AI-Native Leadership 90-Day Checklist (Operators’ Edition)

Purpose: Turn AI usage into measurable throughput with quality, bounded risk, and predictable costs.

WEEK 1–2: INSTRUMENT REALITY
1) Map current AI usage
- List every AI tool in use (GitHub Copilot, ChatGPT Enterprise, Claude, Notion AI, Gemini, etc.).
- Identify use cases: code, support drafting, content, analytics, incident response.
- Name an owner per function.

2) Baseline outcomes (pick 6–10)
- Engineering: lead time, deploy frequency, change failure rate, MTTR, PR review time.
- Support: AHT, CSAT, escalation rate, re-open rate.
- Product: time-to-decision, time-to-launch, rollback rate.
- GTM/content: cycle time, correction rate, compliance review time.

3) Establish cost visibility
- Require tagging for all model/API usage: team, product, environment.
- Set alerts at $500/month and $2,500/month per team (adjust to your scale).

WEEK 3–6: STANDARDIZE THE SAFE PATH
4) Publish a one-page AI policy
- Data classes: public / internal / confidential / regulated.
- Allowed tools per class.
- Rules for customer-facing output (human approval required).

5) Implement an “AI gateway” plan
- Decide: managed (Bedrock/Vertex/Azure OpenAI) vs custom proxy.
- Minimum logging: request_id, team tag, model, latency, cost estimate.

6) Define quality gates
- Code: tests required for AI-assisted changes; mandatory review.
- Agentic workflows: must have rollback + audit log.

WEEK 7–10: BUILD REUSABLE LEVERAGE
7) Create an internal workflow library
- Store approved prompts/templates.
- Add “how to verify” notes (what to check, common failure modes).

8) Select two high-ROI workflows to productize
Examples:
- Bug triage + reproduction steps
- Support reply drafting with policy constraints
- Release note generation from merged PRs
- Incident summaries + action items

9) Add evaluation (evidence, not vibes)
- Define pass/fail criteria (accuracy, policy compliance, tone, safety).
- Run weekly spot checks on samples.

WEEK 11–13: OPERATIONALIZE
10) Launch a monthly AI throughput memo
Include:
- Topline outcomes (lead time, MTTR, AHT, CSAT)
- Model spend by team
- Quality indicators (incident rate, escalations)
- Two wins + one risk

11) Update hiring and performance signals
- Add an “AI critique” exercise (audit an AI artifact for correctness/safety).
- Recognize AI workflow ownership in promotion criteria.

12) Meeting stack hygiene
- Require a decision statement + proposed answer for any meeting.
- Convert status updates to async AI summaries.

Definition of Done (by Day 90)
- 80%+ of AI/model usage is tagged, logged, and budgeted.
- Each function has 6–10 outcome metrics tracked monthly.
- Customer-facing AI output has explicit approval rules.
- At least 2 workflows are standardized with owners, evals, and rollback.
- Leadership reviews AI throughput and spend as routinely as headcount and uptime.