The lifecycle
1
Framing & POC
Pick a scope that proves business value and is feasible. Validate the idea quickly before investing. Recommendation: agree on shared selection criteria so the next use cases are chosen the same way.
2
Audit & MVP specification
Define the target scope, run a security review, and specify the MVP. Bring in the systems and data the agent will need.
3
MVP
Build an industrializable version: connectors, authentication, evaluation, and monitoring in place. Iterate with real users (UAT).
4
Industrialization
Apply engineering discipline: Git versioning, CI/CD, observability, roles and permissions, security and performance tests.
5
Go-live
Promote to production and make the solution auditable (AI Act, GDPR). Onboard the teams who will use it.
6
Operation & continuous improvement
Keep it running and secure: monitoring, KPIs via Insights, and updates to keep pace with the state of the art.
Deploying it
Deployment runs on two independent cycles: the platform moves through its own Dev, Pre-production, and Production environments with CI/CD, while each use case has its own dev and prod workspaces on the production environment, versioned by Git. This is covered in detail in Deployment strategy, including the workspace-or-self-care choice.Next steps
Deployment strategy
Platform environments, and the two-workspace Git model per use case
Who does what
The teams and rights behind this lifecycle
Opening strategies
How to let people build without chaos