← All cases · System builds system · in operation · since July 2026
Build once, use everywhere
Consistent rules and tools across nine workspaces — the second reconciliation run finished without a single file deviation.
Figure: 9 Arbeitsbereiche im zweiten Abgleichlauf ohne Dateidrift (je rund 15 Kern-Agenten und knapp 30 Kern-Skills) · measured 12 September 2026
Request: „Pull the updates.“
- Agents, skills, hooks and security rules are built once, centrally
- Workspaces pull the current state; deviations are reported instead of silently overwritten
- Local capabilities can be handed back into the catalogue — checked beforehand for secrets, user paths and personal names
- A routing registry decides which workspace and which lead agent takes a request; unknown intents are learned on request
As soon as several teams build AI agents, any organisation ends up with ten variants of the same rule; versions drift apart.
Approvals for sharing; choosing what gets installed.
On install, security rules were assigned a person instead of “everyone” as owner — fixed and added to the consistency check.
Any organisation with several teams using AI agents. Governance becomes infrastructure instead of a recommendation.
When every team builds its own AI tools, ten variants of the same rule appear quickly. Here agents, skills and security rules are built once, centrally, and distributed to all workspaces; what emerges locally can flow back — after a check for secrets and personal names. Intent routing decides which workspace and which lead agent takes a request. The proof: two reconciliation runs across nine workspaces with around 15 core agents and almost 30 core skills each, the second without any deviation. Governance becomes infrastructure instead of a recommendation.
Evidence in my own archive · as of 25 October 2026.