Eric J. Ma describes his personal knowledge management (PKM) system built on Obsidian and enhanced with AI coding agents, which he uses to manage twelve direct reports across two teams. His core insight: choosing plain text files in 2022 turned out to be prescient, as the format proved natively compatible with AI agents that arrived later — no migration or conversion layers needed. His vault is organized around distinct note types (daily bullet journals, meeting notes, people dossiers, and project control towers), all documented in an AGENTS.md file so coding agents understand the system's structure.

The workflow spans three phases: ingestion, maintenance, and production. For ingestion, he converts meeting transcripts, PowerPoints, Word docs, PDFs, and Excel spreadsheets into plain text using Python libraries and vision-language models, achieving roughly 90-95% accuracy. For maintenance, he runs periodic "sweeps" where the agent updates people and project notes from source material, with hallucinations occurring only about once every four or five cycles. For output, agents help publish to Confluence, generate HTML presentations, create Jira tickets, and produce PowerPoint decks — all from markdown source files.

The system reduced his knowledge management overhead from 30-40% of his time to under 10%, with the key enabler being agent skills — procedural knowledge encoded as executable markdown that compounds over time. Ma frames the post not as a prescription but as an invitation, emphasizing that the combination of plain text and coding agents is powerful, while acknowledging that the specific implementation should vary per individual.