On the many pedagogical and amazingly revealing panel discussions about AI company integrations with industry leaders Venturebeats Transform 2025 Conference This week one among the Google Cloud Platform Vice President and Chief Technology Officer (CTO) wants Grannis And Richard Clarke, Highmark Health'S Senior Vice President and Chief Data and Analytics Officer.
This session “provided a practical overview of how the 2 organizations work together to remove AI in the quantity of greater than 14,000 employees of the Great US health system Highmark Health (based on Western Pennsylvania).
In addition, the collaboration has led all of those employees on board and made them lively users without losing sight of complexity, regulation or clinic confidence.
How did Google Cloud and Highmark go? Read on to seek out out.
A partnership based on prepared foundations
HighMark Health, an integrated payer system system that serves over 6 million members, uses Google Cloud's AI models and infrastructure to modernize Legacy systems, increase internal efficiency and improve patient results.
What distinguishes this initiative is the deal with platform -enering – AI as a fundamental shift in doing work and not only one other technical level.
Richard Clarke, Chief Data and Analytics Officer from Highmark, emphasized how essential it’s to construct flexible infrastructure early. “There is nothing legateer than an employment platform encoded in Cobol,” said Clarke, but Highmark even integrated the systems with cloud-based AI models. The result: as much as 90% work load replication without systemic disorder, which enables smooth transitions and real-time inspections into complex administrative processes.
Google Cloud CTO is repeated Grannis that success begins with basics. “This may take three, 4 or five years,” he said, “but when your data is finished, you possibly can perform the experimentation loops and reviews that make AI useful on a scale.”
From the Proof-of-Concept to on a regular basis use
More than 14,000 of the over 40,000 Highmark employees often use the corporate's internal generative AI tools, that are operated by Google Cloud by the Vertex AI and Gemini models.
These tools are utilized in a variety of applications – from the generation of personalized member communication to the retrieval of documentation for the processing of injury cases.
Clarke emphasized a provider page that included login information and contract check. Before that, an worker would search several systems manually to envision the willingness of a provider.
AI now aggregates that data checks the necessities and that the tailor-made edition generates complete complete with quotations and contextual recommendations.
What drives this high adoption rate? A mix of structured input development libraries, lively training and user feedback loops. “We don't just enter tools and hope that folks use them,” said Clarke. “We will show you methods to make your work easier after which scale them based on what traction achieves.”
Aggy architecture about chatbots
One of probably the most future-oriented topics from the meeting was the shift of chat-based interactions to multi-agent systems that may do the tasks of end-to-end. Grannis described this as a withdrawal of fast chat models within the direction of tasks and automation.
“Think less about having a chat interface and saying more about it:” Do it, bring it back and let me determine, “said Grannis. These agents coordinate several models that could be cascading across various functions – from translation to the execution of workflows.
Highmark is currently exhibiting individual use for certain workflows, and the long-term goal is to embed them into backend systems with a view to autonomously perform actions. This reduces the necessity for several interfaces or connections and enables centralized control with a broad range.
Task-first, not model-first
Both speakers emphasized a very important mental shift for corporations: start with the model. Instead, start with the duty and select or orchestration models accordingly.
For example, Highmark uses Gemini 2.5 Pro for long, research-intensive queries and Gemini Flash for fast real-time interactions. In some cases, classic deterministic models are also used in the event that they fit higher – resembling the interpretation of patient communication into several languages. As Grannis put it:
To support this flexibility, Google Cloud invests in model routing functions and open standards. The latest initiative “Agent Protocol”, which was introduced with the Linux Foundation, is meant to advertise interoperability and stability in environments with several agents.
Practical advice for company managers in areas
For those that wish to repeat the success of Highmark, the discussion participants offered specific guidance:
- Lay the muse early: Invest now in data readiness and system integration. Even if the total AI use is removed years away, the payment of early basics depends.
- Avoid constructing your individual basic models: If your enterprise models don’t construct, this is dear. Concentrate on orchestration and nice -tuning for certain applications.
- Take a platform –: Centralizing model access and usage tracking. Create a structure that supports experiments without affecting governance.
- Start with tasks, not with tools: First define the result. Then adapt it with the model or agent architecture that matches best.
- Measure and share: The internal adoption is growing when employees see practical results. Follow the use, record success stories and constantly update libraries of approved input requests and rivers.
- Design for motion, not only information: The way forward for Enterprise AI is task execution, not static insight. Create agents that may safely and safely trigger real actions in your systems.
Look ahead
While the partnership between Highmark and Google Cloud continues to be developing, progress has to this point offered a model for others in healthcare -and beyond -that wish to construct scalable, responsible and highly usable AI systems.
As Clarke summarized: “It's not about striking functions. It's about what actually helps people do their work higher.”
Company leaders who missed the session can fulfill this on this comfort: Success within the generative AI shouldn’t be reserved for those with the best budgets, but for those with the clearest plans, flexible platforms and patience to construct strategically.
