Frame the Problem Before Picking Tools
A learning roadmap should never begin as a shopping list of tools—it starts with one real thing you want to change. Clarify why you are learning, how much time you have, and what concrete action you want to complete.
Replace “I want to learn AI” with “I want to turn a long article into verifiable bullet points.” The first has no boundary; the second defines inputs, actions, and acceptance criteria.
Key points
- What is the exact input?
- What outcome do you expect?
- How will you verify that the output works?
Exercise Pick one repetitive task from this week and write a Problem Card answering these three questions—without naming any tool yet.
State Context and Explicit Boundaries
Context, constraints, steps, and verification matter far more than model names. Any background the AI cannot see must be stated upfront.
Specify who the audience is, which facts must be preserved verbatim, what must never be guessed, and the target format. Observable constraints beat vague requests like “make it better.”
Key points
- Context: audience & purpose
- Boundaries: what to keep & what not to do
- Output: structure & acceptance criteria
Exercise Add audience context and two explicit “do not” rules to your Problem Card, then draft the full prompt.
Cross-Check Your First Output
Fluent prose does not equal reliable truth. Cross-verification means testing outputs against source materials and real execution rather than trusting a single explanation.
Split the response into facts, suggestions, and unverified claims. Trace facts back to primary sources and keep explicit question marks where evidence is incomplete.
Key points
- Are stated facts traceable to source text?
- Do recommendations fit your real constraints?
- Which claims still require verification?
Exercise Tag an AI output with three labels: Verified, My Judgment, and Pending Check. Trace at least one key claim to its source.
Leave a Starting Anchor for Next Time
Saving only the final answer throws away the most valuable part of learning. Record the initial question, what failed, how you adjusted, and what is reusable.
Keep one reproducible mini-example and one open question you still cannot explain cleanly. Let real feedback reshape your roadmap.
Key points
- What did you ship this round?
- Which step required human judgment?
- What is the smallest next experiment?
Exercise Write a five-sentence retro and shrink your next step into something you can start within 30 minutes.