Skip to content

The 7-mission universal learning path

Last verified: 2026-05-06 · Time: 7 missions × ~30 min each, spread across 1–2 weeks

If you have one AI subscription and want to actually maximize it, do these seven missions in order. They work for any AI — ChatGPT, Claude, Gemini, Grok, Perplexity, GitHub Copilot, or anything else.

Each mission is small enough to finish in one sitting, and each one raises the floor of what you'll get from your AI for the rest of the year.

How this is structured

  • One mission per session.
  • ~30 minutes each.
  • Each mission has a goal, what to do, what good looks like, and a graduation step.
  • All missions assume no API key, no code, no terminal.
  • After Mission 7 you'll have: a portable AI profile, a reusable assistant, a researched-with-citations workflow, a saved Project, a scheduled or calendar-driven workflow, an evaluation habit, and a switching plan for when one AI isn't enough.

Mission 1 — Build a portable AI profile

Goal: every AI you use should know your tone, role, and constraints without you re-typing them.

Do:

  1. Open Memory and preferences § Portable AI profile.
  2. Copy the template into a notes file (e.g., ~/Notes/ai-profile.md).
  3. Fill it out. Be honest about how you actually like output.
  4. Paste into your AI's memory / Custom Instructions / Saved info / Profile preferences.
  5. Open a fresh chat. Ask: "Repeat back what you remember about me." Verify.

Good looks like: the AI responds in your tone the next time you open it, without you reminding it.

Graduate: keep the file under version control (Drive, GitHub, Obsidian sync). Update monthly.

Mission 2 — Run your first useful workflow

Goal: complete one real task, end-to-end, in under 10 minutes.

Do:

  1. Pick one of the beginner examples. Recommended starters: Write a difficult email or Meeting prep brief.
  2. Follow the Layer 1 (chat-only) steps.
  3. Save the prompt to your notes file under "Prompts — ."
  4. Note one thing the AI did well and one thing you'd change.

Good looks like: something you can use today (an email you'd send, a brief you'd open before a meeting), produced in under 10 minutes.

Graduate: repeat with a second example next week.

Mission 3 — Make it reusable as a Project

Goal: stop re-typing system prompts.

Do:

  1. Pick the workflow you'll repeat the most.
  2. Open your AI's Project / Space / Gem slot:
  3. ChatGPT: New Project.
  4. Claude: New Project.
  5. Gemini: New Gem.
  6. Perplexity: New Space.
  7. Grok / Other: open your notes app and pin the prompt — that's your "project."
  8. Paste the system prompt into Instructions.
  9. Add 1–3 reference files (anything from your existing notes — keep it small).
  10. Run the workflow inside the Project. Confirm it inherits the instructions.

Good looks like: you can re-run the workflow with a one-line message inside the Project.

Graduate: add 1 reference file per use until the Project is genuinely "yours."

Mission 4 — Research one thing with citations

Goal: verify everything you ever ask an AI.

Do:

  1. Pick a question whose answer you'd want to cite.
  2. Run it in Perplexity if you have it (best fit). If not, run it in your AI with web search enabled.
  3. Read every citation. Click through. Note any that don't resolve or don't say what the AI claimed.
  4. Write a 3-sentence summary of what you learned, with cited sources, in your own voice.

Good looks like: you can answer "where did you get that?" for every sentence in the summary.

Graduate: when you come across a fact you'd otherwise just believe, run this exercise on it.

Mission 5 — Schedule (or fake-schedule) a workflow

Goal: stop being the trigger.

Do:

  1. Read the No-code automation guide before doing anything.
  2. Pick a workflow you want to run weekly (a research watch, a status report, a meeting brief).
  3. Run it manually 3 times. If it succeeds 3/3, schedule it:
  4. ChatGPT Tasks / Gemini scheduled actions / Claude Cowork if your plan has it.
  5. Otherwise: a calendar reminder + saved prompt — the manual repeat-run fallback. Equally effective.
  6. Document where the off-switch is.

Good looks like: the workflow runs without you for one full week.

Graduate: add a second scheduled task only after the first has earned its place for a month.

Mission 6 — Build an evaluation habit

Goal: know when your AI is wrong.

Do:

  1. Pick the workflow from Mission 3 (your reusable Project).
  2. Write down 5 inputs:
  3. 2 easy (the AI should nail these).
  4. 2 ambiguous (the AI should ask, flag, or skip per your preference).
  5. 1 out-of-scope (the AI should refuse).
  6. Run them through. Score each: pass / partial / fail.
  7. If 4/5 pass, you're ready to share the workflow with others. If not, tighten the system prompt and re-run.

Good looks like: you can show a colleague the eval set and the scores. The colleague would trust the workflow.

Graduate: re-run the eval set monthly. Watch for drift.

Mission 7 — Decide when to switch tools

Goal: pick the right AI per job — not the AI you happen to have open.

Do:

  1. Open the Capability map.
  2. List the 3–5 workflows you do most.
  3. For each workflow, mark:
  4. Which AI you currently use.
  5. Which AI the Capability map suggests is the best fit.
  6. Whether the gap is worth the cost of adding a second subscription / using a free fallback.
  7. Make one decision: keep your current AI; add a second; or switch.
  8. Update your portable AI profile to reflect the decision.

Good looks like: every recurring workflow has a deliberate AI choice — not a default.

Graduate: revisit annually, or whenever a vendor ships something new and meaningful.

Where to go after Mission 7

A few rules for the road

  1. No API keys until you outgrow the chat product. Most users never need them.
  2. Every assistant is drafts-only by default. Humans send, post, push, pay.
  3. Three clean manual runs before scheduling. Always.
  4. Keep your portable AI profile up to date. It's the single file that travels with you.
  5. When in doubt, ask the AI to interview you before answering. It saves a round-trip.