Code handles 80%+ of execution work, enabling them to move 5x faster than companies twice their size — a concrete productivity benchmark for agent-driven services firms.
06Operational blueprint for agent-native companies: (1) map processes with owners/tools/automations/SOPs, (2) connect 25+ MCPs and CLIs (Slack, HubSpot, Instantly, HeyReach, Apollo, n8n, Notion, GitHub, Supabase) to Claude Code as the execution layer.
07Company knowledge layer pattern (/hwiki): an auto-updating markdown 'company brain' stored in GitHub documenting every client, SOP, transcript, and campaign — mirrors the vault-as-foundation pattern but applied at org/company scale rather than personal.
08Behavior layer via Skills: SOPs are converted into reusable Skills that complete tasks 80%+ of the way autonomously; this team maintains ~20 core skills and 50+ total, organized under a strict root folder structure (CLAUDE.md, /wiki, /clients, /raw-context, /.claude) that requires constant upkeep.
Adjacent fields
Latest evidence
Shann³
A promotional post introducing GPT-6 Astra with a benchmark claim and three concrete workflows for using it: auditing agent session history for automatable patterns, auditing a personal knowledge base/second brain, and reviewing stalled 'evergreen' projects with a plan-then-delegate execution model.
Dan Rosenthal
Dan Rosenthal outlines a 7-step operational blueprint (process mapping, 25+ MCPs/CLIs, an auto-updating wiki, ~50 reusable skills, strict folder structure, team-wide access with guardrails) that lets Claude Code handle 80%+ of execution work, enabling a 20-person team to move 5x faster than larger competitors.
Sebastian Kehle
Sebastian Kehle details a ~900-file Karpathy-style Obsidian vault in a private git repo, split into an unmodified raw/ source layer and an agent-owned wiki/ knowledge layer, retrieved via folder indexes instead of embeddings, with automated maintenance hooks keeping it from rotting.
Linas Beliūnas
A tweet summarizing an Anthropic Member of Technical Staff's talk on building self-improving Claude agents — covering the shift from static memory files to self-written memory, the limits of in-band memory, 'dreaming' review cycles, and production guardrails like versioning and concurrency control.
EP
A workflow for turning a self-hosted Hermes agent into a lifelong knowledge system: instead of relying on capped built-in memory, memory acts as an index pointing to an Obsidian-style markdown vault on a VPS. Six copy-paste prompts build the vault structure, a self-filing skill with contradiction handling, an evidence-card answering method, an inbox-based braindump/import pipeline, and a weekly staleness-check cron job.
Thariq
Anthropic's Thariq details how prompting and context engineering practices have shifted for Claude Opus 5/Fable 5: over 80% of Claude Code's system prompt was removed with no eval loss, as newer models handle judgment, tool design, and progressive disclosure better than rigid rules, repeated examples, or upfront specs.
CyrilXBT
A detailed 6-step system for building an Obsidian vault that auto-captures content via Readwise, Airr, Whisper, and a Telegram bot, routes everything through N8N pipelines, and uses Claude to generate daily briefings and weekly syntheses. The core thesis: knowledge systems fail because they optimize for input, not output—feedback loops (daily briefs, weekly synthesis) are what turn a dead archive into a thinking partner.
Charly Wargnier
Tolaria is a free, open-source (AGPL-3.0) desktop app for Mac and Linux that implements Karpathy's LLM wiki concept — a shared knowledge environment for humans and AI agents. It uses plain markdown, Git-backed vaults, and includes a built-in MCP server for Claude Code integration. Built with Tauri, React, and Rust; 100K+ lines, 85% test coverage, 9.9/10 code health.
Aakash Gupta
Hannah Stulberg (DoorDash PM) built a shared git repo where every team function checks in context—call summaries, decision logs, analytics queries—queryable in natural language via Claude Code. Four independent implementations (DoorDash, Pendo, Google, solo) converged on the same 3-layer architecture, suggesting a generalizable pattern for solving institutional knowledge loss.
Shann³
Comprehensive operator guide for Hermes Agent by Nous Research—an open-source autonomous agent framework with 150K GitHub stars and #1 OpenRouter usage. Covers architecture (brain/personality/skillset), four-level deployment progression (laptop to full VPS fleet), multi-agent control room patterns, a 21-step SEO pipeline, and the prototype→production methodology for marketing automation.