I Built Karpathy's AI Knowledge Base in Claude: Try it!

You are helping me turn one specific YouTube video into real, durable learning.

Source video:
- Title: I Built Karpathy's AI Knowledge Base in Claude: Try it!
- URL: https://www.youtube.com/watch?v=ib74sLgjIBM
- Topic: Agent Architecture
- My current learning frame: Build the raw/wiki/outputs folder structure with a CLAUDE.md schema, dump in a batch of notes, have Claude compile and link the wiki, ask it a question and save the answer, then schedule a monthly health check.
- Why this matters: New playlist item from Systems Made Better; queued for transcript-backed review, topic mapping, and a practical learning artifact.

Transcript anchors from this exact video:
- 0:30 / Evidence 1: "is genuinely the most useful AI setup I've seen in months and implemented in Claude. And it takes probably 45 minutes to build over a weekend. No obsidian, no vector databases, no code, just a brilliant self-improving knowledge..."
- 8:45 / Evidence 2: "ask it to do that and this is how the system works and how they are independent. Nice. And then the detailed behavior is for each system. That's great. Then it should be working on one in here."
- 10:54 / Evidence 3: "will be doubling as a systems memory. It talks it through how to do things. You don't need to worry too much about that right now. Uh that is the plan and you can ask Claude to do..."
- 12:56 / Evidence 4: "also worth saying that in my system I have an about me section and a context map. And that context map shows all of the key databases in notion which it can read from. So in many ways..."
- 15:47 / Evidence 5: "Corey Gam. Cool. So we've got our raw input. Uh my Claude system also created an ingested uh registry. So it talks about when everything went in, which is useful. Okay, step three is build the wiki. This..."
- 18:51 / Evidence 6: "usage, we're 39% into my current session. And actually, great news. Claude recently announced that they are doubling usage limits across sessions and during peak hours. That's not weekly limits, but it is session limits. And you can,..."
- 29:58 / Evidence 7: "little um blue dot for something and you can go and look at that and find the report and the brief. So this for example is another scheduled task that I'm running and essentially draft stuff so I..."

Your task:
1. Use the transcript anchors above as the primary source packet. If you add outside context, label it clearly as outside context and keep it secondary.
2. Create a source-check table with columns: timestamp, claim, what the demo proves, confidence, and what still needs verification.
3. Extract the actual teachable claims from the video. Do not invent claims that are not supported by the title, lesson frame, or transcript anchors.
4. Build a reusable learning artifact: A one-page agent harness map with tool boundaries and proof signals.
5. Include:
   - a plain-English definition of the core idea
   - a diagram or structured model using this sequence: Intent -> Model -> Harness -> Tools -> Verifier -> Artifact
   - 3 concrete examples that apply the video idea to real agentic work
   - 2 failure modes the video helps prevent
   - a checklist I can use the next time I run Codex or Claude
   - one practical exercise with a clear done signal
6. Add a "learning transfer" section: what changes in my workflow tomorrow if I actually learned this?
7. Add a "source check" section that cites which transcript anchor supports each major takeaway.

Quality bar:
- Make this specific to "I Built Karpathy's AI Knowledge Base in Claude: Try it!", not a generic Agent Architecture essay.
- Prefer operational examples, failure modes, and reusable artifacts over broad definitions.
- Call out uncertainty instead of smoothing over weak evidence.
- If evidence is weak, say what transcript segment or timestamp needs review instead of guessing.
- Finish with a concise artifact I could paste into my learning app.