PKM Is Dead Long Live PCM: Obsidian Is Different Now
Video by Obsidian Explained (No Code Required) on YouTube · 14 min · uploaded June 2026
Notes and analysis by BrainTube, written from the video. Not affiliated with Obsidian Explained (No Code Required). Watch on YouTube
In short
The video argues that the filing and linking work of PKM can now be handed to an AI agent inside your Obsidian vault, so your job shifts to giving that agent good context: who you are, your voice, your decisions. The reframe from knowledge to context is useful, and the vault tour shows a sensible set of files. But it never names the agent or shows it working, and it treats maintenance as free when it is really delegated and still needs review.
The argument
Every PKM framework, from Zettelkasten to PARA, is in the creator's words an instruction manual for a human to act as a librarian. That maintenance work grew until it crowded out the output it was meant to support. An AI agent in the vault can now read notes, keep links, catch contradictions, route the inbox and write session summaries. So the discipline should change from personal knowledge management to personal context management: store what an agent needs to act for you right now, treat the vault as the agent's working memory rather than a deliverable, and move your own effort up to judgment and shipping. He demonstrates with a demo copy of his own vault, run by an agent he calls Pragma.
Key ideas, with timestamps
- 0:49
Frameworks as librarian manuals
He groups Zettelkasten, PARA, BASB, ACCESS and others as one thing: a set of rules for a human to index, catalogue, link and maintain notes. His claim is that the maintenance consumed the benefit, and that vaults grew while client lists, audiences and income did not.
- 2:45
What the agent takes over
He lists the jobs an agent inside the vault can do: build a taxonomy, keep wiki-links, surface forgotten backlinks, flag contradictions between two files, route the inbox, tidy daily notes, write an end-of-session summary and run a connectivity audit. He is explicit that this replaces the filing, not the thinking.
- 4:43
Context, not knowledge
His first PCM principle: knowledge is stored for later, context is what an agent needs now. That means who you are, who you talk to, your voice, your hard constraints, what you tried last week and what you decided three months ago, and why.
- 5:32
Vault as substrate, you as principal
The second and third principles: the vault is the agent's working memory, judged by whether the agent has what it needs, not by how it looks. Your role moves up to deciding what deserves attention, reading the room and shipping.
- 7:03
The agent contract and voice spec
The demo vault has no PARA folders, tags or nested categories. A contract file tells Pragma who he is, what voice to use, which mistakes never to make and where everything lives. It is read at every session start and never changed without his permission. A separate voice file is a specification for the agent, not a style guide for him.
- 8:11
Task board, daily note and log
The task board has five lanes, and the agent enforces a maximum of three most important tasks, which he links to his ADHD. Running a begin command creates the daily note: the agent loads memory, reads yesterday's handoff, walks the board and suggests where to start. An append-only activity log is written at the end of each session.
- 9:37
People files and a pruned memory file
A network folder holds one file per person, created when he mentions someone and cross-referenced with tasks and daily notes. A memory file holds active projects, open threads, decisions and recent contacts, and is pruned when things resolve. The last 90 seconds promote his paid community.
What it gets right
The reframe is useful. Asking "does the agent have what it needs?" is a better design question for 2026 than "where should this note go?" It pushes you toward a handful of high-value files (identity, voice, constraints, decisions, open work) instead of an ever-growing archive.
The file set is sound and copyable. A contract file read at session start, a voice spec, a capped task board, an append-only log and a pruned working-memory file is a good minimal pattern. The split between the log (never edited, full history) and the memory file (pruned, current) is the most thoughtful detail. One preserves the record, the other keeps the agent's context small.
Decisions with reasons. Including what you decided and why is the part most notes systems miss. It is what lets an agent avoid re-proposing ideas you already rejected.
What it leaves out
Which agent? The video never says what runs Pragma, how it is connected to the vault, or what the begin command actually calls. For a channel subtitled "No Code Required", that is a large gap. A viewer cannot reproduce the setup from this video alone.
No live run. We see files, not the agent creating a daily note, filing a person or catching a contradiction. Claims like a self-maintaining graph are shown only as finished output.
Delegated is not free. He calls the low-level PKM work basically free. In practice it costs model usage, and every automated edit is a chance for a wrong link, a duplicate person file or a mis-filed note. Someone still has to spot-check. The fair claim is that maintenance becomes review instead of manual labour.
Pruning loses memory. A memory file that is pruned aggressively stays small, but the reasons behind old decisions drift out of it. His activity log partly covers this, though only if the agent searches it. Long-lived reference material, such as what you have read and watched, needs a home that is not pruned.
Knowledge still matters. Context tells an agent how to act. It does not supply the substance of what you know about a field. Dropping the knowledge side entirely would leave the agent sounding like you while saying generic things.
How this applies to memory across AI tools
His setup ties context to one vault and one agent. The same idea works better when the context is reachable from whichever assistant you open, through MCP. Keep the small, private files (contract, voice, open work) where you control them, and put the large reference layer somewhere every client can query.
That reference layer is what BrainTube covers: it saves YouTube videos with their transcripts, articles, PDFs and podcasts into one compiled knowledge base and exposes it as an MCP server, so Claude, ChatGPT, Cursor and other MCP clients can search it alongside your own context files.
For background, see what an AI memory layer is and persistent memory vs the context window.
Who should watch it
Watch it if you keep rebuilding your vault structure and want a different way to think about what it is for. The argument is 0:28 to 6:35 and the file tour is 6:41 to 11:00; the most reusable part is the contract, task board and memory file between 7:20 and 10:27. Skip from 13:16 onward, which is a pitch for his community. Skip the video entirely if you want instructions: it does not name the agent or show how to connect one, so you will need another source to build it.
Frequently asked
- What is personal context management (PCM)?
- It is a term for organising notes around what an AI agent needs to act for you, rather than around storing knowledge for later. That means files describing who you are, your writing voice, your constraints, your open work and past decisions. The agent reads them at the start of each session and handles most filing.
- Can an AI agent maintain my Obsidian vault?
- An agent with file access can create notes, add links, file inbox items and write summaries in a vault, because Obsidian notes are plain markdown files. It works best with written rules for where things go, and it still needs regular review, since wrong links or duplicate notes are easy to miss.
- What should an AI agent's memory file contain?
- A useful memory file holds active projects, open threads, recent decisions with their reasons and key people. Keep it short and prune it when items close, and keep a separate append-only log so the full history is not lost.
Keep videos like this one searchable
BrainTube saves YouTube videos with their transcripts into one compiled knowledge base that Claude, ChatGPT, Cursor and other MCP clients can search.
Go deeper
More video notes
- Build an Obsidian BACKEND Not a Second Brain! · Eric Michaud
- Complete Guide to building an AI Second Brain with Claude · Ray Fu
- The NEW Way to Build a Second Brain w/ Claude (2027) · Nick Saraev
- Build a Self Improving Claude Knowledge Base · Paul J Lipsky
Are you the creator and want this page changed or removed? Contact us and we will act within 48 hours.