AI-NATIVE ORG DESIGN PACK (2026)

Use this to redesign for “agent-operated processes” (AOPs) in 90 days.

PART 1 — PROCESS INVENTORY (1–2 hours)
List your top recurring processes. For each, fill:
- Process name:
- Team:
- Volume/month (tickets, PRs, invoices, leads):
- Fully loaded cost/month ($):
- Cycle time today (p50 / p90):
- Error rate today (%):
- Customer impact (low/med/high):
- Compliance impact (none/low/high):

Pick the top 20 by cost or pain.

PART 2 — SELECT 3 PILOTS (30 minutes)
Choose:
1) Tier 0 (internal-only): e.g., meeting summaries, internal docs
2) Tier 1 (customer-facing text): e.g., Tier-1 support replies
3) Tier 2 (workflow actions): e.g., CRM updates, refunds under $50

Selection rules:
- Must have clear “done” definition.
- Must have measurable outcomes.
- Must have an obvious escalation path.

PART 3 — AOP SCORECARD TEMPLATE (copy/paste)
AOP ID:
Owner (DRI):
Escalation owner:
Risk tier (0–4):
Allowed tools/actions:
Disallowed tools/actions:
Inputs (systems, data):
Outputs (artifacts, actions):
Success metrics (choose 3–5):
- Cost per unit ($):
- Cycle time (p50/p90):
- Error rate (%):
- Escalation rate (%):
- CSAT/NPS impact:
- Incident count:
Guardrails:
- Logging/audit: yes/no
- Golden-set evaluation: yes/no (size: ___)
- Rate limits: ___
- Human approval required: when ___
Rollback plan:
- How to disable:
- How to restore previous workflow:

PART 4 — GOVERNANCE BY RISK TIER (leader checklist)
Tier 0: logging + access controls.
Tier 1: golden-set evals + brand/style checks + human override.
Tier 2: tool allowlist + audit trails + sampling QA.
Tier 3: CI gates + code owners + feature flags + rollback plan.
Tier 4: human approval + compliance sign-off + adversarial testing + quarterly audits.

PART 5 — 90-DAY EXECUTION PLAN
Days 1–14: inventory, pick pilots, assign DRIs, create scorecards.
Days 15–30: build eval set, instrument logs, ship a constrained pilot.
Days 31–60: expand coverage to 30–50% of volume, add sampling QA, measure unit economics.
Days 61–90: decide: scale, re-scope, or kill. Publish learnings and update role expectations.

PART 6 — COMMUNICATION SCRIPT (reduce fear)
- “AI is not a headcount plan. It’s a capacity plan.”
- “Agents can author work; humans (and automated gates) approve and own outcomes.”
- “We will reward people who make the system safer and more reusable.”

If you can’t name the owner, the metrics, and the rollback plan, it’s not a pilot—it’s a risk.