Google embeds AI agents deep into its data stack - here's what they can do for you

Joan Cros/NurPhoto via Getty Images

ZDNET's key takeaways

• Google is introducing powerful tech for agents and data.
• They are also introducing a series of data-centric agents.
• A new command-line AI coding tool is now available.

I am no stranger to hyperbolic claims from tech companies. Anyone who's on the receiving end of a firehose of press announcements related to AI understands. Everything is game-changing, world-changing, the most, the best, yada, yada, yada.

And then there's Google. Google is no stranger to hyperbole. But when a company so steeped in data management as part of its core DNA talks about "fundamental transformation," and says that the world is changing because, "It's being re-engineered in real-time by data and AI," we can consider those claims as fairly credible.

Also: Got 6 hours? This free AI training from Google can boost your resume today

Just in time for Google Cloud Next Tokyo 2025, Google is making a series of announcements that herald a major change in how enterprises manage data.

Yasmeen Ahmad, Google's managing director of Data Cloud, says in a blog post, "The way we interact with data is undergoing a fundamental transformation, moving beyond human-led analysis to a collaborative partnership with intelligent agents."

She calls this agentic shift, which she describes as, "A new era where specialized AI agents work autonomously and cooperatively to unlock insights at a scale and speed that was previously unimaginable."

Also: 5 ways to successfully integrate AI agents into your workplace

From almost any other company, claims like this would seem like just so much hot air. But Google is dropping a series of announcements about new offerings that provide real-world capabilities to data scientists and engineers in pretty tangible support of the claims.

The agentic shift

There's a fairly fine line between AI chatbots and AI agents. Chatbots are conversational, while agents are tools that perform autonomous tasks. Some users employ chatbots to perform tasks, as I did when I used ChatGPT to analyze some business data. Agents, like ChatGPT Agent, use a conversational interface to receive instructions.

A good way to think of agents is as surrogate team members. Perhaps one agent does data normalization (cleaning up data), while another does migration. Each agent does one or more defined tasks using AI capabilities.

Also: Want AI agents to work together? The Linux Foundation has a plan

In this context, Google is looking at agents that can automate and simplify tasks for data workers, can communicate with each other, and can free professionals from tedious work so they can focus on "higher-value tasks." Google is also trying to get agents to work together in virtual teams.

There are, of course, questions about whether agents aren't actually freeing up the time of senior professionals, but are instead taking work away from more junior employees. On the other hand, I don't have anyone to do the grunt work when I'm fully immersed in a project. So anything I can hand off to an agent is more time for projects and writing.

Cognitive foundation

With all these agents running around, traditional databases just aren't up to the task of keeping them fed. Agents do their reasoning or automation processes across silos. They need access to both historical and live data.

Classic data management methods like real-time OLTP (online transaction processing) and deep-dive OLAP (online analytical processing) isolate data too much for AIs to gain insights from trends and current activities.

One way to help unify all of these capabilities is by enhancing their database offerings. A few years ago, Google added a columnar engine for AlloyDB. AlloyDB is the company's fully managed database service on Google Cloud Platform that focuses on PostgreSQL users, which is ideal for those who require a PostgreSQL-specific solution.

Also: How AI agents can generate $450 billion by 2028 - and what stands in the way

A columnar engine is one where workloads query specific columns of data, reading only the fields needed for analysis. This leads to faster queries and allows for vectorized execution, where operations are applied to an entire column of data at once.

Now, Google is adding a columnar engine to Spanner, its globally distributed, strongly consistent database service that offers high availability and scalability, designed for enterprises needing global reach and high transactional integrity.

This also adds power to BigQuery, Google's serverless, highly scalable, and cost-effective multi-cloud data warehouse designed for business agility. As the name implies, BigQuery is ideal for those who need to run fast, SQL-like queries on large datasets.

The company says this new columnar capability in Spanner speeds up analytical queries by something like 200x on live transactional data. With performance like that, we're talking instant responsiveness to real-time situations.

When building enterprise-based AI systems, you need agents to make decisions based on real data. Performing real-time actions based on hallucinated data can get ugly very quickly. This is where RAG (retrieval augmented generation) comes in. Essentially, RAG combines large language models with real-time data access.

Also: 5 ways to be a great AI agent manager, according to business leaders

You can start to see how vectorizing search in Spanner and BigQuery becomes necessary when you're feeding in real-time data along with historical information. But getting vector search to work efficiently has traditionally been painful. Google is adding adaptive filtering in AlloyDB to automatically maintain vector indexes and optimize for fast queries on live operational data.

Google is also introducing autonomous vector embeddings and generation to BigQuery, which automatically prepares and indexes multimodal data for vector search. This is a key step in creating a sort of semantic memory for agents.

The company is also introducing the ability to run AI queries right inside of BigQuery. This is, ahem, big. Now, BigQuery users can have AI do its magic across giant gobs of structured and unstructured data, ask complex questions (including subjective ones like "Which customers are frustrated?"), and get answers directly within existing analytics tools.

New agentic capabilities

In addition to building a foundation for agentic cooperation and data access, Google is announcing a series of new capabilities that embed agents in their biggest data tools. Let's look at each in turn.

Data engineering agent: Built specifically for data engineers, this agent within BigQuery can simplify and automate complex data pipelines. The entire workflow can be driven by natural-language prompts, from data ingestion to transformations to data-quality assessment to normalization.

Spanner migration agent: Related to the data engineering agent, the Spanner migration agent can simplify data migration from legacy systems to BigQuery. This sort of migration is normally exceptionally tedious and potentially dangerous, but now the agent can do most of the heavy lifting.

Data science agent: Data scientists focus on analyzing and interpreting complex data, while data engineers focus on data infrastructure. According to Google, the new data science agent "triggers entire autonomous analytical workflows, including exploratory data analysis, data cleaning, featurization, machine-learning predictions, and much more. It creates a plan, executes the code, reasons about the results, and presents its findings, all while allowing you to provide feedback and collaborate in sync."

Code interpreter: Built as an enhancement of the conversational analytics agent introduced last year, the code interpreter takes in business-analysis questions and converts them to Python code to prepare custom analysis for users. This all runs within Google Data Cloud and uses the Google Data Cloud security infrastructure. It also includes an API available for developers to incorporate conversational analytics agent and code interpreter capabilities in custom code.

New command-line coding tool

As part of this big series of announcements, Google is introducing an extension to Gemini CLI called Gemini CLI GitHub Actions.

CLI stands for command line interface, basically a terminal interface to your computer. Even though most users left the terminal behind when MS-DOS migrated to Windows, coders to this day make heavy use of the command line. Working in terminal mode lets coders add tools and control the coding process much faster than when they have to find and select items from menus and icons.

Also: Bad vibes: How an AI agent coded its way to disaster

Last month, when Google introduced Gemini CLI, it basically made the features of the Gemini chatbot available in the terminal. Now, Google has extended that capability, providing some agentic features within the terminal environment.

Some of you may be wondering how this compares with Jules, the Google coding agent I wrote about in May. First, Jules works in a secure cloud VM, while Gemini CLI GitHub Actions runs in terminal and integrates with GitHub Actions (the GitHub-based workflow tool).

Google says there's a fairly narrow scope to Gemini CLI GitHub Actions compared to Jules. Jules can read your entire codebase, plan and present an approach to a coding challenge, and then execute on it. Gemini CLI GitHub Actions is specifically targeted to intelligent issue triage, accelerated pull-request reviews, and on-demand collaboration.

Also: Most developers use AI in their daily workflows - but they don't trust it, study finds

The issue-triage capability helps coders manage specific bug reports and feature requests. Pull requests are the way GitHub asks coders to confirm integrating coding changes into branches and master codebases. On-demand collaboration is essentially setting up a chat session whenever you want to talk about your code.

I could easily see a programmer use both. Jules would be great for bigger projects and larger swings, and Gemini CLI GitHub Actions would work well for quicker updates and fixes.

Are agents a game-changer?

What do you think about the agentic shift Google is promoting? Have you started integrating intelligent agents into your own workflows? Which of Google's new data tools or capabilities intrigues you most -- the data engineering agent, the in-query AI reasoning, or something else? Do you see agents as helping senior professionals, replacing junior roles, or both? And how do you feel about running AI workflows directly in BigQuery? Let us know in the comments below.

Also: AI agents will change work and society in internet-sized ways, says AWS VP

Want more stories about AI? Check out AI Leaderboard, our weekly newsletter.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.