AGENTIC LEADERSHIP OPERATING TEMPLATE (2026)

Purpose
Use this template to roll out AI agents in a way that increases validated throughput (real outcomes) without expanding security, reliability, or reputational risk.

1) Choose the First Workflow (pick ONE)
Workflow name:
Owner (human accountable):
Agent manager (ops owner):
Quality owner (acceptance owner):
Inputs (systems/data the agent can read):
Outputs (artifacts the agent produces):
Out of scope (explicitly prohibited actions):

Good starter workflows (low blast radius):
- Issue -> PR with tests for non-critical services
- Support ticket -> draft reply with citations + handoff rules
- On-call -> incident summary + timeline + next steps

2) Define Acceptance Criteria (must be measurable)
Quality gates (check all that apply):
- Tests pass (unit/integration)
- Lint/static analysis clean
- No secrets/PII in output
- Tone/compliance rules satisfied
- Includes citations/links to internal sources

Truth metrics (pick 3–5):
- Lead time change (%)
- Change failure rate / rollback rate
- Escaped defects (customer-reported) per week
- Rework time per agent output (minutes)
- CSAT and recontact rate (support workflows)

3) Permissions + Blast Radius
Agent permissions (least privilege):
- Read scope:
- Write scope:
- Production access: (Yes/No)
Approval gates:
- Who approves merges/actions?
- What thresholds require escalation (e.g., refunds > $100)?
Rollback/escape hatch:
- How can a human stop the workflow immediately?
- Who is on-call for agent incidents?

4) Observability + Audit
Logging requirements (100% coverage target):
- Inputs captured (where allowed)
- Tool calls recorded
- Output artifact stored
- Human approvals recorded
- Unique run ID tied to ticket/issue

Weekly review ritual (30 minutes):
- Review 5 random runs
- Categorize failures (hallucination, missing context, policy violation, formatting)
- Update templates/runbooks

5) 30-Day Rollout Plan
Days 1–3: Design
- Write workflow spec (1 page)
- Define acceptance criteria + truth metrics
- Set permissions + audit logging

Days 4–14: Shadow mode
- Agent produces outputs, humans do real work
- Sample at least 50 runs
- Measure: acceptance %, rework time, violation count

Days 15–21: Limited write access
- Agent can open PRs / draft replies
- Human approval required for every action
- Add alerting for missing logs or policy flags

Days 22–30: Expand scope cautiously
- Increase volume OR increase complexity (not both)
- Only expand if: >=90% accepted outputs AND 0 critical policy violations

6) Autonomy Readiness Checklist (go/no-go)
Go = all true:
- >=90% outputs accepted with minor edits
- 0 critical security/compliance violations in 30 days
- 100% runs have audit traces
- Humans can override easily; <5% “blocked by agent” incidents
- >=20% end-to-end cycle time improvement with stable quality

If No-Go:
- Reduce scope
- Tighten templates/tests
- Add approvals
- Improve retrieval/context and rerun shadow mode

Accountability Statement (paste into team charter)
“AI agents may execute tasks, but humans remain accountable for outcomes. The workflow owner is responsible for quality and risk, the agent manager is responsible for reliability and governance, and the quality owner is responsible for acceptance criteria and evaluation.”