AI/ML

Google Cloud Next ‘25: All-in on AI and the emergence of agentic models

Posted by

Sirsha Dey

We are still buzzing from all the excitement at this year’s Google Next event in Las Vegas. This year, the event showcased how Google customers and partners have benefited from building innovative products using Google Cloud and AI-powered platforms.

Here are some of the exciting takeaways from the 2025 Google Next event.

Multi-agent AI models

The event marked Google Cloud’s addition of multi-agent AI models to its Vertex AI product. Available for public review on the Gemini mobile app and Vertex AI, Google Gemini’s 2.5 Pro model is capable of performing complex coding tasks and improving application quality. Google also announced the forthcoming release of Gemini 2.5 Flash as a cost-effective and low-latency AI model.

Among other improvements, Google’s text-to-image model, Imagen 3, has enhanced its image generation capabilities, as well as inpainting to reconstruct damaged or missing parts of any image. Similarly, Chirp 3, Google’s audio generation AI model, has added the “Instant Custom Voice” feature to generate customized voices in less than 10 seconds of receiving the audio input.

Multi-agent system development

Besides AI models, Google is also simplifying the development of multi-agent models with its newly released Agent Development Kit (ADK). As an open-source framework, ADK provides the same capabilities powering Google’s Agentspace tool.

What’s more, ADK supports the “Model Context Protocol” (MCP) that unifies various AI models to interact with data sources and tools. This is more efficient than implementing custom integrations for each model.

Besides ADK, Google Cloud also introduced the industry’s first open-source Agent2Agent (A2A) protocol built for multiple agents to communicate with each other (across diverse frameworks and models).

Enhancements to AI and cloud infrastructure

As part of its continuous advancement in Tensor processing units (TPUs), this year’s event announced the forthcoming release of Google’s 7^th generation TPU, Ironwood, which can improve chip performance by 3,600X. Besides powering the next generation of Google’s AI models, this TPU also improves energy efficiency by 29x.

In addition to Ironwood, Google also announced the availability of its global private network to enterprises across 200 countries. Referred to as the Cloud Wide Area Network (WAN), this network enables “near-zero” latency for Google services like Gmail and online searches for global users. It can also be used to train Google’s Gemini AI model. Cloud WAN can deliver 40% better app performance and reduce the total cost of ownership by 40%.

Google Unified Security

In this year’s Google Next event, the company also introduced the Google Unified Security solution (powered by Gemini AI). This solution combines the benefits of Google’s threat intelligence and cloud security with enterprise browsing.

As an example, this security solution leverages the following Gemini-powered security agents:

The “alert triage” agent performs dynamic investigations into security alerts, thus reducing the manual workload of human agents triaging hundreds of daily alerts.
The “malware analysis” agent investigates any code for safety and identifies any malicious code.

Enhancements to Google Agentspace

Successfully launched in December last year, Google Agentspace has generated a lot of interest from leading enterprises like Banco BV and Gordon Food Services. Matt Jansen of Gordon Food Services explains how the company “recently began our rollout of Google Agentspace to US employees at Gordon Food Service, with the goal of empowering them with greater access to our enterprise intelligence.”

Here are some of the promised enhancements to Agentspace that customers can look forward to:

Integration with Google Chrome

This allows Chrome users to directly find information within their current workflow.
Agent Gallery

The Agent Gallery provides employees with a direct view of all available AI agents – from Google, its business partners, and internal teams – across the entire enterprise.
Idea Generation Agent (available in preview)

This AI-powered agent promotes workspace innovation by enabling employees to develop creative ideas autonomously in their work domain.
Deep Research Agent

This is another AI-powered agent that empowers employees to explore complex topics from internal and external sources, and synthesize this information into reports (with a simple prompt).
Agent Designer (available in preview)

A no-code interface for creating custom agents that connect to enterprise data sources and automates everyday knowledge work tasks. This is the future of Agents. Being able to make fit for purpose agents to each individual or team, we will see massive productivity gains and Agent adoption.
Agent2Agent Protocol

Agentspace now supports the new open A2A Protocol, designed to let agents across different ecosystems communicate with each other. Google is the first to do this and will allow developers to choose the tools and frameworks they want to work with.
AI Agent Marketplace

Watch this dedicated section of the Google Cloud Marketplace for ideas across verticals, quickly getting the best of both enterprise data and the creative agents that are being written and published. Onix can help create and publish your agents to the Agentspace Marketplace.

Onix at Google Next ‘25

For the Onix team, the 2025 Google Next event was the perfect avenue to showcase our capabilities with AI-powered agents.

The highlight of this year’s event was the launch of Onix’s multi-capability agentic AI platform, Wingspan, which has a native integration with Google Cloud AI products and Agentspace. Wingspan combines the power of:

Agentic intelligence
Dynamic decision-making
Enterprise-level governance

With its enterprise intelligence and human-AI collaboration, Wingspan is providing an interesting glimpse into the future of AI agents. In a related session (moderated by Dr. Phil Shelley), Stephen Lee of TELUS, Ninish Ukkan of Arvest Bank, and MaiLin Jan of Wells Fargo discussed how enterprises can bridge the AI gap using Wingspan.

Talking about Wingspan, our global CTO, Niraj Kumar, gave an insightful presentation on “No Data, No AI” at this year’s event.

Looking ahead

Every year, the Google Next event provides a preview of the exciting happenings in the world of AI and cloud technologies. With the proliferation of its AI-powered tools, Google is promising innovation that can transform enterprise-level productivity. For a full breakdown of everything announced at Google Cloud Next ‘25, check out Google’s recap blog here.