Google Cloud unveils self-learning context tools for AI

Google Cloud has introduced self-learning context enrichment and a redesigned Knowledge Studio in preview for Knowledge Catalog, expanding its tools for managing the business context used by AI agents.

The update focuses on how organisations maintain definitions, rules and relationships across internal data and documents. The new enrichment feature analyses interactions between users and agents to identify missing or outdated context, then proposes updates drawn from business documents, code repositories and other sources.

Google positions Knowledge Catalog as a context engine for agents, linking structured metadata such as tables and schemas with unstructured material including playbooks and internal documentation. The aim is to reduce the manual work required from context engineers and data stewards as companies move from a small number of agents to much larger deployments.

The self-learning system can connect to structured data across first-party and third-party sources, as well as unstructured content in Cloud Storage, Google Workspace, GitHub, Confluence and SharePoint. It can propose business definitions, table and column descriptions, join relationships, query examples and task-specific instructions, each with source citations and a confidence score.

According to Google Cloud, the system can run on demand or on a schedule, detect changes in source material, retire obsolete definitions and flag conflicts for review. It can also analyse interaction traces from Gemini agents, Conversational Analytics in BigQuery and custom agents while redacting personally identifiable information.

The tool is designed to identify points where agents struggled, such as ambiguous column lookups, missing definitions, ungrounded metrics or repeated user rephrasings. It then stages proposed fixes so they can be reviewed and, where approved, reused across other agents in the organisation.

Google Cloud also introduced a redesigned Knowledge Studio interface, describing it as a central workspace for the people responsible for curating and governing context. The interface includes a visual context graph, conversational exploration of data relationships and a review process for proposed changes before publication.

Governance controls

Customers can choose whether proposed updates are published automatically and can set the confidence threshold for auto-approval. Changes below that threshold, as well as sensitive updates such as deletions, are routed as change requests for review, editing or rejection by a steward.

Every change is logged and can be reversed, according to Google Cloud. It also said customer data is not used to train its models.

A key part of the system is the use of context sets, curated subsets of the wider context graph for a specific team, domain, user group or agent. Google Cloud said this allows organisations to maintain different definitions for similar business terms across departments, such as sales and finance, while still managing them within one catalogue.

Access to those context sets is controlled through identity and access management rules, with permissions determining which users and agents can read, use or edit them. Agents retrieve context only from the sets they are authorised to use, while access to underlying data remains subject to existing permissions.

Operational focus

The redesigned Knowledge Studio also includes monitoring tools for enrichment runs. Users can track schedules, results, token usage and run duration, and receive email notifications when a run completes, Google Cloud said.

Google Cloud tied the launch to a broader problem in business use of AI agents: keeping organisational knowledge current as definitions change and data estates grow. In many companies, business logic is spread across documents, source code, wikis and data models, making it difficult to maintain a single, trusted set of context for automated systems.

The objective is to shift that work from manual upkeep to a feedback loop based on how agents and users interact with data, Google Cloud said. Internal benchmarks showed Knowledge Catalog made AI agents nearly four times more accurate, with up to 33% lower latency and up to 75% fewer tokens.

Both new features are in preview and will be rolled out to customers over the coming weeks.