The promise of a “second brain” — an external system that captures, organises, and retrieves everything you know — has been in productivity circles for decades. AI has finally made it genuinely achievable. Here is what works in practice, not in theory.
The Core Problem With Traditional PKB Systems
Most personal knowledge base systems fail not for lack of tools but for lack of workflow discipline. Notion databases get set up with elaborate structures and then fall into disuse. Obsidian vaults grow into impenetrable forests of untitled notes. Roam graphs become dense without becoming useful. The common failure mode: the capture-and-structure step requires too much friction at the moment of capture, so it only happens intermittently, and the system fails to represent the full scope of what you actually know and read.
The AI-Augmented Approach
AI changes two specific parts of the PKB workflow: capture processing and retrieval. For capture: instead of requiring yourself to immediately categorise, tag, and summarise every note, capture raw — a voice note, a URL, a PDF — and let AI do the processing later. Tools like Readwise Reader (for articles), NotebookLM (for document collections), and Claude or ChatGPT via file upload can summarise, extract key points, and suggest connections across your existing notes. For retrieval: conversational AI search is more natural than tag-based search. Ask “what do I know about protein synthesis?” across your notes and get an integrated answer.
The Minimum Viable Stack
For most people: Obsidian for the vault (local files, portable, Markdown format), Readwise for article and book highlight capture (syncs to Obsidian), and Claude or ChatGPT for processing unstructured captures and generating summaries. The Obsidian Smart Connections plugin and Obsidian Copilot plugin both add AI-assisted retrieval and note generation within the vault. This stack has enough friction removal to maintain consistent capture, and enough structure to remain retrievable.
What AI Cannot Do
AI cannot determine what is worth capturing for you — that requires your judgment about what connects to your goals. It cannot ensure the quality of your source material. And the “AI will connect my notes for me” aspiration is partly true but requires a critical mass of notes with sufficient specificity; general notes about “productivity” connect to nothing meaningfully. The system works when the inputs are specific: “chapter 4 of this book,” “this specific paper’s methodology,” “this conversation’s three takeaways” — not vague impressions.
