Google intros array of technical AI agents across its apps

Agentic AI buildup

Google's focus on agentic AI comes as the excitement -- and what some would call hype because autonomous AI agents are largely unproven -- surrounding the technology shows few signs of abating.

AI experts predict that most sizeable enterprises will pursue agentic AI initiatives in the coming years.

"The long-term journey that enterprises are on is to reimagine, essentially deconstruct, their enterprises and reconstruct them in an AI and agent native context," said Chirag Dekate, an analyst at Gartner.

Part of reaching that long-term goal is finding ways to help data engineers and scientists use agentic AI technology to increase productivity. Google, AWS, Microsoft, and foundation model providers like OpenAI are among the vendors considering this problem.

Previously, agentic AI has focused on customer service and developer-centered applications and applications. This is mainly because data transformation workflows can be complex, Dekate said.

Agentic AI for data scientists and engineers

"What you're now starting to see emerge, when you see data engineering agents, data science agents and such ... is to start attacking the more complex problems that we often see in an enterprise context," he said. Such agents help data scientists and engineers build data pipelines and processes.

"The use of generative AI as a means of both augmenting these processes and, to whatever degree possible, automating them is, I have to say as an analyst and as a practitioner, a Godsend," said Bradley Shimmin, an analyst at Futurum Group.

He added that not long ago, data scientists and engineers spent most of their time looking for data in data repositories -- instead of creating applications, workflows and data infrastructure.

Therefore, he said, agentic AI tools, such as Google's, will significantly benefit data scientists and engineers.

Google’s new agents

One new tool is Google's Data Engineering Agent in BigQuery in preview. The agent lets data engineers simplify and automate complicated data pipelines. With the agent, users can use natural language prompts to streamline workflows, from ingesting data from sources such as Google Cloud Storage to transforming data and maintaining data quality.

Another of the new capabilities is the Spanner Migration Agent. This AI-powered service enables data engineers to simplify their migrations from database systems such as MySQL.

Meanwhile, Data Science Agent, powered by the Gemini large language model, includes features such as exploratory data analysis, data cleaning, machine learning predictions and featurization. The agent can plan, execute code, reason about results and present its findings, while a data scientist provides feedback and collaborates, Google said.

Code Interpreter is for business users and analysts; it translates challenging natural language questions into Python code.

Google also launched Gemini Data Agents APIs to orchestrate the different agents and enable data scientists and engineers to connect the agents to their systems. The first API in this group is the Conversational Analytics API. It lets users integrate Google’s analytics platform Looker's natural language processing and code interpreter capabilities into their applications and products.

Data scientists and engineers can create custom agents with the Agent Development Kit and the Data Agents API.

A starting point

Enterprises considering Google's agentic AI toolkit will notice that it allows data science and engineering teams to be more agile and precise for simple workflows. However, for complex workflows such as integrating ecosystems, the vendor will likely address those complexities in the future, Dekate said.

"This is a good starting point for enterprises to think about ways of improving the productivity of their data engineering and data science teams," Dekate said.

Shimmin said the trick with these agentic AI and generative AI (GenAI) models is that users must trust what they're doing.

"These frontier-based models are getting good at tool use, they're getting good at structured output learning," he said.

He added that while AI agents are understanding data better than before, users need to remember that businesses run on semistructured information sitting in or moving between agentic processes.

"It's all this sort of semistructured content and data and large language models that understand how to take things in sequence and make sense of them for context, understand the meaning and take action on them," he said. "It's brilliant."

In addition to the agentic tools for data scientists and engineers, Google introduced Gemini CLI GitHub Actions, a free AI coding agent..

Other developments

Google also expanded its collaboration with Wells Fargo. The companies are partnering to help the financial institution use Google Agentspace to build AI agents in different departments, including corporate and investment banking and customer service.

Meanwhile, Google also recently rolled out some Gemini advances.

On July 29, Google revealed that its video generation model, Veo 3, is now generally available on the Vertex AI GenAI platform. Veo 3 Fast, a faster way of turning text into video, is also now generally available on Vertex AI. Both Veo 3 and Veo 3 Fast will offer image-to-video capabilities this month.

On Aug. 1, Google DeepMind revealed that Gemini 2.5 Deep Think is now more widely available to Google AI Ultra subscribers.

Esther Shittu is an Informa TechTarget news writer and podcast host covering AI software and systems.