At VentureBeat’s Transform 2025 conference this week, one of the numerous educational and startling panel discussions on AI enterprise integrations featuring industry leaders was moderated by Google Cloud Platform Vice President and Chief Technology Officer ( CTO ) Will Grannis and Richard Clarke, Senior Vice President and Chief Data and Analytics Officer of Highmark Health.

The New AI Stack in Healthcare: Architecting for Multi-Model, Multi-Modal Environments, a program that examined how the two organizations are collaborating to build AI at a level across the large U.S. healthcare system Highmark Health ( based out of Western Pennsylvania ) delivered a rational look at how the two organizations are collaborating to do so.

Additionally, the partnership has onboarded all of these workers and made them into effective users without losing sight of difficulty, regulation, or doctor trust.

How did Highmark and Google Cloud approach this? Find out more in this article.

A Partnership Built on Prepared Funds

Utilizing Google Cloud’s AI models and equipment to upgrade outdated systems, increase domestic productivity, and increase patient outcomes, Highmark Health, an integrated payer-provider program serving over 6 million members.

This initiative’s emphasis on program engineering distinguishes this effort from other initiatives that treat AI as a fundamental shift in how work is done, rather than just another tech layer.

The importance of developing a flexible system earlier was stressed by Richard Clarke, Chief Data and Analytics Officer at Highmark. Nothing is more traditional than a COBOL-coded job system, Clarke remarked, but Highmark has integrated cloud-based AI models into yet those techniques. The outcome is a 90 % load replication without widespread disruption, which makes for smoother transitions and real-time insight into complex administrative processes.

Did Grannis, the CTO of Google Cloud, said that success starts with doing the right thing. He said,” This may take three, four, or five years, but you can conduct the experiments and tests that make AI helpful at a size.”

From Proof-of-Concept to Actual Use

More than 14, 000 of Highmark’s 40, 000+ people make regular use of the company’s internal conceptual AI tools, which are powered by Google Cloud’s Vertex AI and Gemini designs.

These tools are used in a variety of employ cases, from obtaining member records to obtaining evidence for claims running.

Clarke made a provider-side case of contract verification and credentialing. A team part used to personally check a company’s readiness in a previous system.

Artificial now aggregates that information, checks requirements, and produces customized output with references and cultural recommendations.

Why is this deployment rate so high? Using user feedback loops, lively training, and structured quick libraries. ” We don’t really fall instruments in and expect people use them,” Clarke remarked. We demonstrate how it makes their work simpler, and therefore range based on what is gaining traction.

Chatbots versus Agentic Architecture

The transition from chat-based relations to multi-agent systems capable of carrying out tasks end-to-end was one of the session’s most forward-looking themes. This is a shift away from quick-response talk models, according to Grannis, toward process synthesis and automation.

” Think less about having a chat interface,” Grannis said,” Go do this, bring it back, and let me decide.” These agents coordinate multiple models, probably cascading across various functions, from research to process execution to translation.

Single-use officials are currently being tested by Highmark for a particular workflow, and the long-term goal is to integrate them into server systems so they can carry out tasks independently. This will lessen the need for multiple plugs or interface, resulting in centralized control with a wider range.

First, put the task, no the model, in.

Both speakers made a significant psychological shift for businesses: prevent beginning with the model. Start with the task and choose or arrange types correctly.

For much, research-intensive questions, Highmark uses Gemini 2.5 Pro and Gemini Flash for fast, real-time relations, for instance. Even traditional linear models are sometimes employed when they are most appropriate for the task, such as translating individual communications into multiple languages. Your company operations are your IP, Grannis put it. Think about creating concepts to accomplish a task.

Google Cloud is investing in open standards and model-routing features to help this freedom. The most recent Agent Protocol effort, which was developed in conjunction with the Linux Foundation, aims to promote interoperability and steadiness in multi-agent environments.

Realistic guidance for business leaders from all sectors

The participants provided practical advice for those looking to recreate Highmark’s success:

Lay the groundwork today: Invest in system integration and data preparation right away. The reward depends on first planning, even if whole AI deployment is in the future.
1. Avoid creating your own basic models because it is expensive to do so unless your company needs to build foundational models. Focuses on fine tuning and instrumentation for particular use cases.
2. Adopt a platform-focused approach: optimize type use and access. Develop a framework that encourages experiment without sacrificing government.
3. Start with jobs, no tools: Specify the desired outcome first. next match it with the best-fitting type or representative structures.
4. Assess and share: Internal adoption increases when people experience tangible results. Use adoption statistics, find out about success stories, and keep up the library of causes and flows.
5. The future of business AI is job execution, not dynamic insight. Design for action, not only information. Create systems that are able to safely and securely button real-world events.

Looking Forward

The progress made so much offers a model for others in healthcare, and beyond, who want to create scalable, responsible, and very usable AI systems. Highmark and Google Cloud are still working together.

It’s not about dazzling features, Clarke said, but it’s about what really makes people perform their jobs much.

Enterprise leaders who were unable to attend the session can take comfort in the fact that success in conceptual AI is not limited to those with the most ambitious budgets but also to those with the most ambitious plans, versatile platforms, and the will to work smartly.

With VB Daily, you get regular insight on business use cases.

VB Daily has you covered if you want to impress your employer. In order to provide you with the most insightful information on what businesses are using conceptual AI, from regulatory changes to real-world installations, we give you the inside information on what companies are doing with it.

Learn our privacy statement.

Thank you for your subscription. View more VB updates around.

There was a mistake.