Google Cloud has launched self-learning context enrichment (preview) and a reimagined Knowledge Studio experience (preview) in Knowledge Catalog, its context engine for agents. Building on Knowledge Catalog‘s existing Gemini-powered enrichment, the new self-learning capability analyzes agent and user interactions to find context gaps and keep organization-wide definitions, rules, and relationships current. Knowledge Studio is the new command center where context engineers and data stewards can explore a data estate conversationally, navigate a visual context graph, and review proposed changes before they are published to users and agents.

The launch addresses a bottleneck: context curation today is largely manual. Context engineers write prompts and mappings agent by agent, and data stewards manually keep definitions current across thousands of assets — a process that breaks down as organizations move from a handful of agents to thousands, and eventually millions. Static catalogs expect subject matter experts to document every asset and update definitions whenever business logic changes, but experts rarely have the time; when the finance team revises how partner credits are treated, static definitions fall behind and every agent with a hardcoded formula starts giving the wrong answer.

Self-learning enrichment works where business knowledge already lives. It connects to structured data across first-party and third-party sources and to unstructured data in Cloud Storage, Google Workspace, GitHub, Confluence, and SharePoint, linking unstructured sources such as playbooks and wikis to structured metadata (tables and schemas) in Knowledge Catalog. On demand or on a schedule, it proposes business definitions, table and column descriptions, join relationships, golden queries, and task-specific agent instructions, each with source citations and a confidence score. On every run it detects what changed at the source, retires obsolete definitions, and flags conflicts for review. With optimized search, each agent gets the context it needs instead of reasoning over bloated prompts and raw schemas, and agents benefit from shared learning across the organization.

The self-learning loop is driven by interaction traces. Enrichment can be configured to analyze traces from Gemini agents, Conversational Analytics in BigQuery, or custom agents while redacting PII. It pinpoints where agents struggled — ambiguous column lookups, ungrounded metrics, missing definitions, or users rephrasing the same question — then finds the authoritative answer and stages a fix so the same mistake isn’t repeated. The article’s motivating example is ‘What was the net revenue in EMEA last quarter?’, which requires knowing how the company defines net revenue, which tables hold the bookings, and how they join; that knowledge is typically scattered across a finance playbook in Google Docs, a data model in GitHub, and a Confluence page.

Automation is gated by steward control. Auto-publishing can be enabled or disabled at any time, and a configurable confidence threshold (for example, 85%) determines when proposed changes are approved automatically. Anything below the threshold, and sensitive updates such as deletions, becomes a Change Request for a steward to approve, edit, or reject. Teams can provide their own evaluation set to test proposed updates end to end, every change is logged with full provenance and can be reverted, and the source states that Google never uses your data to train its models.

Context sets provide governed, scoped subsets of the context graph — definitions, metrics, join logic, and golden queries for a specific domain, team, agent, or group of end users. A sales agent can be bound to a Sales context set while a finance analyst’s assistant uses a Finance context set, so a term like ‘net revenue’ can legitimately mean different things in different domains; agents can also search across context sets dynamically instead of being bound to a single set. Access to each context set is managed with IAM, agents retrieve context only from sets they are authorized to use, and access to underlying data continues to follow existing permissions. Context sets are built on the Open Knowledge Format (OKF), so an OKF bundle can be imported as a context set and exported back out, keeping context portable.

In Knowledge Studio, the catalog becomes an interactive command center rather than a static inventory. It provides a visual map of the cross-cloud data estate — operational databases, warehouses, semantic models, context sets, and documentation; natural-language exploration of the enterprise context graph that shows exactly how agents reason over metadata; a single place to inspect proposed changes, source citations, and confidence scores, test context sets against real questions, and approve, edit, or roll back change requests with version control; and monitoring of enrichment runs including schedule, results, token usage, and run duration, with email notifications when a run completes.

Based on internal benchmarks, the post reports that Knowledge Catalog makes AI agents nearly 4X more accurate, with up to 33% lower latency and up to 75% fewer tokens. The new capabilities will roll out to customers in the next few weeks; the announcement links to a demo, an early-access signup, and documentation.