Google enters new era of AI for data with launch of agents
Agentic AI applications for data engineering and data science along with a new underlying knowledge engine aim to make data management and analytics easier and more efficient.
The agentic AI era has arrived for users of Google Cloud's data management and analytics suite.
On Wednesday during Google Cloud Next 25, the tech giant's user conference in Las Vegas, Google launched generative AI-powered agents aimed at making it faster and easier for data engineers, data scientists, analysts and business users to develop applications and analyze data.
The data engineering agent embedded in BigQuery pipelines and data science agent embedded in Google's Colab notebook are now generally available, while the Looker conversational analytics agent is in preview. All the agents are available to BigQuery and Looker users within their existing tiered pricing models at no extra charge.
These are nice enhancements for data engineers in particular. We see strong demand for data management agents. ... Google clearly sees a similar demand among BigQuery users.
Kevin PetrieAnalyst, BARC U.S.
Given that the agents are designed to make it easier and faster to work with data, they are significant for Google Cloud users, according to Kevin Petrie, an analyst at BARC U.S. He noted that his firm's research shows that half of all organizations are considering using agents to address data management needs such as data quality.
"These are nice enhancements for data engineers in particular," Petrie said. "We see strong demand for data management agents. ... Google clearly sees a similar demand among BigQuery users."
Agents, however, are not a panacea for all complexities associated with data management and analytics, he continued. BigQuery remains an intricate platform and takes training to learn.
"Skills and training are a top obstacle to success with all types of AI," Petrie said. "Given this, I'll be interested to see how intuitive these new [agentic] capabilities are and how easily users can learn how to manage them."
New capabilities
Like many enterprises that have invested in developing generative AI tools given their potential to make workers better informed and more efficient, many data management and analytics vendors have developed generative AI capabilities to simplify and speed up using their platforms.
For example, Google Cloud rival AWS provides Amazon Q, a generative AI-powered assistant integrated throughout its data management and analytics offerings.
Agents are applications built using generative AI models in conjunction with an organization's proprietary data to understand the organization's unique attributes. Unlike assistants, which only respond to user prompts, agents can act autonomously to surface recommendations and take on certain tasks.
Google developed its agents for data engineering, data science and conversational analytics -- and will develop more agents in the future -- based on interactions with customers, according to Yasmeen Ahmad, Google's managing director of data analytics.
"The big theme for us is agents, and in the data context, it's agents for every user," she said. "What we've learned over the past year is you need tailored, domain-specific agents that can help with specific tasks that data teams are doing and accelerate their work."
The data engineering agent provides support for developing data pipelines and preparing data by transforming and enriching it. In addition, in preview are capabilities that will help maintain data quality, such as anomaly detection and automated metadata generation.
The data science agent assists with the entire model development process and includes automated feature engineering, intelligent model selection and scalable training.
The Looker conversational agent, built on top of Looker's semantic layer to address accuracy, enables users to interact with data using natural language while providing transparency through explanations about how it arrived at responses.
Each agent is built on top of the BigQuery Knowledge Engine -- now in preview -- and uses semantic search across BigQuery now generally available to assist users.
Aided by the agents, data workers can do in minutes what would have taken perhaps weeks to do manually, according to Ahmad. In addition, agent outputs are improved 50% by the Knowledge Engine, while responses to business questions are two-thirds more accurate due to semantic search, she continued.
Google Cloud's Conversational Analytics agent in action.
Google's approach -- agents tailored to assist specific roles -- is wise and perhaps unique, according to Stephen Catanzano, an analyst at Enterprise Strategy Group, now part of Omdia. However, how much developers can customize the agents to fit their organization's needs will help determine how effective they truly are, he added.
"They hit the high level, but not detailed needs based on environments," Catanzano said. "It's a good start, [and] marketing to personas rather than the task might be a first, from what I have seen."
Meanwhile, adding a semantic layer to all BigQuery data could prove significant, he continued.
"This would make all data much more accessible, especially unstructured data, which would be a significant change to data management," Catanzano said.
Petrie likewise noted that by developing agents geared toward different roles, Google is taking a smart approach to agentic AI.
"Data engineers have different knowledge, skills and problem-solving methods than data scientists or business analysts," he said. "It makes sense to tailor the user experiences for these distinct personas."
Other new capabilities aimed at simplifying data management and analysis that Google unveiled on Wednesday -- some of which are still in preview -- include the following:
Features such as multimodal tables, automated cataloging, metadata generation and BigQuery continuous queries aimed at making it easier to operationalize unstructured data.
BigQuery AI Query Engine to process structured and unstructured data together.
Google Cloud for Apache Kafka to enable development of real-time data pipelines.
BigQuery tables for Apache Iceberg to provide users with an open data lakehouse that can be integrated with BigQuery.
AI capabilities such as intelligent SQL cells in BigQuery notebooks.
Catanzano noted that Google competitors such as Oracle have already added support for Iceberg, so Google is keeping up with the market by doing so as well.
Looking forward
As Google's data management and analytics suite evolves, making BigQuery an autonomous data foundation is a focal point, according to Ahmad.
Some of the features unveiled on Wednesday -- for example, agentic AI and automated cataloging -- are related to the idea. Others, such as additional capabilities that target operationalizing unstructured data, are part of Google's roadmap.
"More and more, you will see us invest in agentic, autonomous capabilities that in five years' time people will look back at and say, 'I can't believe we used to do that,'" Ahmad said.
However, before adding more agentic AI capabilities and features aimed at making data management more autonomous, Google would be wise to make sure those revealed during Next perform as intended, according to Catanzano. Some are still in preview, and it's important to make sure the ones being developed now work properly.
"I'd like to see how these work out," Catanzano said. "It's good thought leadership, but are they really ready for real time?"
Petrie, meanwhile, suggested that Google do more to foster collaboration between data personas such as engineers and data scientists as their organizations invest more in building AI and machine learning tools.
"While it makes sense to have distinct agents for each group, these folks also need to help, teach and learn from one another as their worlds overlap," he said.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.
