Pre-deployment checklist for Agentic AI

This checklist covers the minimum controls to validate reliability, safety, security, observability, and maintainability of agentic AI solutions before production rollout. It assumes the system can plan actions, call tools, and operate across multiple turns and sessions. Items are grouped by area so teams can assign ownership and produce evidence (tests, dashboards, runbooks) for each control.

Minimal “Go/No-Go” evidence pack (what to collect before prod)

Golden dataset results + thresholds met
Guardrails are applied
Prompt/template versions + change log
Observability dashboard links + alert rules
Runbooks + kill switch verified
Provider failover test report
Data/privacy sign-off (PII, retention, consent)
A2A support from the start