How Agentic AI Is Changing Healthcare

In January Owkin, a 10-year-old French AI company, launched its agentic infrastructure for biology based on curated, multimodal patient data (medical imaging, health records and genomic and molecular data) from diverse geographies, collected from more than 800 hospitals over a decade. Using organized, real-world patient information, Owkin says its AI agents can identify biomarkers, interpret complex datasets, and guide clinical trial decision making -while preserving proprietary privacy.

The aim is to accelerate drug discovery, improve clinical success rates, and ultimately deliver better treatments to patients faster by linking patient data from hospitals, labs, and academic institutions to the pharmaceutical industry through intelligent agents.”

It is just one example of how agentic AI — artificial intelligence systems capable of autonomously planning, reasoning, and executing multi-step tasks with minimal human intervention — is reshaping healthcare at every level.

Along with biological reasoning models, pure-play startups and established tech giants are introducing agents that accelerate drug discovery, improve operational efficiency and reduce costs, address the global healthcare workforce crisis, and improve patient outcomes and experience.

Unlike earlier generations of AI that functioned as passive assistants or decision-support tools, agentic AI systems can independently initiate workflows, coordinate across platforms, learn from outcomes, and adapt in real time. This represents a fundamental pivot from AI that suggests to AI that does.

Predictive models powered by agentic AI can identify patients at risk of disease progression or complications, resulting in fewer hospitalizations, reduced healthcare costs, and better outcomes., says a 2025 article in Frontiers, the open science publisher. For instance, AI-based monitoring systems can detect subtle changes in vital signs, predict deterioration, and alert clinicians before critical issues develop, enabling timely intervention. Because agentic AI is goal-driven and adapts over time, it is especially well-suited to handle the complexity of hospital environments and ever-changing patient needs. In addition, agentic AI can optimize hospital resource management by dynamically adjusting staffing, supply distribution, and patient flow based on real-time data , says the Frontiers article.

Building Biological Artificial Super Intelligence

“This is the future of biopharma discoveries,” Thomas Clozel, M.D., CEO and co-founder of Owkin Clozel, said in an interview with The Innovator. “Everything will be agenticized.”

The infrastructure, which is being built in collaboration with academic partners France’s Gustave Roussy and Germany’s Charite Comprehensive Cancer Center, will initially focus on supporting the harmonization and structuring of biomedical data across Europe to enhance scientific collaboration.

Positioned as a flagship pilot for European digital sovereignty in health, the biological AI aims to deliver open, reusable and “immediately impactful” outputs for researchers, clinicians and innovators across the EU, says Owkin. It seeks to demonstrate that European can lead globally in the emerging field of AI-native biology/

“Europe can be number one in biological AI,” says Clozel. While the race for general-purpose LLMs has largely been won by American companies, the field of biology-native reasoning systems remains wide open and it plays directly to Europe’s strengths: its healthcare systems, its academic excellence and its unmatched biomedical data, he says.

“We have the right data and the right tools,” says Clozel. “Our platform, domain expertise and the critical partnerships with top AI companies will put us in a position to deliver tangible impact for the healthcare ecosystem globally.”

Accelerating Drug Discovery

Drug development is notoriously slow and expensive – an average of 10 years and $2.6 billiion per new drug. Agentic AI is beginning to compress this timeline at multiple stages. In January 2025, Owkin dosed its first patient in a Phase I clinical trial of OKN4395, a first-in-class triple inhibitor for solid tumors. The company says its AI operating system was instrumental in advancing this compound from asset selection through to clinical development.

Owkin’s AI has prevented Phase 2 clinical trial futility by identifying high-risk populations, discovered new multimodal disease subtypes, and reduced trial duration by up to three years (35%) through better patient segment identification.

The French company has invested heavily in privacy-enhancing technologies and already has active collaborations with 8 of the top ten global pharma companies, including AstraZeneca, Bristol Myers Squibb, and Sanofi.

Other companies are also using agentic AI to try and accelerate drug discovery. Google DeepMind, creator of AlphaFold, which has been used by over three million researchers in more than 190 countries, demonstrating the scale at which AI can transform biological research, is now using AI agents for drug R &D and discovery tasks.  AI Co-Scientist — a Google DeepMind multi-agent system based on Gemini 2.0 — has already been used at Stanford to identify drugs that could be repurposed to treat liver fibrosis. TxGemma, Google DeepMind’s  initiative enabling developers to write bespoke AI models for drug R&D tasks — including target validation and ADME prediction —  links them into agentic workflows ;and AMIE (AI Medical Interview Engine) DeepMind’s conversational medical agent, published in Nature, can now reason through multimodal evidence and support longitudinal disease management.

Causaly, which focuses on AI for the life sciences industry and competes with Owkin, launched Causaly Discover, an end-to-end agentic AI platform for drug discovery, in March of last year. Its platform autonomously synthesizes vast biomedical data, reasons over research questions and formulates hypotheses with the aim of reducing the time and effort researchers spend sifting through fragmented resources. Meanwhile, Recursion Pharmaceuticals has built, with Nivida’s help, supercomputing infrastructure to enable agents that autonomously design and execute cellular experiments, creating a rapid learning loop that the company says significantly outpaces traditional research methods.

Automating End-to-End Pharmaceutical Workflows

Not just drug discovery, but every part of pharmaceutical company’s operations, are being infused with AI agents. The UK’s causaLens, a member of the World Economic Forum’s Global Innovators community, has built a platform for deploying what it calls “Digital Workers” – orchestrated sets of multi-agentic AI systems designed to automate complex, end-to-end- pharmaceutical workflows. The causaLens agent platform powers applications across pharma engagement optimization, medical writing, pharmacovigilance (detecting, assessing, understanding, and preventing adverse effects or any other drug-related problems)  and regulatory compliance, automating workflows that traditional required manual teams.  The company’s approach emphasizes causal reasoning and full auditability, a critical requirement in regulated healthcare environments.

The company focuses on three areas, CEO Darko Matovski, said in an interview with The Innovator. The first involves the process of taking drugs to market such as how to get doctors to prescribe them, how to monetize them, how to achieve consumer awareness through social media platforms, and accelerating clinical trials. “Clinical trials have so many processes and each one can take weeks and months which together compound to years and decades,” says Matovski. “Our aim is to take something that takes three weeks and make it three hours and do this across the board and shave years off the process.” One area that causaLens is focusing on is procurement.  Bringing in a new vendor to provide syringes for the trial to packaging for a drug can take three to four weeks per supplier. “These are things that have nothing to do with the drug or the patients, but rather all the things around it,” he says. “Our digital workers can make that go much faster.”

The second area of causaLens’ focus is on the manufacturing side. It has created an AI first digital engineer that can do 10% to 20% better than humans at ensuring high-end manufacturing goes as planned.  Matovski cites the example of a personalized cancer drug: patients donate their plasma, which is put into a manufacturing process in which the T cells are trained to kill the cancer, and then their own plasma is inserted back into the patient. The drug has a 60% probability of success. Digital worker AI agents increase that figure to 70%, increasing the odds of getting treatments to more patients.

The final area is running clinical trials. “There is a lot of data that needs to be analyzed everyday,” says Matovski. “Our digital AI scientist can automate this analysis, reducing the time from months to hours.”

causaLens has spent a decade focusing on causal reasoning, AI that understands the physical world and cause and effect, the way humans do. Other companies are now rushing to develop similar technology, referred to as “world models”

“Our superpower is getting the right balance between nondeterministic and deterministic abilities into our digital workers,” says Matovski. “In practice, this means our AI can be creative in solving problems while staying anchored in the physical realities of drug manufacturing — it won’t hallucinate a solution that violates the laws of chemistry.”

CausaLens builds three safeguards into its agentic AI, he says. One is human- in-the loop. “Our digital workers learn from humans on a continuous basis. What used to be in people’s heads is now being captured in our system. The second thing is really important – causal grounding – decisions are tied to areas of physical reality –and then finally when you create multigenic systems there are control loops at each level so that even small errors will be detected.”

The company’s clients include Johnson & Johnson and Syneos Health.

Patient Engagement and Care Delivery

The international healthcare sector is projected to face a shortfall of approximately 10 million health workers by 2030. Agentic AI is increasingly seen not as a replacement for healthcare staff, but as a mechanism to augment their capacity — freeing clinicians to focus on the complex, high-judgment work that requires human expertise.

Hippocratic AI, a U.S. company that is part of the World Economic Forum’s Innovator communities, is positioning itself as the leader in generative AI agents for patient-facing clinical tasks, including post-discharge follow-ups, chronic care management, medication adherence monitoring and care gap identification.

Withing 15 months of commercialization, Hippocratic AI says it completed over 115 million clinical patient interactions with no reported safety issues, across partnerships with more than 50 health systems, payers and pharma clients in six countries.

For example, Universal Health Services deployed Hippocratic AI agents for post -discharge phone check-ins at multiple hospitals. Agents review discharge and medication instructions, screen for new or worsening symptoms and answer patient questions. Average patient satisfaction rating reached 9.0 out of 10, according to the company.  Measurable outcomes reported across deployments indicate a 30% reduction in hospital readmission rates, a 360% increase in chronic care management team capacity and a 2.6x improvement in engagement with Spanish-speaking patient populations.

Assort Health, a San Francisco-based company, has built a patient experience platform powered by specialty-specific agentic AI that focuses on replacing manual, high friction patient touch points – particularly in scheduling and inbound call handling – with intelligent, autonomous agents. The platform handles patient triage, scheduling, FAQ responses, prescription refill requests, lab result delivery, and billing inquiries across phone, text, email and portal channels. In documented deployments, Assort, which has raised $102 million in total funding, says it has reduced patient hold times from over 11 minutes to approximately one minute and cut call abandonment rates by 81%. In addition to improving patient access the company report increasing labor capacity by 48% for healthcare organizations using its platform and improving patient access.

Bonsai Health, a Santa Monica, California-based company, build an agentic platform focused on one of healthcare;s most persistent challenges: patients missing care because practices lack the tools to consistently reach them. Since launch, Bonsai says it has engaged more than 235,000 patients across over 100 healthcare groups and specialty practices, scheduling over 36,000 appointments. The platform uses specialty-trained AI agents to proactively reach out to patients, manage appointment reminders and reduce no-show rates – automating tasks that previously consumed significant staff time.

Innovaccer, a San Francisco, California-based company, applies agentic AI to referral management, automating end-to-end workflows to ensure patients connect with appropriate specialists. The company says it has delivered more than $1.5 billion in documented customer savings across its deployments to date.

Epic, the largest electronic healthcare record vendor in the United States, is making agentic AI  central to its next phase of development. For example, Emmie,  a digital concierge appearing in Epic’s MyChart patient portal that can answer patient questions about test results, educates patients before appointments.

There are a lot of interesting opportunities to make a hospital system more efficient, notes causaLen’s Matovski. AI agents can help with figuring out how many patients are going to show up, how many nurses will be needed, ordering the right amounts of drugs, and automating processes such as patient check in, he says. “AI agents can also help doctors spend more time being doctors and less time taking notes, fill in blind spots, and become a physician partner. The entire hospital system is so inefficient, literally every little process can be automated with AI agents at some point.”

Key Challenges

The opportunity is huge. An estimated $3.8 billion was invested in AI agent startups in a single year, a market projected to grow tenfold by 2030. While health systems are already reporting measurable returns across operational efficiency; patient satisfaction and workforce capacity, the use of agentic AI in the health system is also facing challenges:

• Safety and clinical risk: Agentic AI systems that act autonomously in healthcare must demonstrate consistent, reliable performance. Industry-wide standards for safety validation remain nascent.
• Data privacy and compliance: AI agents must access large volumes of sensitive health information to function effectively. Compliance with HIPAA, GDPR, and other frameworks requires robust safeguards and transparent governance.
• Bias and equity: AI systems trained on non-representative datasets risk amplifying existing healthcare disparities. Diverse training data and transparent algorithms are essential. Owkin’s multimodal patient data drawn from over 800 hospitals globally is designed to address this, but the challenge of equitable AI performance is one that still needs to be addressed by all actors.
• Regulatory uncertainty: The FDA, EMA, and other bodies are still developing frameworks for evaluating autonomous AI systems in clinical contexts. The gap between technological capability and regulatory clarity creates uncertainty for companies and health systems.
• Technology maturity: Most healthcare organizations are not yet digitally mature enough to deploy fully autonomous agents. Infrastructure, interoperability standards and integration capabilities remain significant barriers to scale.
• Clinical validation: While early results are promising, long-term clinical validation data — particularly on patient outcomes over extended periods — is still limited.

Defining The Next Era Of Healthcare Delivery

Governments are beginning to respond to the agentic AI healthcare shift. The U.S. Department of Health and Human Services released a comprehensive AI strategy in December 2025, built on five pillars: governance and risk management, infrastructure design, workforce development, research reproducibility, and care delivery modernization. And, the FDA launched its own agency-wide agentic AI platform on December 1, 2025.

In the European Union, the AI Act’s risk-based framework places particular scrutiny on AI systems used in healthcare, particularly those classified as “high-risk.” This regulatory environment is shaping how European companies — including Owkin and causaLens — design their systems, with an emphasis on transparency, auditability, and human oversight.

The most consequential question is not whether agentic AI can automate healthcare workflows- the evidence suggest it already can – but how it will be governed, validated and integrated into systems that prioritize patient safety and equity. The companies that get this balance right are likely to define the next era of healthcare delivery.

This article is content that would normally only be available to subscribers. Become a subscriber to see what you have been missing