Lesson 03 — A2 Delegation: Deciding What The Rollout Delegates To Claude

Generation-time decisions (logged)

Learning objective

(Bloom's: Analyze) For each of your rollout's three pilot workflows, decide whether Claude runs in Manual, Auto, or is ruled out of Skip mode entirely — from stated capability, cost-of- being-wrong, and reversibility reasoning — replacing your plan's single blanket default.

SSD cycle 1 — the three modes, and the reasoning frame that picks among them

• SAY: Doc-set §7 names three permission modes, verbatim: "Manual" ("Claude pauses and asks for approval for actions"), "Auto" ("Claude keeps working... reviews each action for safety before it runs and blocks anything it determines to be unsafe"), and "Skip" ("Claude doesn't pause to ask and nothing checks its actions automatically"). AIHC.1.A2's reasoning frame is not "which mode sounds convenient" — it's capability evidence (has this workflow actually produced trustworthy output before?) and cost of being wrong (what happens if a bad draft ships unreviewed?), the same reversibility logic doc-set §7 itself uses when it recommends manual approval "when the task touches sensitive files, accounts, or sites" or hard-to-undo actions.
• SEE: A static 2×2 grid: capability-evidence (low/high) × cost-of-being-wrong (low/high), with Manual/Auto/Skip placed in the quadrants where doc-set §7's own safety guidance would put them — Skip in no quadrant Devon's org currently occupies.
• DO: Before rating any of your three workflows, write one sentence per axis: what would high capability evidence look like for a workflow Claude has never run for your org yet? What would high cost-of-being-wrong look like for a document headed to a funder versus one that never leaves your building?

SSD cycle 2 — the Grants-team and Development-team workflows: outward-facing, hard to undo

• SAY: Your grant report goes to a funder; your donor letters go to real donors. Both are outward-facing and hard to reverse once sent — exactly doc-set §7's stated trigger for manual review ("when the task touches sensitive files, accounts, or sites" or the action is hard to undo). Neither workflow has run under your org's supervision yet, so capability evidence is not yet established either.
• SEE: Your two workflows plotted on the Cycle 1 grid: both land in low-capability-evidence × high-cost-of-being-wrong — the Manual quadrant.
• DO: Write the one-line reasoned decision for each: "Grants-team Outcomes drafting: Manual — funder-facing, unreviewed capability, hard to retract once submitted." "Development-team donor letters: Manual — individually addressed to real donors, hard to retract once mailed." Recommended path: Manual for both, for now — (the alternative — Auto once a review streak is established — is available on request, not offered as an open menu yet).

SSD cycle 3 — the Program-team workflow: where Auto is actually defensible

• SAY: Not every workflow lands in the same quadrant — that's the point of deciding per workflow instead of by blanket rule. The Program team's monthly board summary is internal, and your own board process already puts a human review step between the summary and any board decision. That downstream review is real cost-of-being-wrong mitigation your other two workflows don't have.
• SEE: The Program-team workflow plotted on the same grid — lower cost-of-being-wrong (given the existing board-review gate), landing in the Auto quadrant.
• DO: Write the reasoned line: "Program-team board summary: Auto — internal, reversible before the board meeting, existing review gate downstream." Note explicitly why this is not the same reasoning as "reduce staff friction" (your plan's original phrase) — the difference is the downstream gate, not the mode's convenience.

SSD cycle 4 — ruling Skip out, on the record

• SAY: A complete delegation decision also states what you're not choosing and why — not just what you picked. Doc-set §7: Skip means "nothing checks its actions automatically." For an org still establishing capability evidence on all three workflows, that's not a live option right now for any of them — ruling it out explicitly is itself part of AIHC.1.A2's "stated reasons," and it forecloses a staffer quietly switching a workflow to Skip later "to save time" without your reasoning being on record to push back against.
• SEE: The reasoning table from Cycles 2–3, with a fourth row added: Skip — ruled out for all three, reason: "no workflow has established capability evidence yet; ruling in Skip before that exists removes the one check (Auto's safety review) every other mode keeps."
• DO: Write your own one-line ruling-out statement for Skip, in your own words, for the record.

Independent at-bat

Devon's data-handling-policy draft assumes "IT will set a 90-day retention window" (flagged unverified in Lesson 01). Decide, unscaffolded, using this lesson's reasoning frame: should verifying whether that retention control actually exists be delegated to Claude, kept strictly human, or handled some other way — state capability evidence and cost-of-being-wrong for this specific task. (This previews Lesson 06's full resolution — reason it out now on your own first.)

Exit ticket (Bloom's climbs to Analyze; graded against doc-set §7 + his own reasoning)

1. (Remember) Name doc-set §7's three permission modes and one defining phrase for each.
2. (Understand) Why is "reduce staff friction" not a capability-or-cost-of-being-wrong reason?
3. (Apply) State which mode you'd start a brand-new, never-run workflow in, and why.
4. (Analyze — objective level) Compare the Grants-team and Program-team decisions: what single factual difference between them (not a vibe) justifies Manual for one and Auto for the other?
5. (Analyze) Your at-bat verdict on the retention-verification task: what evidence did you use, and would your answer change if your org later moved to the Enterprise plan?

Ledger write

ledger_write:
  learner_id: L2-ADMIN-DEVON
  lesson_id: L2-admin-devon-03-a2-delegation-20260719
  anchors: [AIHC.1.A2]
  exit_ticket: {score: "", bloom_reached: analyze}
  auto_mastery: "0.20 -> "
  auto_score: ""
  self_score: ""
  calibration_gap: " — populates when a learner takes this lesson live"
  journal_prompt: >
    Which of the three workflows was hardest to place on the capability/cost grid, and what would
    have to be true for you to move it to a different mode six months from now?
  structure_used: PBL
  referents_used: [gospel choir]
  next_lesson_seed: "Lesson 04 adds a fourth move on top of these Manual-mode workflows: asking Claude for its own read on rollout risk before the Manual review happens, not just during it."

RUBRIC SELF-AUDIT

Escalation check: no load-bearing indicator fails → auto-ships.