What is go-live readiness?

Go-live readiness is the point at which a new system, and the people who run it, are fit to start real work, judged against agreed criteria rather than a date. For an AI agent, it means the agent holds a current a certificate for each agent it will use, and the people who review its held actions are trained and assessed.

Why does go-live readiness matter for agents?

Most go-live checklists cover the technology: connections tested, access granted, data loaded. With an agent, those checks are necessary and nowhere near enough. The agent will start producing held actions on day one, and someone must decide each of them.

If the reviewers are not ready, one of two things happens. They approve too much because they do not know what to check, or they hold everything and the agent adds work instead of removing it. Either way, the go-live fails on people, not software.

What does go-live readiness check for an agent?

Readiness has two halves, and both must pass:

The agent - A current certificate from replay against the customer's own cases, at the depth it will use. - The depth set by evidence and capped by the role's ceiling, with the lower applied. - The customer's own limits entered and tested. - Sources named, with their provenance class.

The people - Every reviewer assessed on held items from this agent's type of work. - A named owner who can change the brief and sources. - An agreed way to stop the agent, practised once. - A rota that covers the expected volume of held actions.

What does go-live readiness look like in practice?

Take a lender planning to switch on an agent that drafts affordability summaries at depth 2. The technical checks pass two weeks early. The people checks do not: two of six underwriters have not done the assessment, and nobody has practised stopping the agent. The lender moves go-live by a week and runs the stop drill. That is cheaper than the alternative, as argued in a readiness assessment is cheaper than a failed go-live.

Timing the training is its own task. How AI agent training should be scheduled around a go-live answers the question of when, and how to schedule training around a go-live gives the steps.

How is go-live readiness different from operational readiness?

Go-live readiness is a gate: are we fit to start on this date? Operational readiness is a lasting state: can we keep running this well, including covering absences, renewing credentials before they lapse and handling incidents? A team can be ready to go live and not ready to operate, for example if only one person is trained to review.

What happens after go-live?

Readiness is checked again. Within a few weeks, a retrospective asks what training missed. A training retrospective template after go-live and how to run that retrospective cover this. The agent's certificate lapses after 90 days, so its next readiness check is already scheduled the day it goes live.

For the full picture, see training teams for AI agents.