If intent changes, systems must change safely.
Brands evolve: new products, new markets, new regulations, new risks. Versioning turns change into a controlled process rather than an uncontrolled ripple across teams.
In AI-enabled operations, this becomes more important, not less. A small change to a policy definition can affect behaviour across multiple tools, channels, and workflows. Without versioning, organisations lose the ability to explain what changed, who approved it, and why the system behaved the way it did.
What to version
- Definitions (concepts and vocabulary)
- Constraints (rules and exceptions)
- Personas and contexts
- Derived policies (bundles for systems)
It is also useful to version the supporting evidence around those changes:
- the tests that were run
- the rationale for the change
- the approver or accountable owner
- the systems or use cases affected
What versioning gives you
- Rollback when something breaks
- Auditability of who approved what
- Reproducibility (why outputs looked a certain way)
- Safer experimentation with temporary campaigns or local-market exceptions
- Clear evidence when a compliance, brand, or legal question appears later
Versioning in a governed operating model
Within the Brando, versioning is not just a repository habit. It is part of how governance is expressed operationally. Definitions, policies, and release states need to move together, so that the organisation can connect specification, deployment, and assurance.
That is also where the implementation layer becomes useful. If deployment is controlled, the organisation can know which version is active in live operation and respond safely when a problem appears.
Without versioning, it is impossible to prove which brand rules were in effect at the time of an incident.
A practical versioning rule
Version the things that change behaviour. That includes brand definitions, claims rules, approval routes, prompt instructions, policy bundles, evaluation rubrics, tool permissions, and exceptions. If changing it could alter what an AI system says, does, blocks, or escalates, it belongs in the versioned operating model.