...
By Jake Cole
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
Health care organizations have accelerated investments in digital transformation as they work to modernize operations and improve service delivery. Recent industry data shows that 40 percent of organizations are committing between $50 million and $100 million annually to digital technologies, and 66 percent report actively deploying new digital solutions. These efforts reflect a broader push to strengthen operational efficiency, enhance patient services and streamline administrative processes.
As these investments expand, artificial intelligence has become increasingly integrated into health care delivery. AI now supports a wide range of applications, from predictive analytics to clinical decision support. Predictive AI, which has been used in health care for decades, applies machine‑learning techniques to estimate future outcomes such as readmission risks, early disease indicators and treatment recommendations.
However, the momentum behind digital and AI adoption has also revealed a significant readiness gap. Despite growing interest in advanced technologies, an HIMSS Market Insights survey shows only 18 percent of health care organizations consider themselves prepared for AI implementation.
Read More on Essays
This disparity underscores the challenge of moving from experimentation to dependable enterprise use, where infrastructure, governance and operational alignment become essential. For actAVA, a health care-native agentic AI lifecycle management platform founded by health care and AI technology leaders, the central challenge is to develop the systems needed to make AI reliable, governed and adaptable within complex health care environments.
As part of this mission, actAVA has introduced Cura, a specialized one‑trillion‑parameter model built specifically for agentic health care. Cura is designed to help enterprises transform their own institutional knowledge, including workflows, policies, tools and operational expertise, into intelligence they fully own. It represents a large‑scale demonstration of a vision to enable health care organizations to stop renting generalized AI and instead build domain‑native systems capable of clinician‑grade communication, expert clinical reasoning and reliable execution across complex administrative and clinical workflows.
“We believe the future of enterprise AI won’t be defined by a single model rented by everyone, but by thousands of specialized systems owned, shaped and continuously refined by the enterprises that rely on them,” says Kevin Riley, CEO and co-founder of actAVA.
“Health care has always been a domain where reliability matters because decisions connect directly to people, processes and regulations. The next phase of AI requires organizations to think beyond model capability and focus on how intelligence is deployed, monitored and improved inside real operational environments.”
The conversation around enterprise AI has often focused on larger models and advanced reasoning capabilities. However, health care workflows introduce challenges that extend beyond model performance. Administrative and clinical operations involve extensive policies, multiple systems, specialized roles and decisions where accuracy and accountability remain essential.
CTO and co-founder Frank Wang describes this transition as a shift from developing AI capabilities to engineering dependable AI systems. “Building a compelling demonstration of AI can happen quickly, but creating systems that perform consistently across thousands of real-world scenarios requires a different discipline,” Wang says. “The future of enterprise AI depends on orchestration, evaluation and continuous improvement as much as it depends on intelligence.”
That perspective reflects a broader industry consideration, wherein health care organizations need infrastructure that connects models, workflows, data and governance processes. Agentic AI systems, which can perform multi-step tasks by coordinating tools and actions, introduce new opportunities for automation, but they also require safeguards to operate within defined boundaries.
...
actAVA focuses on building the systems needed to manage AI agents throughout their lifecycle, from how they are created to how they are evaluated and governed. This includes tools that help agents follow policies, work across multi‑step processes and operate safely within health care settings. It is also developing specialized language models shaped around the workflows and requirements common in health care organizations.
A key theme behind these developments is ownership. Health care organizations often possess extensive operational knowledge within their data, processes and institutional expertise. Wang argues that future enterprise AI systems will increasingly involve organizations maintaining greater control over their workflows, models and knowledge assets. The leadership team notes that regulated industries tend to rely on predictable, repeatable operational patterns, which may lead to an emerging “90/10” architectural shift, where roughly 90 percent of enterprise AI workloads run on commodity or open‑weight models for repeatable workflows, while only 10 percent rely on frontier models for novel or highly complex use cases.
CAIO and co-founder Weiran Yao believes this shift reflects a broader evolution in how organizations build with AI. He says, “AI is changing the relationship between expertise and technology. The opportunity is creating systems where organizational knowledge can become an active capability while maintaining the governance and evaluation needed for responsible deployment.”
The importance of evaluation has become increasingly visible as researchers examine how AI agents perform in realistic health care settings. As part of this broader effort, actAVA researchers collaborated with health care professionals and academic partners to develop χ-Bench (CHI-Bench), a benchmark designed to evaluate whether AI agents can complete complex, long-horizon health care workflows. The benchmark examined 75 tasks across provider prior authorization, payer utilization management and care management processes.
Rather than focusing on isolated AI capabilities, χ-Bench was designed around conditions that reflect enterprise health care operations, including multi-step processes, policy requirements, role transitions and interactions with simulated health care applications. The study found that the strongest-performing agent framework resolved 28 percent of tasks at pass@1, while performance declined further when consistency across repeated attempts was required.
These findings highlight why deployment infrastructure has become an important area of focus. Health care workflows are not limited to answering questions or generating information; they require systems to interpret policies, manage context across multiple stages and execute actions within established operational boundaries.
According to Yao, the development of benchmarks such as χ-Bench represents a growing effort to evaluate AI systems against practical enterprise requirements. By testing long-horizon workflows, researchers can better understand where AI agents perform effectively and where additional engineering, governance and oversight mechanisms are needed.
For health care organizations exploring AI adoption, the next phase may depend on building a foundation that connects technological capability with operational responsibility. As AI systems become more embedded in health care environments, the ability to evaluate performance, manage risks and adapt workflows may become as important as the models themselves.
The continued evolution of health care AI will likely depend on how effectively organizations combine advanced models with governance, evaluation frameworks and operational readiness. As the field develops, progress may be measured not only by the sophistication of AI systems but also by their ability to support reliable, accountable and practical use across complex health care environments.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines