Sanity Knowledge Bases and Workflows: Governance for AI Content Operations

Sanity's September product update brings two complementary ideas into focus: an agent needs reliable context, and a team needs a clear process for deciding what happens to its output. Knowledge Bases and Workflows were both described as beta features in the company's announcement.

Knowledge Bases accepts sources such as datasets, owned website URLs and files, builds a map of facts and surfaces contradictions. Workflows describes content stages in TypeScript, with state held in the Content Lake and accessible to both interfaces and APIs. The announcement describes the same permissions and audit trail applying to people and agents.

An agent can retrieve a brand document and a product page successfully and still find two conflicting answers. Retrieval alone does not tell it which price, availability statement or positioning claim should control the draft.

This is where a content team can make a concrete improvement. Choose the owner of a source, record when it was checked and define the scope in which it applies. A regional product condition should not silently become a global promise. A retired presentation should not outrank the current product record because its wording is easier to reuse.

An agent can help identify these inconsistencies, but resolving them remains an editorial decision. The value of a knowledge system depends on how readily a team can correct the underlying facts and understand the sources behind an answer.

Take a multilingual product announcement. An agent prepares translations, a regional reviewer checks terminology, and a product owner approves the claims. The meaningful part of the process is the agreement about responsibility at each stage.

Encoding that process makes the agreement visible. A team can ask where a document is waiting, which review it has passed and what should happen when a check fails. A person and an agent can participate in the same process without pretending they have identical judgment or authority.

Our analysis is that this makes automation easier to expand gradually. A team can start with agents preparing drafts and people approving every change. Later, it can automate a low-risk check while preserving human review for claims that require interpretation. The decision belongs to the content process rather than to an assistant's enthusiasm.

Sanity's announcement says Workflows was free during beta and that early customers were using it in production. That is vendor-reported experience, not a universal readiness guarantee or a promise about future pricing. Organizations should verify access, current documentation and the requirements of their own project.

A practical pilot should include a failure as well as a success: a conflicting source, a rejected translation or a missing approval. The team should be able to locate the affected content, explain the state and resume work without losing accountability.

The industry development here is a closer connection between knowledge, permission and process. Faster drafting becomes more useful when a team can trace its inputs and decide what may move forward. That combination deserves attention from CMS architects and content leaders alike.