I Kept Forgetting My Own Stories

Here’s a problem I kept running into. I’d be in a group setting, someone would hit a topic I had a story for, and I’d freeze. Those stories were sitting in meeting transcripts and podcast archives. I just couldn’t pull them up when I needed them for writing or a presentation.

The usual fix was a better filing system. Keyword-stuffed titles. A tag taxonomy you’d actually maintain. That was the messy middle of personal knowledge management: organizing, tagging, and linking notes by hand.

Brooks and I agree on why people quit. The friction of manual organization is the main barrier to PKM adoption.

The messy middle is dead. AI agents can handle the organization now. You spend your time capturing insights and retrieving them.

Ask for the meaning, not the keyword

You no longer need to keyword-stuff titles or create complex tag taxonomies. You can simply ask, “Find a story about leadership under pressure,” and the AI will surface a note about a canceled flight where you rallied the team, even if the word “leadership” never appears in the text.

That’s semantic search. It finds the story by what it means.

Retrieval used to be the hard half of PKM. Now retrieval is easier than capture. So capture everything you might want later, then ask in plain language. Write the note. Dump the transcript. Ask later.

So should you throw the whole pile at an AI and walk away?

No.

The Two-Click Rule

Even with AI, basic organizational skills remain invaluable.

A clean folder structure reduces the tokens and the complexity an AI agent needs to search and process your data. Same for naming conventions. Separate Book Notes from Meeting Transcripts. Do you need a fancy taxonomy on top of that? No.

Here’s the test. If you can’t get to a file in two clicks, the agent works harder than it should, and accuracy drops. That’s the Two-Click Rule.

AI makes you faster, but structure makes it accurate. The “just throw it all in AI” advice skips that. Human structure still helps the machine.

Own the files. Then let the model see them.

Brooks flagged a privacy and security pitfall that shows up the minute you start sending notes and calendars into cloud tools. If that bothers you, start file-first.

Obsidian stores notes as plain Markdown text files on your device. That eliminates vendor lock-in. Combined with local LLMs, you get a private, future-proof PKM system. You own your data. You can run AI locally to protect privacy. That capability was impossible just a few years ago.

MCP servers are the bridge. Model Context Protocol lets AI assistants like Claude or ChatGPT interact with your local files, calendars, and task managers without custom coding. The AI can see your Obsidian vault or your Google Calendar and take action. Same idea as Zapier connecting apps, except you drive it with natural language.

That’s the universal translator. You don’t have to be a developer to care. Your notes, calendar, and task manager can sit in the same conversation as the model.

Don’t switch tools because the feature is new

The productivity community is currently in a new wave of shiny object syndrome. This time it’s about AI integrations and MCP servers.

Brooks’s closing advice was the one to keep. Experimenting is fine. If your current system works, don’t migrate just because a new AI feature was released. The new tool has to solve a specific pain point.

The community already did this with task managers. Now it’s AI tool churn.

Feeling overwhelmed by new AI tools? Take a breath. Write the pain down in one sentence. If the integration doesn’t kill that sentence, stay put.

Stay in the loop

Relying entirely on AI to organize or summarize your knowledge can lead to vibe-coding-style disasters. Subtle errors you don’t notice. A summary that sounds right. A note in the wrong folder. Then you trust the library and it serves you a slightly false version of your own history.

Brooks brings up the Co-Intelligence idea for a reason. Don’t passively consume AI output. Always verify AI-generated organization or summaries so your knowledge base stays accurate and trustworthy.

Spot-check the filing. Read the extracted stories. Fix the misses while you still remember what actually happened.

Turn transcripts into a story bank

Forgetting my own experiences in group settings was the pain. Automated extraction plus semantic search is the fix.

Use an LLM to extract anecdotes from meeting transcripts and podcast archives. Build a searchable library of personal stories. When you need one for writing or a presentation, ask in plain language.

Capture first. Get the transcripts and archives into a Meeting Transcripts folder. Don’t pause to tag them.

Extract second. Point an LLM at that folder and tell it to pull personal anecdotes into a Story Bank. Keep Book Notes in a different folder so reading notes don’t get mixed with stories you lived.

Retrieve third. Ask in plain language, then check that the story it returns is actually yours and actually true.

Today, do one thing. Pick a single meeting transcript or one podcast archive. Run it through an LLM with instructions to extract the personal anecdotes. Drop those into a Story Bank folder. Then ask it for a story you know is in there, and see if it finds the right one.

If it misses, fix the folders and the names before you install anything new.