Google Cloud introduced updates to its Vertex AI Agent Builder on Wednesday, providing enterprises with more ways to build, scale and govern AI agents.
With the updates to its agentic AI platform, Google said developers can use the Agent Development Kit API — a tool that enables developers to create and deploy AI agents — to build agents more efficiently. They can also use the new managed Vertex AI Agent Engine to scale their agent in production, while governing the agents with new features like native agent identities and security safeguards.
In addition, Google said its AI Mode search service will now be available on the Chrome browser for iOS and Android devices.
Also on Wednesday, Bloomberg reported, based on anonymous sources, that Apple plans to pay Google about $1 billion a year to use Google’s Gemini 1.2 trillion-parameter foundation model to overhaul its Siri AI assistant system.
Strategy to Dominate
The updates to Vertex AI, the larger platform that houses the agent builder tools, demonstrate Google’s intent to continue dominating the web interface market and also compete at the top level in AI software, said Bradley Shimmin, an analyst at the Futurum Group.
The Gemini model family now competes head-to-head with leading generative AI systems from ChatGPT maker OpenAI and its rival, Anthropic, Shimmin said.
“Gemini has very quickly rocketed to the top of the U.S.-based frontier model makers,” Shimmin said.
To remain in the top group of AI vendors, Google has been investing in AI developer tools.
“Google recognizes that they need to build a developer ecosystem if they’re going to succeed,” Shimmin said. “What was sort of a disparate collection of tools sitting under Vertex AI … is turning into what I would say is a very highly visible tool set that makes up that tool chain that is gaining a lot of traction with the developer ecosystem.”
New Tools
Google is expanding that tool set with the release of these new features.
For example, the Build capability enables developers to build more capable agents using Google’s adaptable plugins framework or a prebuilt plugin, including a new one for tool use that helps agents “self-heal,” Google said. The self-heal feature enables an agent to recognize when a tool call has failed and attempt it again. The build capabilities also include more language support, enabling developers to build agent development kit (ADK) agents alongside Python and Java. Another new feature is a set of observability tools that let users track how agents perform, find and fix production problems and interact with the deployed agents.
Developers can now simulate agent performance using an evaluation layer. After scaling, developers have new tools to govern the agents. For example, agent identities enable users to assign their own identities to agents. Users can enforce agent privilege access and establish policies and resource boundaries to meet compliance and governance requirements.
Enterprise Challenges
With the tools, Google is addressing some of the problems enterprises are experiencing with AI, said Torsten Volk, an analyst at Omdia, a division of Informa TechTarget.
“It’s often the small things like ensuring observability for production models, simple identity and access management, and the ability to easily and reliably chain together tools for production use, ideally across the whole organization,” Volk said.
He said Google is going deeper than some competitors in offering a comprehensive developer experience that includes both automation and orchestration capabilities. This is important because many developers are just starting to build agentic tools and applications, so they are not familiar with elements such as self-healing or robust agent workflows, Volk said.
“Winning over developers by allowing them to create production applications successfully is just as important as convincing the operations people of the viability of running enterprise workloads on Google infrastructure,” he said.
Google is providing tooling for developers and wrapping it up within a platform, which is essential, Shimmin said.
“That platform means a lot to the enterprise developer,” he said, adding that this is why AWS has done well with the Amazon Bedrock generative AI platform. “In the enterprise, the way that money gets spent, it often begins with investments by individual practitioners, whether data professionals or data science professionals or just software developers.”
