Asking well, in brief
Lesson 2 opens the track’s first arc, the four lessons that answer the question of how to use generative AI well. It adapts the prompting class from the same Harvard Kennedy School course, and it pays the promise lesson 1 left open: you learn why the same question can get different answers twice.
The lesson asks nothing technical. Its raw material is a task from the reader’s own week: a cover letter, a pile of reader comments to make sense of, a conversation worth rehearsing before it happens.
The capability: after this lesson, you can build a prompt with three deliberate parts around a real task, explain answer variability and put it to work instead of fearing it, ask for visible steps and show labeled examples when a task turns analytical, and set up a persona that rehearses a hard conversation and then critiques your performance.
What the lesson covers. The anatomy first: task, instructions, context, TIC for short, with context as the part almost everyone skips and the part that helps most. A sticky note of habits follows: start simple, be specific, say what to do rather than what not to do, ask for many options, fence off pasted material, start fresh when a thread turns to sludge, and ask the machine to draft the prompt itself. Then the payoff of lesson 1’s promise: the machine picks among plausible next words, so one answer is a sample, not a verdict, and variability becomes brainstorm fuel, illustrated by Google DeepMind’s December 2023 FunSearch result, where generating huge numbers of candidate programs and keeping the best produced a genuinely new discovery on a long unsolved math problem. Two upgrades come next, asking the machine to walk through its steps and showing it labeled examples, taught through the course’s story of classifying forty reader comments on a Boston Globe article, along with an honest note on how the first upgrade has aged since the course was filmed in 2024. The persona move closes the lesson: mentor, critic, tutor, student, interviewer, simulator, and the four student-tested tips that make a rehearsal work.
Why this order. Lesson 1 built the mental model, a prediction machine writing one word at a time; this lesson turns that model into the field’s most useful everyday skill, since context is exactly the material the machine predicts from. The practice in Clawless runs the whole loop on the reader’s real task. Lesson 3 steps beyond the chat window: assistants that arrive already knowing their job, AI that works from your own documents, and systems that can see images and speak out loud.