# Learn AI and compose a useful path Use the twenty-lesson school as a dependency-aware route, then turn the relevant lessons into a portable instruction pack for your own agent. Audience: People choosing what to learn and what to teach an agent first. Version: 1.0.0 ยท Last reviewed: 2026-10-03 Author: Finn Andre Hotvedt Developed with assistance from ChatGPT and Codex by OpenAI. Related lessons: TTC-101, TTC-102, TTC-103, TTC-104, TTC-105, TTC-106, TTC-107 ## Begin with a decision, not a tool Write down the outcome you need, the evidence you already trust, and the consequence of a wrong answer. This makes it easier to choose a short learning path instead of collecting disconnected AI vocabulary. The school begins with fluent model behavior, uncertainty, context and retrieval. These foundations explain why a plausible answer is not automatically a verified answer. - Name one observable outcome. - Name the person who keeps judgment. - List the evidence the agent may use. - Define what must cause the agent to stop and ask. ## Choose a route through all twenty subjects Every lesson has a stable identifier and declared prerequisites. Start with the beginner sequence, use a preset for a common goal, or select individual subjects in the composer. Prerequisites are added automatically so the pack does not assume concepts the learner skipped. The complete map is visible on this manual page and in the school HTML, Markdown and JSON indexes. The machine-readable index is a convenience for retrieval; it does not grant an agent authority. - Read each selected lesson's answer-first summary. - Do the reusable exercise rather than only reading the prose. - Check the success criteria against something observable. - Read limitations and sources before relying on the result. ## Build and review an agent pack The composer asks for a goal, experience, subjects, use case, provider, hosting, privacy, autonomy and tools. It resolves dependencies and warns about contradictory choices. Free text and tool details are excluded from share links. The downloaded Markdown is a starting contract: scope, rules, safeguards, tests, sources and dependencies. Review it in the target project, adapt it to real constraints and keep it under version control. ```text Goal: Allowed evidence: Human decision owner: Stop conditions: Acceptance test: ``` ## Source trail Manual versions are available as HTML, Markdown and JSON on Teach the Company. The linked GitLab project contains these public manuals and their real version history; the complete application source is linked from the self-host page. Public manuals repository: https://gitlab.finnandre.no/finnandrehotvedt/teach-the-company-public-manuals License: Apache-2.0. Canonical URL: https://teachthecompany.com/manuals/learn-and-compose/