Google updates agents in BigQuery to further automate analytics tasks

Google has made a series of incremental updates to its managed BigQuery data warehouse service to help data practitioners across enterprises further automate data analytics tasks.

These updates were made to the data engineering and data science agents in BigQuery that the hyperscaler announced in April during its annual Google Cloud Next event.

Data engineering agent moves from basic to advanced

The data engineering agent, according to Google, now includes more features than just basic data preparation capabilities.

“We have now evolved to a full end-to-end agent capability that spans pipeline building, data transformation and pipeline troubleshooting,” Yasmeen Ahmad, product manager of data and AI at Google Cloud, told InfoWorld.

This means that the agent, while accepting input in natural language, can now understand schemas, learn from existing metadata, and grasp the relationships between different data assets, enabling data practitioners to engage with it across the entire data pipeline lifecycle, she added.

These engagements could include asking the agent to perform tasks such as generating a data pipeline, modifying existing pipelines, and even troubleshooting issues, since it can analyze code and logs to identify the root cause of a problem and suggest or apply a fix.

Data science agent integrated into BigQuery Notebook

The data science agent, which was made accessible via Google’s free, cloud-based Jupyter notebook service, Colab, to help data scientists automate feature engineering, is now integrated into BigQuery Notebook, Ahmad said.

This integration will enable the agent to support enhanced capabilities for creating automated end-to-end data science workflows, from creating multi-step plans through generating and executing code, reasoning about the results, and presenting findings, she said.

Autonomous vector embeddings and generation within BigQuery

In order to help enterprises automatically prepare and index multimodal data in BigQuery for vector search, Google has introduced autonomous vector embeddings and generation inside the data warehouse.

“When we say ‘autonomous’, we are referring to the automation of the complex, undifferentiated heavy lifting of data engineering and MLOps,” Ahmad explained. “In a traditional workflow, a data science team would have to manually extract data, set up compute, batch data for API calls, and build and tune the vector indexes.”

This feature will help enterprises free up data science teams to focus on the higher-value work of selecting the right models and validating effectiveness against business outcomes, she added.

Further, she believes that these embeddings will help enterprise users build a long-term semantic memory for data agents.

HyperFrame Research analyst Stephanie Walter agreed with Ahmad’s assessment of the capability.

“Autonomous vector embeddings transform unstructured and multimodal enterprise data into vectors, making semantic search, similarity comparisons, content recommendations, and anomaly detection possible at scale. These capabilities are crucial for building and accelerating advanced AI-powered solutions,” Walter said.

She added that rival hyperscalers and vendors also have similar offerings, such as Microsoft’s Azure Cognitive Search and Synapse, AWS’ Amazon OpenSearch Serverless, Snowflake’s Cortex, and Databricks’ Lakehouse AI.

AI query agent now in preview

At Cloud Next, Google also introduced an AI query engine inside BigQuery to help data practitioners analyze structured and unstructured data together.

This AI query engine, which had been in early tester phase for select customers, is now officially available in preview for all customers.

Looker’s conversational analytics agent gets a code interpreter

In addition, at the April conference the company announced the addition of a conversational analytics agent to Looker to help business users ask questions of their data in natural language.

Google is currently updating that agent with a new code interpreter, to allow business users to ask more complex questions of their data in natural language, without any support from IT.

The new Gemini-powered code interpreter capability, which is currently in preview, will help the agent generate code, provide clear natural language explanations, and help create interactive visualizations, according to Ahmad.

“Code interpreter allows users to ask more complex ‘what if’ and scenario based natural language questions which can now be processed through advanced code and workflows written in Python by the agent,” she said.

The interpreter is also being made available to the conversational analytics API that can be used to integrate Looker’s natural language processing capability into an enterprise’s applications and workflows.

This API, which was announced in private preview at Cloud Next ’25, is now being made available in public preview to all customers and partners.