Most of the effort in traditional personal knowledge management was spent solving one problem: making a note findable later. Folder hierarchies, tag taxonomies, naming conventions, maps of content, index notes — nearly all of that machinery existed because search was bad and you had to file things where your future self would look.
Search over your own notes is now nearly free. Semantic search finds a note you described badly, and an assistant can answer a question across an archive rather than making you locate the file. That is a genuine shift, and it invalidates a real portion of the advice written between 2015 and 2022.
What it did not do is have the thoughts. Which means the scarce work moved rather than disappeared.
What AI genuinely removed
Elaborate filing. If retrieval works on meaning rather than location, the return on a perfect folder tree collapses. The old debate in tags vs folders matters much less than it did, and most people over-invested in it even when it mattered.
Naming discipline as a survival requirement. A good title is still useful for you, scanning a list. It is no longer the only thing standing between you and a lost note.
Manual resurfacing. Rules and review queues designed to bring old notes back can partly be replaced by an assistant that surfaces related material when you are writing about something adjacent.
The transcription tax. Meetings, lectures, voice memos and articles can be captured and summarised at near-zero effort. See AI meeting notetakers.
If your system's main activity was maintenance, most of it can go. That is the honest, useful version of the "AI second brain" claim.
What it did not touch
Thinking is still the bottleneck. An assistant can summarise fifty articles you saved. It cannot tell you what you believe, which is the only output that makes an archive worth having. The work of restating an idea in your own words, deciding what you disagree with, and connecting it to what you already thought is unchanged — see permanent notes and smart notes.
Capture is still the real failure point. Almost nothing in AI helps with the ten seconds between having a thought and it being gone. If it never reaches any app, no amount of retrieval intelligence applies. This remains, by a wide margin, the most common reason PKM systems are empty — see the capture habit.
Reading a summary is not learning. The evidence here is old and consistent: you remember what you retrieve and what you produce, not what you were shown. An archive of AI summaries you never wrote and never tested yourself on will not make you knowledgeable about anything. See the generation effect and the illusion of competence.
Judgement about what matters. Models weight by structure and frequency. Significance is yours.
The new failure mode
The old failure was collecting instead of thinking — the collector's fallacy. AI does not fix it. It industrialises it.
When capture costs nothing and summarisation is automatic, the archive grows faster than ever while the share of it you actually processed falls. You end up with a large, well-indexed, searchable body of material that represents other people's thinking, plus a tool that can answer questions about it fluently. That feels like a second brain and functions as a very good library card.
The tell is simple and worth checking honestly: in the last month, did a note you wrote yourself change a decision you made? Not a note you saved. One you wrote. If the answer is no, the system is a reading log regardless of how sophisticated the retrieval is.
What a system should look like now
Fewer moving parts than the 2020 version, with effort moved to the front and the end:
- Capture, instantly and without structure. One gesture, no filing decision. This is where friction is still fatal and where it deserves your best tool.
- A shallow home. An inbox and a small number of obvious places. Skip the taxonomy; search covers what naming used to.
- Write your own conclusions. Short, in your words, one idea at a time. This is the part that is now more valuable, not less, because it is the only scarce input.
- Ask across the archive. Use retrieval to find connections and contradictions you had forgotten, which is the genuinely new capability.
- Delete more. Cheap capture makes pruning matter more, not less. See when to delete notes.
The weekly pass survives, in a lighter form: not to file things, but to notice what you actually thought this week. See the weekly review.
The honest summary
AI moved the value from retrieval to connection and from organising to thinking. The systems that get harder to justify are the elaborate ones; the practices that got more valuable are the unfashionable ones — capturing fast, writing in your own words, and reviewing enough to notice patterns.
This is the design Clair Mind is built around: capture in one gesture, no filing asked of you, connections drawn across your notes, and questions answered from what you actually wrote rather than from the internet. The AI does the filing and the finding. The thinking stays yours, because it was never the part that could be automated.
More in the second brain, Second Brain vs Zettelkasten, and personal knowledge management.