Google Cloud says AI agents grounded in its Knowledge Catalog use “up to 75% fewer tokens”, with “up to 33% lower latency” and “nearly 4X more accurate” answers, citing “internal benchmarks” in a blog post announcing general availability of expanded semantic model support.
The catalog now ingests metric definitions from six systems, by our count. Two are Google’s own, Looker and BigQuery, and four come from other companies: dbt, Snowflake, Databricks and SAP.
Business Engineer Course · The FDE Playbook
An evergreen Business Engineer course on forward-deployed engineering, how AI gets put to work inside organisations. It teaches the general idea only and does not discuss Google or Knowledge Catalog.
The key insight:Google Cloud’s Knowledge Catalog now reads metric definitions from six systems, four of them from other companies (dbt, Snowflake, Databricks and SAP), and Google says agents grounded in it use up to 75% fewer tokens on its internal benchmarks.
Google says Knowledge Catalog “now natively ingests semantic models and metric definitions from Looker LookML, BigQuery Graph (including BigQuery measures), dbt MetricFlow, Snowflake Semantic Views, Databricks Metric Views, and SAP Core Schema Notation (CSN)”. It connects them with schema, lineage, data profiles and a business glossary “in a single context graph, without moving data or rewriting business logic in a new language”.
The post, by product managers Maggy Hu and Paulina Trzeciak, lists each of the six connectors as GA. Google says the catalog grounds “Gemini agent, BigQuery Conversational Analytics, and your custom agents” in the metrics, filters and join paths teams have already defined. A separate tool, Knowledge Studio, is in preview.
Google Cloud’s internal benchmark claims for agents grounded in Knowledge Catalog, indexed with ungrounded agents at 100: nearly 4x accuracy, up to 33% lower latency and up to 75% fewer tokens, per its blog post. The index values are our arithmetic.
The Problem Google Describes
Google opens with an example of an agent asked for net revenue that “counted cancelled orders as revenue and pulled from an unvalidated staging table”. Its diagnosis: the agent “didn’t understand your business”, while that understanding “usually already exists in the form of a semantic model” in tools such as Looker, dbt, Snowflake, Databricks or SAP.
When two tools define the same metric differently, Google says, “both definitions appear with their provenance” so data stewards can reconcile them, and “nothing is silently overwritten”.
The Benchmark Claims
Google says that, “based on internal benchmarks”, grounding agents in Knowledge Catalog “makes them nearly 4X more accurate, with up to 33% lower latency, and using up to 75% fewer tokens”. As an index with ungrounded agents at 100, that is down to 67 for latency and down to 25 for tokens, by our arithmetic.
Why It Matters
Metric definitions are the business logic agents need to answer questions about revenue or costs. Google says the catalog brings semantic models “into one context graph, so every agent works from the same trusted definitions”, reading them from Snowflake, Databricks, dbt and SAP as well as its own Looker and BigQuery. Data plumbing for AI is a recurring theme in our coverage, from Apple’s deal with data-pipeline startup Ascend.io to Google Cloud’s OpenAI model additions.
The Structural Read
As we read the announcement, the product is a shared context layer: the catalog keeps each tool’s metric definitions with their provenance rather than replacing them, and agents read from that graph.
As we read the connector list, four of the six semantic-model sources come from companies other than Google, which places Google’s catalog on top of data platforms it does not own.
As we read the benchmark wording, the latency and token figures are ‘up to’ values from Google’s internal tests, so they describe the best case Google measured.
Google Cloud blog, Knowledge Catalog semantic model support
“both definitions appear with their provenance”
Three Implications
FOR DATA TEAMSGoogle says Knowledge Catalog ingests semantic models from Looker, BigQuery, dbt, Snowflake, Databricks and SAP without moving data or rewriting business logic.
FOR AGENT BUILDERSGoogle says the catalog grounds Gemini agent, BigQuery Conversational Analytics and custom agents in metrics and join paths teams have already defined.
FOR BUYERSGoogle’s post gives benchmark claims from internal tests and no pricing for the expanded support.
The framework says “Salesforce isn’t a CRM; it’s a data layer other systems depend on.”
The framework is the owner’s general argument about enterprisesoftware, not evidence about Google’s announcement; it does not discuss Knowledge Catalog.
What Is Not Established
The accuracy, latency and token figures are Google’s internal benchmarks; the post does not publish the test set, the number of queries or the baseline agent configuration. The latency and token figures are stated as “up to” values, so they describe the best case Google measured. The post gives no pricing for the expanded support.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Google Cloud has made generally available a Knowledge Catalog feature that reads metric definitions from Looker, BigQuery, dbt, Snowflake, Databricks and SAP into one context graph for agents. It says grounded agents use up to 75% fewer tokens and are nearly four times as accurate on its internal benchmarks, which it has not published in detail.
A note on sourcing. Product and benchmark details come from Google Cloud’s blog post on Knowledge Catalog and its X post of 29 October 2026; the benchmark figures are Google’s internal results. Figures labelled our arithmetic are our own calculations.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
