@meta
  v: 1
  route: /lesson-2/
  generated: 2026-10-24T10:19:12Z
  ttl: 1d

@intent
  purpose: Lesson 2 teaches: Give AI the whole picture
  audience: visitor, business-owner, ai-agent
  capability: learn, practice, bookmark, take-notes

@state
  lesson: 2
  title: Give AI the whole picture
  key_idea: Give it what a smart new hire would need on day one.
  core_moves[5]{move,how,example_prompt}:
    The material itself,"Paste the thread, the transcript, the file, the screenshot. A description of a document is a guess about a document.",Here's the full email thread. Read all of it before you suggest anything.
    "Who's involved, and why they matter",The same message lands differently with a ten-year client and a new one. Say who's who.,Dana is a ten-year client and hates long emails. Mark is new and watching price. Write to each of them differently.
    What good looks like,"Show an example you liked. It's faster than describing a style, and far more accurate.","Here's a proposal that won us a job last year. Match its structure, not its words."
    The constraints,"Deadline, length, tone, budget, and anything that's off-limits.",Under 150 words. No pricing. Don't promise a date.
    Where it's going,"Who reads the result, and where. A phone between jobs needs something different from a board meeting.","This goes to the owner, who reads it on her phone between site visits."
  why_it_works[3]{reason,explanation}:
    It fills every gap with a guess,"Anything you leave out, the AI invents. A confident guess reads exactly like a fact, which is how bad answers slip through."
    "Say it once, not every time","ChatGPT Projects, Claude Projects and Gemini Gems hold standing context: your business, your customers, your style. Put it there once and stop retyping it."
    More isn't always better,Forty files with no direction buries the two that matter. Hand over what's relevant and say which part is most important.
  common_mistakes[4]{mistake,fix}:
    Describing the document instead of pasting it,"Paste it, attach it, or screenshot it. Let the AI read the real thing."
    Dumping everything with no direction,"Say which part matters most, and what to ignore."
    Forgetting the audience,Name who will read it and where. Then the tone and length take care of themselves.
    Retyping your background in every chat,Put your standing context in a project or custom instructions once.
  exercise: "Think of the email you answer most often. Write down everything a smart new hire would need to answer it well without asking you. That list is your context. Write yours first, then compare."
  exercise_model_answer: "Who usually writes it (existing clients, mostly office managers). What they're really asking (when will it be fixed, and what will it cost). Our policies (same-day response, flat rate for under an hour, after-hours fee in the contract). Our tone (plain, friendly, no jargon, sign with first name). What we never do (promise a time we can't keep, quote over email for anything over $500). An example reply we were proud of."
  further_reading[3]{title,url}:
    Effective context engineering for AI agents (Anthropic),https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents
    What are Projects? (Claude),https://support.anthropic.com/en/articles/9517075-what-are-projects
    Gems in Gemini (Google),https://support.google.com/gemini/answer/15146780
  what_jason_did: "I run two inboxes with different jobs, and I told the AI exactly that. One is where the alerts and vendor noise land. The other is client-facing, and I gave it the names of the people who matter to me and what we're building for each of them. Same AI, very different behavior, because it knows which inbox it's in and who's writing."
  try_today[5]:
    - Pick a task you hand to AI often.
    - Write down what a new hire would need to do it well.
    - "Paste the real material next time, not a description."
    - Say who will read the result and where.
    - Move your standing background into a project or custom instructions.
  storage: browser-only

@actions
  - id: open_practice_lab
    method: GET
    href: /lesson-2/#lab
  - id: back_to_field_guide
    method: GET
    href: /#guide

@context
  > What you get back is capped by what you hand over. Most weak AI answers aren't model problems. They're context problems.

@nav
  self: /lesson-2.agent
  parents: [/.agent]
  peers: [/lesson-0.agent, /lesson-1.agent, /lesson-3.agent, /field-notes-001.agent, /about.agent]