Lesson 11 — D1 Orchestration: Running Two Crews on One Job

Learning objective

Bloom's level: Apply, with an Analyze step. Run two separate Claude conversations for two parts of one real week's work at the same time, keeping each on its own job, and chain draft → critique → revise inside at least one of them.

Plain language: Same as running a rough-in crew and a finish crew on the same job in different rooms — two threads, one week's work, each doing its own job without stepping on the other's.

Standard: AIHC.1.D1 — "Chain AI assistance across the steps of one task — draft, critique, revise — rather than accepting single-shot output." Plain words: chaining and splitting the work, instead of one-and-done.

Your real artifacts today: a new quiz question for an upcoming module, and a fresh news clipping or claim you've brought in this week to check.

SSD cycle 1 — one thread, three passes

• SAY: Anthropic's own getting-started guidance says not to "hesitate to refine a request or ask a follow-up rather than starting over" (doc-set §3, source S1). The standard's own language — draft, critique, revise — is the same three-pass move on a wire run: rough it in, inspect your own rough-in, correct it before the walls close up. One thread, three turns, not one shot and done. Real term: chaining ↔ trade version: three passes on the same run, not one pass and hope.
• SEE: A described single thread shown as three stacked turns: Draft (a new quiz question) → Critique (Frank or Claude flags a weak wrong-answer choice) → Revise (the fixed version).
• DO: Run this three-pass chain live, in one thread, on one new quiz question for your upcoming module.

SSD cycle 2 — a second thread, its own job

• SAY: The free plan includes "memory across conversations" (doc-set §4, source S15) — meaning a second, separate chat can still draw on context Claude has already picked up, without you re-explaining your classroom's format from zero. Same as a job-site logbook carrying context from one crew to the next, even though they're different shifts. Real term: separate thread ↔ trade version: a second crew in a different room — same job, different task, doesn't need to know every hammer-swing the first crew made, just the parts of the plan that affect them.
• SEE: Two described thread icons side by side: Thread A, "quiz-question chain" (from cycle 1), and Thread B, "news-claim check" — different content, same week.
• DO: Open Thread B now, separate from Thread A, and run a fact-check pass on your fresh real clipping or claim — using your trust-rule sheet and labeling habits from earlier lessons — while Thread A sits untouched.

SSD cycle 3 — keeping the crews from tripping over each other

• SAY: Anthropic's own advice to "plan your conversations" and batch similar requests (doc-set §5, source S3) applies just as directly across two threads: decide up front which thread a new question belongs to, rather than dumping quiz material into the fact-check thread. Real term: scope discipline ↔ trade version: keeping the rough-in crew's material list off the finish crew's punch list — a mixed list slows both crews down.
• SEE: A described decision rule: "Is this about apprentice materials? → Thread A. Is this about a claim you need to check? → Thread B."
• DO: Take three new real to-dos from this week and sort each one into Thread A or Thread B before acting on any of them.

Independent at-bat (unscaffolded — his own two-thread split, no cycle text to reference)

Name a real upcoming task of your own that has two distinct parts — for example, prepping for next month's union meeting: agenda talking points on one side, a specific technical code question on the other. Set up your own two-thread split for it, and write one sentence defending why you split it that way rather than running it in a single thread.

Exit ticket (Bloom's climb: Remember → Understand → Apply → Analyze → Apply, objective level)

1. (Remember) What are the three passes in a single-thread chain?
2. (Understand) Why does splitting into two threads risk "tripping over each other," and what's the fix?
3. (Apply) A new to-do lands mid-week: "ask Claude to double-check a claim from the newsletter." Which thread, and why?
4. (Analyze) Compare your cycle 1 chained quiz question to a quiz question you'd have accepted single-shot a few lessons back — what changed?
5. (Apply — objective level) For your at-bat's two-thread split, name one thing that would have gone wrong if you'd used just one thread instead.

Graded against your actual two threads and their real content — never against how organized the split merely sounded before you tried it.

Ledger write

ledger_append:
  learner_id: L4-PUB-FRANK
  lesson_id: L4-lesson11-d1-orchestration-20260719
  standard: AIHC.1.D1
  exit_ticket: {score: "", bloom_reached: apply}
  auto_mastery: "0.30 ->  (projected ~0.55 — profile baseline noted 'no multi-agent experience'; this lesson is single-user multi-thread, a lighter lift than full multi-agent orchestration, and draft-critique-revise was already exercised informally in earlier lessons)"
  self_score: ""
  calibration_gap: " — not computed until a real self-score exists"
  journal_prompt: >
    Same as running two crews — what's the one thing you'd tell a crew lead to keep
    separate jobs from getting tangled, if it were threads instead of rooms?
  structure_used: PBL
  referents_used: [electrical-trade crew-scheduling culture — rough-in crew / finish crew framing]
  next_lesson_seed: "D2 Regeneration, capstone — treat this lesson's own facts as a dated snapshot; build a re-check sheet and run the full six-lesson loop alone on a brand-new artifact."

Rubric self-audit (R1–R13)

Escalation: all load-bearing indicators pass → auto-ship (per playbook §5 policy).