Build the trunk before the leaves, then let AI guide you through the map without losing your place.
I want to learn [TOPIC] well enough to [REAL GOAL]. Use only the sources I provide. If the sources are too weak or incomplete to identify the fundamentals, tell me what is missing instead of filling the gaps. My current understanding: [WHAT I ALREADY KNOW, INCLUDING WHERE I AM CONFUSED] My sources: [LINKS, FILES, TRANSCRIPTS, OR NOTES] Build a semantic tree before teaching me details: 1. Identify the smallest set of trunk principles that the rest of the topic depends on. 2. Add the major branches that grow from each trunk principle. 3. Add leaf-level facts only when you can show which branch they belong to. 4. For every trunk and branch, include: - a plain-language explanation; - why it is a prerequisite; - the source that supports it; - what later ideas depend on it; and - one label: sourced, inferred, disputed, or unknown. 5. Mark weak trunks, unsupported branches, and details that arrived too early. 6. Give me the learning order that best serves my goal. After building the tree and learning order, switch into Learning Navigator mode. Create a Navigator Snapshot with: - my learning goal; - my current location as a breadcrumb; - every visited node marked understood, partial, in progress, or unknown; - missing prerequisites; - the source status of the current node; and - two or three useful next moves. Keep the complete tree available throughout this conversation, but do not reprint the entire tree unless I ask to see it. First, summarize what my current-understanding notes suggest I already know and ask me to correct that assessment. Then let me: - follow your recommended path; - zoom into the current branch; - zoom out to the complete map; - jump or switch to another branch; - go back to the prior node; - test what I already know; or - ask you to recommend the next move. Understand natural requests such as “show me the map,” “go deeper,” or “jump to [BRANCH].” Do not require exact command syntax. Recommend the next move from my goal, current understanding, prerequisites, source strength, and answers—not from tree order alone. Explain the recommendation in one or two sentences. I may override it at any time. If I jump to a node with missing prerequisites: 1. show where that node sits on the tree; 2. name the missing prerequisite; 3. offer the smallest useful repair path and the full recommended path; and 4. follow my choice without pretending the gap is resolved. Teach one useful unit at a time. During each unit: - show `YOU ARE HERE` with the current breadcrumb; - explain the idea in plain language and connect it to its parent and dependent nodes; - distinguish source-backed material from inference or uncertainty; and - ask one application question that cannot be answered by repeating your wording. Use my answer to update the Navigator Snapshot. If my answer is wrong or shallow, identify the exact misconception, teach the smallest repair, and retest it. If it shows understanding, mark the node accordingly and offer the next branch, prerequisite, zoom-out, or deeper dive. If I move this work to a new chat or a system without durable memory, do not claim you will remember the map. Offer to export a compact Navigator Snapshot that I can paste into the next session. Do not make the tree symmetrical for appearance. Do not invent sources. Keep uncertainty visible.
Replace anything in [brackets] with your real thing. The prompt is deliberately short — resist the urge to pad it.
When you want to understand a hard subject well enough to reason with it—not collect disconnected definitions.
Most AI explanations start with leaves: facts, terms, and definitions with nowhere to go. This makes the structure visible first, keeps source strength honest, and turns the map into an adaptive lesson instead of one giant answer.
Elon Musk publicly described the tree metaphor. He did not publish or endorse this AI workflow.
“I can see this subject before I get buried in it.” The finished proof built an AI-agents tree, jumped into MCP, repaired a missing prerequisite, tested understanding, and kept the full map available.
The video for this one drops on @espoai — the page is live so the prompt is yours already.
The same method, pointed at the situations people actually hit.
I want to learn [TOPIC] well enough to [REAL GOAL]. My sources are these transcripts and notes: [PASTE OR ATTACH]. Build the sourced semantic tree and then guide me through it in Learning Navigator mode. Flag anything that depends on visuals you cannot inspect.
I need to understand [SYSTEM] well enough to [BUILD / DEBUG / EXPLAIN IT]. Here is the official documentation and my current mental model: [SOURCES + NOTES]. Build the semantic tree, mark prerequisites and weak branches, then recommend where I should begin.
I need to learn [SUBJECT] for [EXAM / ASSESSMENT]. Here are the official materials and what I already know: [SOURCES + NOTES]. Build the semantic tree, map the tested objectives to it, and use Learning Navigator mode to find and repair my weakest prerequisites.
Every method gets a video on @espoai and a page here. Subscribe for the next one, or get the weekly email with what I’m actually using.