ICMD 90-Day Agent Launch Kit

Purpose
Ship one agent workflow to production with measurable value, bounded risk, and predictable cost.

1) Pick the workflow (Week 1)
- Choose a workflow with a clear object (ticket, invoice, alert, pull request) and a defined finish line.
- Write acceptance tests in plain language (example: “Resolve most routine password reset tickets without escalation”).
- Estimate value using your own inputs: minutes saved per task × tasks per week × blended hourly cost.
- List irreversible actions (send, delete, refund, deploy). Mark them as “confirmation required.”

2) Set boundaries and write policy (Weeks 1–2)
- Choose the agent mode: Suggest / Act with confirmation / Autonomous within policy.
- Create a tool allowlist (keep it small) and a blocklist for dangerous capabilities.
- Define permissions: OAuth scopes, service accounts, and any break-glass elevation flow.
- Specify PII rules: fields to redact, retention limits, and who can access logs.

3) Build tool wrappers and observability (Weeks 2–6)
- Implement typed tool interfaces with strict validation for inputs and outputs.
- Add rate limits, timeouts, retries, and idempotency keys to tool calls.
- Log the full story: user intent, plan, tool traces, retrieved evidence, final action, and timestamps.
- Add replay: the ability to rerun a failed task deterministically with the same inputs.

4) Evaluation and rollout (Weeks 4–10)
- Build a golden set of real tasks with expected outcomes.
- Track: Task Success Rate (TSR), cost per successful task, intervention rate, rollback rate, evidence coverage.
- Define launch gates before rollout (example structure: success stable, rollbacks rare, cost within your ceiling for a sustained period).
- Canary release a small slice of traffic and compare against a control cohort.

5) Packaging and pricing (Weeks 8–12)
- Price on objects (per ticket/invoice/alert) or autonomy tiers (Suggest vs Act).
- Publish cost ceilings per workspace/tenant and give admins controls to cap or pause usage.
- Ship an agent report dashboard: time saved proxies, completion rate, and errors/rollbacks by team.

6) Production readiness (Weeks 10–13)
- Write an incident runbook: disable tool access, stop runs, roll back actions, notify affected users.
- Complete a security review: audit log integrity, least privilege, retention, and vendor risk.
- Create enablement docs: policy templates and a “what the agent can/can’t do” page.

If you complete these steps, you’ll have one shippable agent: narrow, measurable, governed, and economically predictable—ready to expand to the next workflow.