Model 07 · The AI Agent Lifecycle

Agents aren't apps and they aren't models — they're a third thing, and the operating models we've built for software and analytics don't quite stretch to fit. Treat an agent like an app and you under-govern it; treat it like a model and you over-engineer it.

This model gives enterprise teams a shared spine across six stages — Ideate, Risk-Tier, Build, Govern, Operate, Retire — so agents move safely from first whiteboard idea to graceful sunset.

The AI Agent

Lifecycle defines what good looks like at each stage — the gates, the owners, the evidence — and answers the question most programmes haven't yet: how does an agent get retired without leaving orphaned permissions, stale integrations, and a quiet trail of unowned automation behind it?

The result is agents as a manageable class of enterprise asset — with the same operational rigour we expect of any other production system, just shaped to the way agents actually behave.

Download the model doc for the stage-by-stage definitions, the gate criteria, the role map (who owns what), and the retirement playbook most programmes haven't written yet.