Google Cloud continues to quickly build out its generative AI capabilities and is advancing technology for the adoption and use of AI agents that can automate tasks and collaborate across organizations.
At Google Cloud Next 2025 today, the company announced new tools and services in Vertex AI, such as an agent development kit and Agent2Agent protocol that enable a multi-agent ecosystem. Google's cloud division also unveiled enhancements to Agentspace, a platform announced in December that provides artificial intelligence-enabled search capabilities and agents to enterprise customers.
"Gen AI really evolved from a buzzword into a business essential. We're not just talking about incremental improvements, we're talking about fundamental shifts, big unlocks that will lead to powerful new use cases across the entire healthcare value chain," Aashima Gupta, global director of healthcare strategy and solutions at Google Cloud, said during a media briefing.
AI advancements are moving from chatbots to single-agent systems to multi-agent systems working across multiple functions, Gupta said.
"One of the areas where we are seeing the most advancement in AI is purpose-built agents. At Google Cloud, we see AI agents not just as tool, but intelligent collaborators that can streamline operations, reduce inefficiencies and improve patient care, she said. "Many healthcare organizations already use AI chatbots for basic task-specific functions like answering patient inquiries. These chatbots provide efficiency gains by automating high-volume repetitive tasks, but these are also fairly limited. Purpose-built agents can manage more complex, multi-step workflows in healthcare. They can not only automate tasks, they can do deep research. They can reason and act as a virtual assistant to employees, reducing administrative burden."
"The next frontier that we are very excited about is multi-agent AI systems, where AI doesn't just assist but collaborates across multiple regions. This shift can dramatically improve workflows. Imagine entire systems powered by multiple AI agents working together autonomously to optimize things like revenue cycles, streamlining claims operations, even proactively addressing bottlenecks. This isn't just about automation. It's about creating self-optimizing ecosystems that deliver significant long-term value," she said.
Through Vertex AI, Google Cloud offers enterprise-ready tools to build multi-agent systems and a portfolio of purpose-built agents.
Multi-agent systems rely on models with enhanced reasoning capabilities, like those available in Gemini 2.5, Google Cloud executives said. They also depend on integration with an organization's workflows and connection to its enterprise data.
At its Google Next event, the company announced enhancements to Vertex AI, including an AI Agent Development Kit, an open-source framework that simplifies the process of building sophisticated multi-agent systems while maintaining precise control over agent behavior along with Agent2Agent protocol, which helps enterprises support multi-agent ecosystems.
"Agent2Agent protocol is an open standard that is a framework that will allow agents, no matter what technology stack or tools were used to build these agents to communicate together," Lilly McNealus, director, outbound product management at Google Cloud said during the media briefing.
Google Cloud worked with more than 50 partners, such as Accenture, Box, Deloitte, Salesforce and SAP to ensure that the Agent2Agent protocol works seamlessly regardless of the underlying technology, McNealus said. "We're really excited about what the Agent2Agent Protocol will unlock in terms of the next level of multi-agent systems," she said.
The company also rolled out Agent Garden, a centralized hub for AI agents, including Google’s agents, third-party agents, or even custom agents built by an organization's IT.
"Employees of healthcare organizations can find the absolute right agent they need at the right time to get the job done," McNealus said.
Within the Agent Gallery, Google Cloud is launching a deep research agent, an AI agent designed to synthesize data, providing cited findings in an easy-to-read report. "You can think about healthcare employees within an organization who have to search and synthesize tons of healthcare research as well as data, this all of a sudden becomes a very powerful tool for them to get their jobs done faster. Additionally, for those who are in the forefront of innovation and actually doing scientific research, this becomes a great companion tool that can even push the boundaries of innovation further and faster," NcNealus said.
The Agent Gallery is a feature of Google Cloud's Agentspace, which brings together enterprise search, conversational AI, Gemini and third-party agents to enable employees to find and synthesize information from within their organization, converse with AI agents, and take action with their enterprise applications.
Vertex AI and Google Agentspace can advance the way organizations build and manage multi-agent systems, scaling AI agent adoption across enterprises and accelerating deployment of AI through out-of-the-box, ready-to-use agents," McNealus said.
"The pace of innovation and the ability to adopt and integrate that innovation in those technologies into their workflows is incredibly challenging, and it's not just about keeping up, it's figuring out how to stay competitive in this ever-evolving landscape," she noted.
Agentspace is the entry point for employees of healthcare organizations to access the power of agents, she said.
"It allows healthcare organization employees to seamlessly search across all the information within an organization powered by Google search technologies. It allows employees to then seamlessly analyze and summarize that information, leveraging the power of Gemini multimodal intelligence, and it allows employees to act on that information powered by AI agents," she said, noting that the tool also enables multi-modal search.
"That means that healthcare employees can seamlessly search across not just text, but also videos, images and audio and find precisely what they're looking for in record time," she said. There are also connectors into document repository and database systems to pull together different information across often fragmented data sources.
"Enterprise readiness is at the core of our development of Agentspace," McNealus added. "This means from security and compliance to access transparency to data privacy and regulation and to responsible AI, these are the cornerstones of the way that we have developed Google Agentspace."
How Highmark Health and Hackensack Meridian Health are pioneering AI
While many healthcare providers and payers are still developing their AI strategies, Highmark Health and Hackensack Meridian Health are forging ahead to deploy AI agents and search tools to reduce administrative time, simplify workflows and improve patient care.
This week, Google Cloud also announced a collaboration with Seattle Children’s Hospital to roll out an AI agent that helps physicians and clinicians easily access information from the system’s clinical pathways at the point of care.
Highmark is an integrated delivery system that operates an insurance arm and a health system, Allegheny Health Network. The organization is focused on broad-based adoption of generative AI tools across the enterprise to increase employee efficiency and improve patient and member experiences.
"We already have more than 14,000 of our 40,000-plus employees regularly using our internal generative AI tools powered by Vertex and Gemini. We've surpassed one million prompts," said Richard Clarke, chief data and analytics officer at Highmark Health during the briefing with reporters. "It's been really a cornerstone for us to get comfortable with that, because we're obviously dealing with incredibly sensitive information and data every single day."
Highmark also is placing "bigger bets" on using AI agents to "transform" operations and processes. "Initially that was areas like software development, the prior authorization process, our call center or contact center, and then just absolutely transformational for the industry is everything in the ambient listening space, and that has been a true gift to bring joy back to practice for many of our clinicians," Clarke said.
To date, Highmark Health has focused its AI initiatives on its internal workforce and administrative tasks. "I am very excited for this next phase, and I think it's going to be through agents where we are actually now directly interacting with our members and patients, providing guidance, and really moving to that 'age of abundance', where that type of high-quality guidance is kind of there all the time, always on, in whatever modality that our members and patients wish to seek it," Clarke said.
Hackensack Meridian Health, a health system with 18 hospitals and more than 500 patient care locations, has "hyper-scaled" its AI-enabled capabilities over the past three years, said Sameer Sethi, senior vice president and chief AI officer at the health system.
The health system, which is staffed by 35,000 team members and 7,000 physicians, is using a Gemini-based app that optimizes medical record summarization with Google Cloud's Vertex AI Search tool. Hackensack Meridian is focused on six key areas, including creating personalized experiences for employees and patients, streamlining clinical and administrative efficiencies and addressing clinician burnout.
Hackensack also is using AI technologies for disease prediction and precision treatment, Sethi noted. "We have built various use cases where we are not just early detecting disease, but also creating signals within patient records that allows us to move a patient to a better setting of care. It could be long-term acute care," he said.
For more precise treatments, the health system is working to use AI to "connect the right dots" to create a personalized treatment for a patient, Sethi said.
AI agents enable organizations to go from "insights to action," he noted. Hackensack has developed a nurse AI agent to help nurses sift through large amounts of data fairly quickly and offer next best action recommendations and an agent to assist with post-operative calls to patients.
The next step will be multi-agent systems that have agent-to-agent interoperability to enable organizations to orchestrate different activities and technologies for more complex processes. He cited the example of an orthopedic patient who needs to make an appointment, needs to arrange transportation to the appointment, needs to use a wheelchair once on site and then needs a ride to the pharmacy. Right now, all those tasks are handled by several different human employees. Multiple AI agents collaborating together can automate all of those tasks, Sethi said.
The biggest barrier to agentic AI adoption is building the technology to emulate human functions, according to Sethi.
"It's trying to get to what a human is doing and what of that can be, or should be, agentized or automated? We went through these journeys as a part of robotics process automation phase which we still are on. The biggest barrier is charting out exactly what's needed and how does a human do it? I think building the automation or the agents to actually do it is not the hardest part. It's figuring out how a human does it and then what of this can be and should be automated?"
He added, "It's a very natural, organic and much-needed hurdle, but that's what slows things down."
The building blocks for AI agents have been developed, Clarke noted, "Frankly, some of the barriers I think we all were worried about a few years ago, be it hallucinations or costs, are really just not turning out to be the barriers that maybe we thought they were going to be."
As healthcare organizations look to deploy AI agents, they are focused on three key technology areas, access to data, access to systems and orchestration, Clarke noted.
"That ecosystem access, there's work still to be done there, especially given the legacy nature of all of our companies," he said. "After that, honestly, it's just going to be and each and every institution's own risk tolerance in terms of how much they are ready for these agents to be truly taking action on the behalf of their institutions."
Establishing strong AI governance and ongoing monitoring and auditing of AI tools will be critical as these technologies get adopted, Clarke said.
"We're just going to have to keep kind of building out that muscle over time. I do not expect it to be something that goes from zero to 100 right away because I do think especially in healthcare, we're going to be appropriately measured," he noted.