Google Cloud on Wednesday announced upgrades to its Vertex AI Agent Builder, giving enterprises more options for building, scaling, and governing AI agents.
With improvements to its agentic AI platform, Google explained that developers can leverage the Agent Development Kit API — a tool that helps them build and deploy their own AI agents — to develop agents more quickly. They can also leverage the new managed Vertex AI Agent Engine to scale their agent in production, governing the agents with new capabilities such as native agent identities and security precautions.
It also said its AI Mode search service will now work on the Chrome browser for iOS and Android.
Also on Wednesday, Bloomberg reported, citing anonymous sources, that Apple would pay Google nearly $1 billion a year so that it can use Google’s Gemini 1.2 trillion-parameter foundation model in reworking its Siri AI assistant system.
Strategy to Dominate
The upgrades to Vertex AI, the broader platform that includes the agent builder tools, highlight Google’s commitment to maintaining its dominance of the web interface space and also competing at the highest level in cloud-based AI software, said Bradley Shimmin, an analyst with Futurum Group.
Now, the Gemini model family is on a par with leading generative AI systems from ChatGPT creator OpenAI and its competitor Anthropic, according to Shimmin.
“Gemini has advanced with amazing speed to become one of the leading U.S.-based frontier model makers,” Shimmin said. To stay in competition with the rest of the top-tier AI vendors, Google had to invest in tools for its developers.
“Google understands that they have to create a developer ecosystem if they’re going to succeed,” Shimmin said. “What was kind of a far-flung set of tools sitting under Vertex AI… is now becoming what I would say is a very high-profile tool set that makes up that tool chain and is seeing a lot of adoption in the developer community.”
New Tools – AI Agent
Google is augmenting that set of tools with the release of these new offerings. For instance, the Build capability allows developers to create agents with more capabilities via Google’s open source plugins framework or pre-built plugins, such as a new one for tool use to help agents “self-heal,” added Google.
The self-heal feature allows an agent to detect a tool call that has failed and retry it. Build capabilities also support additional languages, so now, in addition to Python and Java, developers can build agent development kit (ADK) agents. Another new feature is a framework of observability tools that allows users to measure the performance of agents, track and debug production problems, and interact with the deployed Ai agents.
Developers can now evaluate agent performance using the simulation layer in Gym-Gazebo. Down the line, after scaling up, developers have a few new tools to control the agents. Instance agent identity, for instance, allows users to attach their instances with an identity of their own choice. Enforce agent privilege access and define policies and resource limits to adhere to compliance, governance , and code requirements.
Enterprise Challenges – AI Agent
The tools also sustain Google’s efforts to deal with the issues that enterprises face in AI, said Torsten Volk, an analyst at Omdia, an Informa TechTarget division.
“It’s often little annoyances like making production models observable, having easier identity and access management in place, or the ability to easily and reliably chain together tools for everyday work that ideally runs not only for you but across the organization,” Volk noted.
Google is going further than some of its rivals in providing developers a full experience beyond just automation or orchestration, he said. That’s a significant hurdle, since many developers are just beginning to introduce agentic tools and applications, making them uncertain about concepts such as self-healing or resilient agent workflows, Volk noted.
“They also want to be winning over developers and have them able to build production apps successfully — it’s not enough just convincing operations folks that they can run enterprise workloads on Google infrastructure,” he said.
In offering the tooling to developers and covering it in a whole platform, Google is making all the difference, Shimmin said.
“That AI Agent / platform is very meaningful to the enterprise developer,” he continued, which is why AWS has done well with the Amazon Bedrock generative AI platform. “In the enterprise, money tends to get spent in the way that it has historically begun, which is with investments from individual practitioners or just people who need something new to do their jobs.
