KENNY CHIEN / PRODUCTS / AGENT LAUNCH CHECKLIST
Agent Launch Checklist
The pre-flight list I run before any agentic app touches production.
What is the Agent Launch Checklist?
The Agent Launch Checklist is a pre-flight checklist for agentic applications — the ordered list of evaluation, security, observability, and rollback gates I run before any agent is allowed to touch production. It turns "are we ready to ship this?" from a feeling into a series of yes-or-no questions with owners.
It exists because most agent failures in production were predictable before launch: no evaluation set, no permission boundary testing, no cost envelope, no rollback path. The checklist is the condensed hardening phase of my AI Application Launch engagements — the unglamorous work that separates demos from systems, in a form any team can run in an afternoon.
A system you cannot score is a system you cannot ship. The checklist starts there and does not let go until someone owns the pager.
What the gates cover
Evaluation gates
Golden tasks, failure taxonomies, and acceptance thresholds — written down before launch, rerun on every model upgrade. If the eval set does not exist, the launch conversation ends here.
Security gates
Permission boundaries tested as hard as task success: what the agent may act on, as whom, within which limits — plus red-team prompts and a data-boundary review that legal can read.
Observability gates
Tracing on every agent decision, cost envelopes with alerts, and a provenance trail that answers "why did it do that?" — for audit, for debugging, and for the angry-customer call.
Rollback gates
Kill switches, graceful degradation paths, and a named human who carries the pager when the agent is wrong. No agent ships until the path back out of production is as clear as the path in.
Who it's for
Teams about to ship their first agentic application and unsure what "ready" means
Security and platform reviewers who need a structured basis for a go/no-go call
Engineering leaders who want launch criteria that survive their absence from the room
Teams with an agent already in production who suspect they skipped some gates
Status and availability
The packaged edition is in development, but the checklist itself is in active use today. It is free and will stay free — email me and I will send you the current version directly, no list-building rituals attached. If you tell me what you are launching, I will flag the gates that matter most for your case.
Frequently asked questions
Is the checklist really free?
Yes. It is free on request and will stay free. There is no download link yet because the packaged edition is still in development — email me and I will send you the current version directly.
What kind of application does it apply to?
Any agentic application — software where an AI agent plans, calls tools, and acts on real systems. The gates are framework-agnostic and model-agnostic: they test whether you can score, secure, observe, and roll back the system, regardless of which stack produced it.
How does it relate to the AI Application Launch engagement?
The checklist condenses the hardening phase of an AI Application Launch engagement into one document. The engagement is the full path from use case to production with me involved; the checklist is the free distillation of what production-ready means, for teams running that path themselves.
Go deeper
Launching an agent soon?
Email me one line about what you are shipping and when. I will send the checklist and point out the gates teams in your situation most often skip.