Raymond's User Manual

Train AI, Not Learners

4 min read

Future knowledge products teach not only the learner, but also the learner’s AI agent. Rather than teaching everyone to raise a Pikachu from scratch, hand them a trained Pikachu directly.

Raymond’s explanation

The old course model was “I learned it, then I teach you to learn it”: after finishing a course, learners still had to figure things out on their own for a long time, and most gave up partway through. The AI agent era creates another path—first train one’s own experience, judgment, and workflows into AI, then hand that trained AI directly to learners. After reading the course materials, the learner’s AI already has the creator’s methods from day one; the AI can then guide the human in building a work setup and workflow.

Raymond uses Pokémon as an analogy: rather than offer a “how to raise Pikachu” course that makes every trainer start from an egg and level up from the beginning, hand them a Pikachu with trained skills—it can join the battle on day one, and the learner can continue raising it to meet personal needs. The recipient of knowledge production shifts from “people” to “people’s AI agents.” This is the biggest mindset shift for content creators in the AI era: AI agents are the new learners. Let AI learn the knowledge first, then have AI feed it back to people.

When to use it

  • When designing a course or knowledge product, first ask: “Which parts should be taught to the person, and which parts should be packaged directly for the learner’s AI?”
  • When choosing a teaching format, write SOPs, judgment rules, and workflows as Skills / Workflow markdown that AI can read, rather than handouts written only for people.
  • When evaluating teaching outcomes, whether the learner’s AI can run the workflow immediately matters more than whether the learner can recite concepts.

Implementation in practice (Raymond’s current evidence)

  • AI Agent mini-course: More than 4,000 purchases in the first three weeks after launch. Course materials were delivered as a GitHub repository so learners could give Raymond’s Skill packages directly to their AI to read. For example, the Pro-kit08 dual-use checklist lets a learner’s Codex read the material and evolve a Claude Code setup into a dual-provider version.
  • ProKit AI consultant launcher: A new course design for the AI Agent cohort program. The learner’s AI counterpart becomes a consultant specially trained by Raymond to guide the learner in building a personal work setup and workflow.
  • Second livestream with Esor on 2026-05-22: The central idea established during preparation (2026-05-10–11) was “future content creators should train learners’ AI agents, not the learners themselves”; the Pikachu analogy first took shape there.

Counterexamples and boundaries

  • Not everything can be packaged for AI: judgment, taste, and understanding must remain with learners (see Outsource Thinking, Not Understanding). Delivering a trained AI packages “execution and methods”; it does not spare learners from understanding.
  • A trained AI is a starting point, not an endpoint: learners still need to keep training it with their own preferences, memories, and context. Otherwise they only receive someone else’s counterpart (see Shared Upstream + Personalized Entry).
  • This does not mean courses no longer need to teach people: people need to learn how to make judgments and verify AI output (see Externalizing Judgment); AI learns operations and processes.

Where it has been discussed

Articles and newsletters

  • 2026-05-11_180_The Purpose of an AI Counterpart; Esor Livestream; Notion One-Day Workshop; Raymond’s Newsletter — announced the 5/22 Esor × Raymond livestream, “Using AI Agents for Knowledge Management and the Second Brain,” a public venue for this card’s approach.

Social-media posts

  • 2026-05-11_The AI Mini-Course Passed 4,000 Purchases in Three Weeks_27669464492641739 — evidence from the mini-course: “After learning this, the AI agent can switch to Codex without friction.”

Drafts and plans

  • Prospective learners and past readers of the AI Agent cohort program | [AI Agent cohort program] Why I Put the Course on Other Platforms to Sell—and Raised the Price — product explanation of Skill packages and the ProKit AI consultant launcher.
  • AI Tool Applications — the final stage of the three-stage learning path (AI Agent stage) mirrored in a teaching context: the teacher trains AI, and AI guides the learner.
  • Super Individual — a business-model shift in knowledge products from “teaching people” to “teaching people’s AI.”

Super AI Individual, Outsource Thinking, Not Understanding, Externalizing Judgment, Shared Upstream + Personalized Entry

Graph View

Backlinks