Skip to content

The 30-Day Learning Plan

Section 15 · Lesson 4 · Level: beginner · ~15 min · Prereq: none

Why this matters

A 135-lesson wiki is easy to start and easy to abandon in week two. This page is the schedule: four weeks at roughly 45 minutes a day, one deliverable every week, and a checkpoint that tells you whether you are actually learning or only reading. Do not follow it perfectly. Follow it unevenly for 30 days instead of perfectly for three.

The plan at a glance

timeline
    title The 30-day plan
    section Week 1 Foundations
        Days 1 to 2 : Workbench, first prompts, real output
        Days 3 to 4 : Git, branches, first pull request
        Days 5 to 7 : First CI pipeline green
    section Week 2 Ship something
        Days 8 to 10 : API, database, first migration
        Days 11 to 13 : Staging deploy and a smoke test
        Day 14 : Checkpoint — public URL
    section Week 3 Harden it
        Days 15 to 17 : Tests, smoke tests, reviewing agent output
        Days 18 to 20 : Secrets, rules file, skills
        Day 21 : Checkpoint — agent-proof repo
    section Week 4 Add the model
        Days 22 to 24 : Context, MCP, retrieval
        Days 25 to 27 : Evals, injection, budget
        Days 28 to 30 : Ship it, measure it, plan the next 30 days

Week 1 — foundations (~45 min/day)

Deliverable: a public repo with a green pipeline and a README a stranger could follow.

Week 2 — ship something

Deliverable: your API and database on a public URL, with a merge that auto-deploys to staging and one promotion you approved.

Week 3 — harden it

Deliverable: a repo with a lean AGENTS.md, three skills, a hook, required checks, and a logged test of all three.

Week 4 — add the model

Deliverable: a model feature behind a flag, an eval set in CI, real numbers in docs/ai-feature-metrics.md, and a written plan for the next 30 days.

Four tracks, if you are not here to ship a product

The four paths on How to use this wiki map onto the same 30 days. Just shipping: weeks 1–2 and Capstone 1; skip the evals week. Professional engineering: week 1 on git and CI, then branch protection, the test pyramid, ADRs on day 19 instead of skills, code review on day 17. AI-product builder: weeks 1 and 4, plus context engineering and MCP, and Capstone 2 instead of Capstone 3. Team lead: Rules of operation, Writing SOPs, Issue-driven development, CODEOWNERS, Human accountability, Managing a skill library, then Case study synthesis.

Checkpoints

  • Day 7. You can clone your repo on a clean machine and run its tests using only the README, and explain what a pull request does.
  • Day 14. Your app is on a public URL. A merge deploys to staging automatically, and you promoted one commit yourself, with a rollback you tested.
  • Day 21. A fresh agent session, given only the repo, can make a small change and open a passing pull request.
  • Day 30. A model feature behind a flag, an eval score in the repo, and one number you can defend: cost per request.

After day 30

The plan does not end, it changes gear: one shipped change per week (small is fine — the point is the loop), one lesson per week in your own docs/lessons.md with the command and its output, and one skill per month capturing a workflow you repeated three times.

Try it

  1. Copy the four weeks into your calendar as 30 blocks of 45 minutes, titled with the day's lesson.
  2. Start a notes file and paste one real command output into it every day. No summaries.
  3. At each checkpoint, write one sentence: what you can do now that you could not on day zero.
  4. At the end of day 30, write the next 30 days yourself.

Common mistakes

  • Reading ahead instead of shipping — finishing week 3's reading in week 1 with nothing deployed.
  • Skipping week 3 because week 2 felt good — you get a deployed app you are afraid to change.
  • Treating the checkpoints as optional — they are the only evidence the month worked.
  • Not recording outputs — "I read about migrations" and "I ran one" are different things a year later; the note with the error message is the difference.
  • Stopping at day 30 — the compounding starts at week five.

Key takeaways

  • 45 minutes a day, four weeks, one deliverable per week, four checkpoints.
  • Every day ends with a run command and a recorded output.
  • Pick a track on day 1 and let it reorder days 17–28; do not do all four.
  • The checkpoints are tests of capability, not of reading: deploy, promote, get a stranger productive, record a number.
  • After day 30: one shipped change a week, one lesson a week, one skill a month.

Further learning