People literacy curriculum

People — Workforce AI Literacy Curriculum

Design a role-anchored AI literacy program that satisfies EU AI Act Art. 4 (mandatory literacy duty, in force since 2 Feb 2025) AND fixes the "chatbot-only" usage pattern that strands organizations on the wrong side of the MIT 95% GenAI divide.

Anchor: BCG 88/25 manager role-modeling — 88% of managers say role-modeling AI matters, only 25% do it visibly; weekly-AI-use rates jumped 55%→72% YoY in cohorts where managers used the tools themselves. Training without manager role-modeling is theatre.

Verdict vocabulary (stable output contract): Compliant-and-effective / Compliant-not-effective / Non-compliant. This designs the enterprise-wide program; for choosing which single tool to put in front of a specific group, use people-tool-selection.

---

Step 1: Mental-Model Taxonomy

Core question: Do learners know which of the four AI patterns fits their problem — or are they defaulting to chatbot for everything?

The four canonical patterns. Every learner must distinguish them by capability, failure mode, and one example in their own job:

1. Chatbot (single-turn / multi-turn LLM). Capability: open-ended generation, summarization, brainstorming. Limit: no memory across sessions, no access to your data, hallucinates facts. Canonical example: "Draft a customer email from these bullets."
2. RAG (retrieval-augmented). Capability: grounded answers over your documents/policies/tickets. Limit: only as good as the corpus; cannot act. Canonical example: "Answer HR policy questions over the employee handbook."
3. Workflow AI (deterministic chain / function-calling). Capability: scripted multi-step automation with AI in specific nodes; predictable. Limit: brittle to off-path inputs. Canonical example: "Extract invoice fields → validate → post to ERP."
4. Agent (autonomous, tool-using, planning). Capability: decomposes goals, calls tools, iterates. Limit: blast radius; needs guardrails (see imda-4-dimensions-agentic.md). Canonical example: "Resolve this Tier-1 support ticket end-to-end."

Diagnostic for the cohort:

1. Can each learner name which pattern they should use for a given task? (Test with 5 job-anchored scenarios.)
2. Do they know what each pattern cannot do? (Hallucination, no-memory, no-grounding, no-tool — name the failure mode per pattern.)
3. Is there a sanctioned example of each pattern already deployed internally? (If not, training has nothing to anchor to — fix this before content design.)

Output: PATTERN COVERAGE | FAILURE-MODE FLUENCY | INTERNAL ANCHOR EXAMPLES | TAXONOMY GAPS

---

Step 2: Role Segmentation

Core question: Which segments need which depth — and where are the highest-leverage cohorts?

Five segments, each with distinct competency needs. Do not collapse into "all employees."

• Executive (C-suite, BU heads). Pain: capital allocation decisions, narrative-setting, vendor selection. Depth: pattern-recognition + ROI gates + governance posture; not prompt craft. Consult deloitte-cheerleader-to-champion.md: 30% of orgs have champion-level execs; the rest under-invest.
• Manager (people leaders). Pain: BCG 88/25 bottleneck — adoption stalls without manager modeling. Depth: must use AI weekly themselves AND coach team usage. Highest-leverage segment.
• Knowledge Worker (analysts, engineers, marketers, ops). Pain: chatbot-only usage; no awareness of RAG/workflow/agent. Depth: full taxonomy + hands-on + role-anchored exercises.
• Frontline (sales, service, field). Pain: in-flow tools matter more than classroom. Depth: pattern recognition for sanctioned tools only + escalation rules.
• Risk-Compliance (legal, audit, infosec, HR). Pain: must evaluate AI without using it. Depth: EU AI Act + NIST RMF literacy + redress mechanisms. Consult eu-ai-act-essentials.md and nist-rmf-functions.md.

Older-worker overlay: Stanford shows 16% headcount drop in 22-25yo with high AI exposure — older workers' job security depends on demonstrating AI fluency. Provide segment-blind extra office hours; do not create an "older worker track" (stigmatizing).

Output: SEGMENT MAP | HIGHEST-LEVERAGE COHORT | OLDER-WORKER POSTURE | RISK-COMPLIANCE DEPTH

---

Step 3: Competency Matrix

Core question: For each role × each pattern, what is must-know vs should-know vs could-know?

Build the 5×4 matrix. Per cell, label MUST / SHOULD / COULD and name the artifact that proves competency (a deliverable, not a quiz score).

	Chatbot	RAG	Workflow	Agent
Executive	SHOULD	MUST	SHOULD	MUST
Manager	MUST	MUST	SHOULD	SHOULD
Knowledge Worker	MUST	MUST	SHOULD	COULD
Frontline	MUST (sanctioned tool only)	SHOULD	COULD	COULD
Risk-Compliance	MUST	MUST	MUST	MUST

MUST = blocks role performance without it. SHOULD = needed within 6 months. COULD = enriching but not required.

Proof artifacts (not quizzes):

1. Knowledge Worker MUST-Chatbot: ship one sanctioned-tool output to a real stakeholder.
2. Manager MUST-RAG: demonstrate one team query answered against internal corpus.
3. Risk-Compliance MUST-Agent: complete one IMDA 4-dimension review (see imda-4-dimensions-agentic.md).
4. Executive MUST-Agent: sign one tech-buy-vs-build decision with explicit governance rationale.

Consult hiten-skill-library.md: capture proof artifacts into a reusable skill library so each completion compounds organizational capability rather than evaporating.

Output: COMPETENCY MATRIX | PROOF ARTIFACTS | SKILL-LIBRARY HOOK | UNCOVERED CELLS

---

Step 4: Delivery Architecture

Core question: What gets delivered as workshop, e-learning, sandbox, or in-flow tutorial — and how does manager modeling get hard-wired in?

Three delivery channels, each with a specific job:

• Sanctioned sandbox (always-on). A safe environment with sanctioned models + sample data + the four pattern templates pre-built. Suppresses shadow-AI by giving a better path. Track usage as the primary leading indicator.
• Role-anchored exercises (workshop / cohort). 90-minute live sessions per role × pattern, using the learner's actual job artifacts. Not generic prompt-engineering decks. Cohort size ≤15 for managers; ≤25 for knowledge workers.
• In-flow tutorial (frontline + knowledge worker). Embedded coaching in the sanctioned tool itself — "why this prompt failed", "try RAG instead", surfaced at the moment of use. Highest retention channel.

Manager role-modeling tie-in (BCG 88/25): managers are the multiplier. Hard-wire by:

1. Manager-first sequencing. Manager cohort completes 30 days before their team's rollout starts.
2. Weekly visible-use ritual. Each manager shares one AI-assisted artifact in team standup; absence is noticed.
3. Manager metric. Team's sanctioned-sandbox weekly active rate is on the manager's scorecard, not the learner's.

Consult bcg-manager-modeling.md: without this, adoption ceiling is ~25%; with it, 72%+.

Allocate budget by 70-20-10 (see 70-20-10-value-split.md): 10% on the deck/e-learning content, 20% on cohort delivery + manager enablement, 70% on the sandbox + in-flow tutorials + integration with real workflows.

Output: CHANNEL MIX | MANAGER-FIRST SEQUENCING | SANDBOX READINESS | 70-20-10 BUDGET SPLIT

---

Step 5: Compliance & Measurement

Core question: What is the minimum-viable evidence to satisfy EU AI Act Art. 4, and what real metrics prove the program works?

EU AI Act Art. 4 minimum-viable evidence (from eu-ai-act-essentials.md):

• Documented training program tied to role × AI-system-exposure
• Completion records per individual interacting with AI systems
• Content covers: technical knowledge appropriate to role, awareness of opportunities and risks, possible harms
• Refresh cadence defined (annual minimum; on material system change)
• Provider-vs-deployer scope: as deployer, train staff who operate or are affected by the AI system
• Evidence retained for regulator inquiry

Note: Art. 4 has been in force since 2 Feb 2025; enforcement via national authorities from 2 Aug 2026. Non-compliance is not penalty-listed in Art. 99, but documented absence undermines defense in any incident.

Effectiveness metrics (lagging completion is not enough):

1. Sandbox weekly active rate by role. Target: 60%+ knowledge worker, 80%+ manager within 90 days.
2. Pattern-fit accuracy. When learners select a pattern for a new task, do they pick correctly? Measure via 5-scenario pulse quarterly.
3. Applied-use artifacts. Count of proof artifacts shipped per learner per quarter.
4. Shadow-AI suppression. Network/DLP signal of unsanctioned tool usage trending down — if not, the sanctioned path is worse than the shadow path; fix the sandbox.
5. Redeployment outcome. Per accenture-redeployment-roi.md: track learners who moved into AI-augmented roles vs attrition. Redeployment ROI is the program's true business case.

Diagnostic:

1. Is completion >90% but sandbox usage <30%? (Compliant-not-effective. Fix manager modeling and in-flow tutorials.)
2. Is shadow-AI usage flat or rising? (Sandbox is failing — content/access problem, not training problem.)
3. Can you produce per-individual training evidence within 5 working days? (If no, you fail Art. 4 on procedure regardless of content quality.)

Output: ART. 4 EVIDENCE STATUS | SANDBOX ACTIVE RATE | PATTERN-FIT ACCURACY | SHADOW-AI TREND | REDEPLOYMENT COUNT

---

Synthesis

Produce three artifacts:

CURRICULUM SPEC. The 5×4 competency matrix with MUST/SHOULD/COULD per cell, proof artifacts named, delivery channel assigned per cell, owner named per segment.

COMPLIANCE FOOTPRINT. Art. 4 evidence checklist with current status per item; gap list with named remediation owner and date; refresh cadence calendar; retention location.

30-DAY ROLLOUT. Day 1-7: manager cohort kickoff + sandbox stood up. Day 8-21: manager cohort completion + role-modeling ritual launched. Day 22-30: knowledge-worker wave 1 begins; in-flow tutorials live; first metrics pulse.

Verdict states:

• Compliant-and-effective. Art. 4 evidence complete AND sandbox active rate ≥60% AND pattern-fit accuracy ≥70% AND shadow-AI trending down. Continue; widen waves.
• Compliant-not-effective. Art. 4 evidence complete but usage/accuracy lagging. Fix manager modeling, sandbox UX, and in-flow tutorials before adding content.
• Non-compliant. Art. 4 evidence gaps remain. Halt new pattern rollout; close evidence gaps within 30 days; do not allow Risk-Compliance segment to start coverage assessments until their own training is documented.

---

References

All files below live in references/ at the plugin root (${CLAUDE_PLUGIN_ROOT}/references/ when installed as a plugin).

• eu-ai-act-essentials.md — Art. 4 mandatory literacy duty, scope, refresh, evidence requirements.
• bcg-manager-modeling.md — 88/25 finding; manager weekly use as the adoption multiplier.
• deloitte-cheerleader-to-champion.md — 30% champion-level executive sponsorship as ceiling on adoption.
• 70-20-10-value-split.md — budget allocation across content, delivery, and integration/sandbox.
• hiten-skill-library.md — capturing proof artifacts as reusable skills compounds capability.
• accenture-redeployment-roi.md — redeployment outcome is the true business case for literacy spend.
• imda-4-dimensions-agentic.md — agent-pattern competency anchor for Risk-Compliance and Executive segments.

Reference files are bundled with this skill — Claude resolves them by filename regardless of install layout (single-skill or plugin).