Skip to main content
Pick one path. Run it for a week. Do not try to finish the internet.

Paths you can take next

Path A · Make your week compound

Goal: one Project that starts every chat already briefed.
  1. Create a Project for your real team or portfolio.
  2. Add a short custom instruction and three source files you actually reuse.
  3. Run five real tasks inside it before you invent a second Project.
Start here: Projects overview

Path B · Ship one skill

Goal: stop retyping the same procedure.
  1. Pick a workflow you did twice this month.
  2. Write a skill with trigger phrases, rules, and one before/after example.
  3. Install it, trigger it from a real request, then tighten once.
Start here: What is a skill · Building skills

Path C · Connect one system safely

Goal: reach without oversharing.
  1. Connect one everyday service.
  2. Set writes to needs approval.
  3. Run one read task, then one deliberate failure probe.
Start here: Connectors · Lab: Connected run

Path D · Practice agent judgment

Goal: get comfortable with loops that can act.
  1. Re-run the sharp vs fuzzy loop on a real ops task.
  2. Write a definition of done before you start.
  3. Read the lap log, not only the deliverable.
Start here: Agentic loop · Agents overview

Books and long reads

O’Reilly ebook-style deep dive on prompting patterns, evaluation, and failure modes beyond any one vendor.Open Chip Huyen’s systems view: evaluation, RAG, agents, and production trade-offs. Good once you outgrow chat tips.Open Broader systems literacy if your role sits next to model deployment, data quality, and monitoring.Open

Practice challenges

Do these on your real work, not toy prompts.
  1. Same brief, three models once a week for a month. Keep a personal scoreboard of when Opus actually earned the upgrade.
  2. One Cowork folder job every Friday: renewals pack, expense cleanup, board appendix draft.
  3. Red-team one workflow from Safety before you automate it.
  4. Write one skill that removes a repeated Slack or email ritual.

Communities and ongoing feeds

Fast, concrete notes on agents, tools, evaluation, and failure modes as the ecosystem moves.Open Product and research announcements. Skim titles weekly; open only what matches your stack.Open API, prompting, tool use, and evaluation references if you start building, not only chatting.Open

A good stopping rule

If you have not used yesterday’s learning on a real task, do not open another guide. Apply first, then come back.