@meta
  v: 1
  route: /field-notes/001-overnight/
  generated: 2026-10-24T04:29:16Z
  ttl: 1d

@intent
  purpose: A real case of an overnight AI assignment that failed and how to structure it so it works
  audience: visitor, business-owner, ai-agent
  capability: learn

@state
  field_note: 1
  title: Why didn't the AI finish it overnight?
  story: "The request sounded simple: find a hundred service companies in each industry around town, build a database, set up an email campaign, have it ready in the morning. In the morning it wasn't done. Or it was done in a way nobody wanted. The reaction was the one most people have: this thing doesn't work. The model wasn't the problem. The assignment was. Nobody had said what counted as a lead, where the data could come from, what quality was good enough, or what to do when it got stuck. And a chat window doesn't keep working after you close the laptop."
  better_way[5]{step,time,what}:
    Questions,~5 min,Which industries? What counts as a lead? Which sources are allowed? Who approves sending?
    Sample,~5 min,Ten leads for two industries. Check the format and accuracy.
    Run,overnight,"Only after the sample looks right, and only on a tool that really keeps working."
    Report,morning,"The list, the drafts, and what failed and why."
    Send,you,Nothing goes out under your name until you approve it.
  lesson: "Don't confuse a big request with a complete assignment. Long work needs questions first, a sample, a tool that really runs it, progress reports, and your approval before anything goes out."

@actions
  - id: read_lesson_1
    method: GET
    href: /lesson-1/

@context
  > A business owner handed AI a huge lead-generation job, walked away, and came back to nothing useful. The fix wasn't a smarter model. It was ten minutes of questions and a ten-lead sample first.

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