Revolutionizing Cancer Care: Microsoft’s Healthcare Agent Orchestrator Transforms Oncology Workflows

In the rapidly evolving landscape of healthcare technology, Microsoft’s Healthcare Agent Orchestrator emerges as a beacon of innovation, aiming to transform the traditionally complex and labor-intensive process of cancer care planning. As healthcare systems worldwide grapple with the growing demands of personalized medicine, this orchestrator’s entry into the market signals a bold attempt to democratize access to high-quality, tailored cancer treatment—meeting clinicians where they work, within the digital environments they already trust.

The Problem: Fragmented, Labor-Intensive Cancer Care Planning​

Cancer care planning is fundamentally a multidisciplinary endeavor. Tumor boards—comprising oncologists, radiologists, pathologists, surgeons, and other specialists—are central to this process. By collaboratively reviewing patient data from various domains, these boards generate personalized, consensus-driven treatment plans that have been shown to improve patient outcomes. Yet, their promise is undercut by their inaccessibility: less than 1% of cancer patients receive the benefit of such individualized planning. The culprit? The painstaking effort required to compile, analyze, and synthesize data for each case. According to the American Society of Clinical Oncology, clinicians can spend up to 2.5 hours per patient preparing for tumor board meetings, a rate unsustainable in high-volume health systems.
The data explosion in medicine—spanning imaging, genomics, pathology, electronic health records (EHRs), and the ever-changing landscape of clinical trials—renders manual information synthesis all but impossible at scale. Fragmentation persists, leading specialists to miss potentially life-altering insights hidden within disparate databases and free-text notes.

Microsoft’s Solution: The Healthcare Agent Orchestrator​

Enter the Healthcare Agent Orchestrator, a software platform that coordinates a suite of specialized AI agents, each capable of handling a distinct component of cancer care planning. Whether parsing genomics data, analyzing radiology images, scanning pathology slides, or cross-referencing clinical trial eligibility, these AI-powered agents work collaboratively—much like a virtual, tireless assistant team.
The orchestrator’s distinctive strength lies in its integration with Microsoft 365 products—Teams, Word, PowerPoint, and Copilot. These tools form the backbone of administrative operations for countless healthcare organizations globally, removing the adoption barriers associated with unfamiliar software. Instead of working outside established digital workflows, clinicians interact with powerful AI within familiar interfaces, minimizing friction and maximizing productivity.
Stanford Medicine, an early adopter, provides a compelling case study. Dr. Mike Pfeffer, CIO at Stanford Health Care, highlights how their clinicians are already leveraging AI-generated summaries in tumor board meetings, using automation to reduce workflow fragmentation and uncover insights previously buried in the data haystack. With 4,000 tumor board patients annually, any reduction in labor—from hours to minutes per case—can have exponential effects on efficiency and patient access.

How the Orchestrator Works​

At its core, the orchestrator acts as a hub that manages many AI agents—each essentially a modular app tailored to a medical data type or workflow. Here are the orchestrator’s key technical features:

• Agent Coordination: The orchestrator assigns subtasks across specialized AI agents (imaging, pathology, EHRs, trial matching, etc.) and consolidates their outputs into digestible reports.
• Plug-and-Play Framework: Any approved agent, including third-party modules with an API, tool wrapper, or Microsoft Cloud Protocol (MCP) endpoint, can be integrated into Teams chats or collaborative threads.
• Multimodal Data Handling: The orchestrator can process vast and diverse data types: structured rows, free-text records, images, PDFs, test results, and genomics data.
• Conversational AI: Integrating with Teams, clinicians collaborate with the orchestrator via natural language, asking questions and receiving synthesized, context-aware responses.
• Open Agent Ecosystem: Microsoft is partnering with other companies—Paige.ai’s “Alba” digital pathology agent, for example—to enrich the system’s capabilities.

This open-ended design enables clinicians and health IT teams to customize workflows and connect the orchestrator to an ever-growing catalog of purpose-built AI agents, available via the Azure AI Foundry Agent Catalog.

Early Implementations and Real-World Impact​

Beyond Stanford Medicine, institutions such as Johns Hopkins, UW Health, Mass General Brigham, and Providence Genomics are piloting the Healthcare Agent Orchestrator, fine-tuning its capabilities and surfacing real-world challenges in live clinical environments. Their experiences underline both the promise and the complexity of integrating agentic AI into the medical workflow.

Tumor Board Transformation​

Perhaps nowhere is the impact clearer than in tumor board preparation. Dr. Joshua Warner, radiologist at UW Health, is exploring how sessions that once consumed two hours per patient might be accelerated to under ten minutes. The orchestrator’s ability to quickly fetch, normalize, and summarize multimodal data—while also surfacing relevant clinical trial matches and guideline references—extends far beyond what a single human reviewer might achieve within a tight timeframe.

Genomics and Evidence Parsing​

Providence Genomics is leveraging the orchestrator’s strength in analyzing both structured (like lab values) and unstructured (like case notes) data. The system’s prowess in genomics matching and real-world evidence parsing means experimental therapies and off-label drug options are surfaced with far greater speed and context, opening new doors for complex cancer cases.

Digital Pathology Collaboration​

Microsoft’s ecosystem approach yields immediate dividends with partners like Paige.ai. Through the Alba pathology agent, clinicians can upload and analyze whole-slide images—often gigabytes in size—receiving real-time, conversational insights. Instead of laboriously scrolling through digital slides, pathologists get summarized findings and can ask for clarifications in natural language, all within their existing Teams threads.

Critical Analysis: Strengths and Challenges Ahead​

The orchestrator’s grand ambition is evident, but its path to universal cancer care personalization is dotted with both impressive innovations and formidable obstacles.

Notable Strengths​

Seamless Integration and Adoption​

Microsoft’s choice to embed the orchestrator within the Microsoft 365 suite is a strategic masterstroke. The barrier to entry for time-strapped clinicians is dramatically reduced, sidestepping the friction historically associated with health IT rollouts. Instead of learning a complex new environment, specialists interact with AI-driven insights within the software platforms they’ve relied on for years. This familiar interface design is likely to drive rapid adoption and workflow stickiness.

Open, Modular Architecture​

The orchestrator’s API-friendly architecture allows healthcare institutions to combine first-party and third-party AI agents. This ensures extensibility—today’s imaging agent may be joined by tomorrow’s precision oncology tool, tailored for rare disease subtypes. The possibility for hospital IT teams (or entrepreneurial developers) to build and plug in their own agents means the orchestrator could become an innovation hub for AI in medicine.

Multimodal, Real-Time Data Synthesis​

By harmonizing diverse data types—from genomic variants to clinical narratives, radiology scans, and complex external databases—the orchestrator approaches the gold standard for comprehensive case review. Rapid, AI-driven summarization and evidence linking has the potential to democratize high-caliber, multi-expert opinions to a much wider patient base.

Partner Ecosystem and Early Institutional Validation​

Collaboration with leading institutions like Stanford and Johns Hopkins confers credibility and enables rapid iteration. These organizations possess the scale, data richness, and clinical acumen to pressure-test the orchestrator, surfacing bugs and refining workflows far more effectively than tightly controlled demos could.

Potential Risks and Limitations​

Research-Stage Deployment​

It’s important to note that the orchestrator’s current implementation, while feature-rich, is squarely focused on research and workflow discovery rather than immediate clinical deployment. Microsoft’s own announcement stresses that the system is presently a vehicle for “studying how AI agents might assist tumor boards, with broader applications to be explored in the future.” Thus, while early pilots are promising, widespread clinical adoption will hinge on peer-reviewed efficacy studies, regulatory scrutiny, and robust real-world performance validation.

Data Privacy, Security, and Governance​

Handling sensitive patient data—especially across multiple integrated AI agents and cloud endpoints—demands rigorous governance. Even though Microsoft is a trusted custodian for healthcare data under HIPAA, GDPR, and other frameworks, the complexity of agent orchestration multiplies the risk surface. API endpoints, especially from third parties, must be carefully vetted for security and compliance. Healthcare organizations will need tight controls and ongoing audit mechanisms to ensure both patient privacy and regulatory alignment.

Algorithmic Bias and Explainability​

AI agents, trained on varied datasets, may inadvertently inherit or amplify preexisting biases—potentially skewing treatment recommendations. Transparency and explainability are essential, especially in high-stakes contexts like cancer care. Clinicians must be able to interrogate how recommendations were generated, understand data provenance, and override AI-driven suggestions when warranted by clinical expertise or patient preference.

Over-reliance on Automation​

While automation can reduce labor, it raises the specter of deskilling—where clinicians might lose critical judgment or forensic skills if they come to trust AI unconditionally. Microsoft’s approach, which positions AI as an assistant rather than a decision-maker, is prudent, but health systems must invest in training clinicians to maintain a healthy skepticism about algorithmic advice.

Integration Complexity​

Despite the promise of “plug-and-play” agent integration, harmonizing data standards, ontologies, and workflows across disparate EHRs and hospital infrastructure remains a formidable challenge. Real-world health systems are notorious for idiosyncratic legacy systems, siloed data, and patchwork compliance approaches. Achieving seamless, error-free orchestration at scale will require ongoing investment, standardized APIs, and—crucially—broad industry cooperation.

The Road Ahead: Democratizing Personalized Cancer Care​

If Microsoft’s orchestrator can surmount these hurdles, its vision is nothing short of transformative. By decreasing case prep times from hours to minutes, hospitals could support more frequent, comprehensive tumor board reviews—not merely for rare or complex cases, but as a standard of care for every cancer patient. The potential downstream effects include:

• Wider Access to Personalized Plans: No longer a perk of elite academic centers, AI-accelerated tumor boards could be deployed in regional hospitals, community clinics, and even in resource-scarce settings globally.
• Accelerated Clinical Trial Enrollment: By programmatically matching patients to studies based on nuanced, up-to-date profile analysis, both patients and researchers can benefit from expanded opportunity and more efficient scientific discovery.
• Smarter, Data-Driven Guideline Adherence: AI agents can continuously surface evidence-based best practices, catching guideline deviations or surfacing new therapies as soon as they become available.
• Reduced Administrative Burden: Freed from hours spent digging through files and databases, clinicians can redirect their focus to patient interaction, nuanced decision-making, and shared care planning.
• Rapid Adoption of New Diagnostic and Therapeutic Tools: Third-party agents—built anywhere, adopted everywhere—create a powerful feedback loop for AI and clinical innovation.

Conclusion: A Blueprint for AI-Augmented Medicine​

Microsoft’s Healthcare Agent Orchestrator embodies the next logical step in the convergence of cloud computing, artificial intelligence, and clinical medicine. By weaving AI-powered agentic workflows directly into the fabric of clinical collaboration tools, it offers a tantalizing model for how “machine intelligence” can augment—not replace—the nuanced judgments of human clinicians.
Still, the orchestrator’s impact will depend on its capacity to prove itself in the demanding, high-stakes environment of real-world cancer care. Early indications are promising: pilot institutions are already reshaping workflows and surfacing insights with a scale and speed that manual review cannot hope to match.
But caution remains warranted. Full clinical deployment will require methodical validation, robust guardrails, and a commitment to transparency, security, and equity. AI must remain a partner, not a substitute, for the domain expertise and compassion at the heart of medicine.
If Microsoft—and its ecosystem of developers, health systems, and AI startups—can deliver on the orchestrator’s foundational promise, personalized cancer care could become not the exception but the new standard, bringing hope and precision to millions more patients worldwide.

---

Source: Maginative Microsoft’s Healthcare Agent Orchestrator Hopes to Enable Personalized Cancer Care for Everyone

 

[ChatGPT]

Microsoft’s bold unveiling of its AI agent orchestrator for cancer care marks a new chapter not only for medical technology but also for how artificial intelligence might soon reshape the fabric of enterprise healthcare environments. This innovation – now publicly available in the Azure AI Foundry Agent Catalog – promises to retool the way providers coordinate cancer treatment, blending modular, customizable AI agents with advanced interoperability in clinical settings. Proponents tout its potential to streamline the complexities of oncology workflows, while skeptics raise valid questions about the risks and readiness of such transformative technologies. This in-depth exploration dissects Microsoft’s latest offering, scrutinizing its design, claims, real-world implementation, broader context, and the high-stakes balancing act between innovation and patient safety.

The Rise of AI Agent Orchestration in Oncology​

Few healthcare challenges demand as much data-driven coordination as cancer care. Multidisciplinary, multimodal, and rapidly evolving, oncology requires seamless collaboration among clinicians, researchers, administrators, and now—according to Microsoft—sophisticated AI agents operating in dynamic tandem.
The agent orchestrator at the heart of this shift functions as a digital maestro, enabling multiple AI agents to “work together to complete a task.” In contrast to siloed AI tools that target narrow problems, the orchestrator supports a broad spectrum of healthcare tasks by combining general reasoners (capable of versatile, cross-domain logic) with hyper-specialized, multimodal agents (designed for niche or complex medical data types).

What Sets Microsoft’s Solution Apart?​

Unlike single-purpose chatbots or analytics tools, the Azure healthcare agent orchestrator is engineered for:

• Multi-agent orchestration: Multiple AIs can collaborate or specialize, breaking down large tasks into subtasks—such as extracting, summarizing, and contextualizing data from medical records.
• Open-source customization: Developers and researchers are empowered to compose, adjust, and extend each agent, leveraging their own models, tools, and domain knowledge.
• Native integration: The orchestrator is natively compatible with Microsoft’s productivity suite (Teams, Word, PowerPoint, M365 Copilot), bridging clinical workflows and existing enterprise tools.
• Interoperable data services: The solution taps into Microsoft Fabric and the Fast Healthcare Interoperability Resources (FHIR) service, promoting seamless, standards-based connectivity across EHRs and healthcare IT infrastructure.
• Explainability and traceability: Outputs from the orchestrator are grounded in source EHR data, promising an unusual degree of transparency—a critical safeguard in high-stakes, regulated environments like healthcare.

Harnessing Multimodal Data for Clinical Impact​

At its technical core, the orchestrator coordinates AI agents capable of analyzing myriad healthcare data forms, from digital imaging and DICOM files (vital for radiology and pathology), to genomics, structured EMR entries, and unstructured clinical notes. This breadth allows the system to:

• Build longitudinal patient timelines by fusing disparate data streams.
• Assess cancer staging across imaging, genomics, and text-based diagnostics.
• Evaluate trial eligibility rapidly by scanning and cross-referencing relevant patient attributes.
• Surface guideline-concordant care insights at the point of need—all in a manner designed to augment, not replace, the clinical decision-maker.

Microsoft describes the orchestration engine as offering “actionable insights grounded in multimodal clinical data,” echoing a growing industry emphasis on holistic, patient-centered analytics.

Real-World Pilots: Stanford, Johns Hopkins, and Beyond​

Microsoft’s early development partnerships have lent credibility to its ambitions. Institutions such as Stanford University, Johns Hopkins, Mass General Brigham, UW School of Medicine, and Providence Genomics are reportedly experimenting with the orchestrator to support complex, clinically relevant workloads.
Mike Pfeffer, Chief Information Officer at Stanford Health Care, provided a notable testimonial: Clinicians at Stanford already employ “foundation model generated summaries in tumor board meetings today (via a PHI-safe instance of GPT on Azure).” Pfeffer pointed to the orchestrator’s potential to “reduce fragmentation and enable new insights from data elements that were challenging to search”—including trial eligibility criteria, treatment guidelines, and real-world evidence. This, he argued, could represent a watershed in moving “the first generative AI agent solution into a production setting for real-world care for our cancer patients.”
Such real-world pilots, especially with prominent academic centers, provide valuable signals about the technology’s maturity and practical impact. However, independent peer-reviewed studies on clinical outcomes, user satisfaction, and safety considerations are still emerging and should be followed closely as deployment scales.

Demystifying the Agent Orchestrator: How Does It Work?​

Modular and Highly Customizable​

At the implementation level, each “agent” is a modular software function—powered by Azure’s stable of advanced AI models—that can be individually tuned or combined into complex workflows. These agents can be generalist (e.g., answering broad clinical questions or summarizing documents) or specialist (analyzing genomics, medical imaging, or extracting specific EHR features).
Developers gain access to:

• APIs for orchestration and messaging between agents.
• Customizable logic for chaining, branching, or hierarchically organizing agent actions.
• Tools to connect and transform enterprise data from Microsoft Fabric and FHIR endpoints.
• Support for packaging and distributing agents for direct use in familiar tools such as Teams or Word, streamlining the clinical workflow.

Grounding and Explainability​

A recurrent theme in Microsoft’s messaging is the orchestrator’s explainability. Outputs produced by the system are explicitly “grounded” in underlying source data—typically EHR records, imaging results, or verified clinical guidelines. This traceability is essential, as black-box AI risks are particularly acute in regulated, safety-critical sectors like healthcare.

Security and Privacy​

Research partners, including Stanford Medicine, emphasize the use of PHI (Protected Health Information) safe instances and explicit data governance protocols for all pilot implementations. Microsoft asserts compliance with HIPAA and other key healthcare data standards, although—like all cloud-driven clinical tools—the robustness of these security frameworks should be routinely audited and validated by independent experts.

Key Use Cases and Workflow Scenarios​

Tumor Board Coordination​

Multidisciplinary tumor boards—where oncologists, radiologists, pathologists, surgeons, and nursing staff confer on complex cancer cases—are a natural proving ground for orchestration. The orchestrator automates the synthesis and summarization of multimodal information (imaging, genomics, prior notes), surfacing relevant highlights for discussion and documentation. By integrating directly into Teams chats, the system can support real-time expert discussions among humans and AI agents alike.

Rapid Literature and Trial Eligibility Scanning​

Locating and cross-referencing clinical trial criteria, guideline updates, or the latest real-world evidence is a notorious time sink for cancer care teams. The orchestrator’s ability to parse, filter, and contextualize this information—potentially in minutes rather than hours—could accelerate clinical decision-making and improve patient access to cutting-edge therapies.

Personalized Clinical Summaries​

By automating the construction of longitudinal patient timelines across vast, variable, and incomplete EMRs, agents can generate summaries tailored to clinician queries, surfacing relevant history, labs, treatments, and outcomes. This personalized data storytelling becomes an essential productivity and safety tool in high-volume clinics.

Strengths: Unlocking the Power of Agentic AI in Healthcare​

1. Multidisciplinary Synergy​

The orchestrator’s multi-agent architecture is well-matched to the inherently multidisciplinary nature of cancer care. By intelligently routing tasks between specialist and generalist agents, the system avoids some brittleness seen in single-point AI tools.

2. Deep Integration into Familiar Tools​

Direct plug-ins to Microsoft Teams, Word, PowerPoint, and the M365 Copilot family lower the barriers to adoption. Clinicians need not abandon existing workflows, making rollout, customization, and scaling easier for IT teams.

3. Championing Explainability​

By grounding AI-generated outputs in source EHR or clinical data and surfacing the contextual logic behind recommendations, Microsoft addresses a critical barrier to clinical trust and regulatory approval.

4. Open-Source Flexibility​

Developers and researchers retain agency to extend, fork, or compose agents for unique institutional workflows. This contrasts with many closed commercial healthcare AI solutions, which restrict modification and transparency.

5. FHIR and Data Standards Support​

Native support for the FHIR specification and integration with Microsoft Fabric enable genuine interoperability—often the Achilles’ heel of healthcare data innovation. This design bodes well for broader compatibility.

Risks and Weaknesses: What Remains to Be Proven​

Despite impressive ambitions and promising pilots, the orchestrator is not without unanswered questions and potential pitfalls.

1. Regulatory and Clinical Safety Hurdles​

The path from pilot deployment to routine use in clinical decision-making is fraught with regulatory scrutiny, especially for AI that could influence diagnosis or treatment. The FDA and equivalent authorities globally retain strict requirements for validation, transparency, and post-market surveillance.

2. Data Fragmentation and Source Quality​

While the orchestrator can connect diverse data streams, the integrity, consistency, and completeness of source data—especially in fragmented US healthcare settings—remains a stubborn challenge. No AI system can reliably reason over incomplete or poor-quality records.

3. Data Privacy, Security, and Compliance​

Even with PHI-safe provisions and HIPAA alignment, the risks of third-party cloud-based data handling are non-trivial. Institution-specific configurations, robust cryptographic protocols, and frequent audits remain paramount.

4. Human Factors and Automation Bias​

Embedding agentic AI into routine workflows carries the risk of “automation bias”—where clinicians over-trust AI-generated summaries or recommendations. Microsoft’s explainability features mitigate this somewhat, but ongoing clinician education and human-in-the-loop safeguards must be rigorously maintained.

5. Vendor Lock-in and Ecosystem Dependence​

Deep integration with the Azure ecosystem unlocks value but may inadvertently create pressure for health systems to centralize around Microsoft’s stack, potentially limiting flexibility or creating long-term switching costs.

Broader Context: Microsoft’s Growing AI Healthcare Portfolio​

The agent orchestrator is part of a concerted push by Microsoft to consolidate its leadership in enterprise healthcare AI. Other major initiatives include:

• Welcoming xAI’s Grok 3 model to Azure AI Foundry: Grok 3 is designed for rapid scientific discovery, trained on a vast GPU cluster and aiming to advance LLM frontiers. The addition provides customers with access not only to Microsoft’s own models but also to powerful external LLMs, multiplying choice for developers.
• Blue Shield of California’s Experience Cube: Joint development of a cloud-based data hub integrating member, provider, and payer data for personalized clinical insights.
• Fujitsu’s Cloud Healthcare Platform: Built atop Azure, offering secure, HL7 FHIR-compliant aggregation and pseudonymization of medical data—a testament to Azure’s positioning as a healthcare IT backbone.

Critical Takeaways and Future Directions​

An Industry Watershed—or Just the First Step?​

If Microsoft’s orchestrator delivers on its promises, the implications are profound: Automated, multidisciplinary synthesis of vast, disjointed clinical information may soon become routine in cancer care and beyond. Early development work at Stanford and peer institutions hints at real, pragmatic wins—accelerated tumor board coordination, reduction in administrative friction, and untapped clinical insights.
Yet structural obstacles remain—regulatory, cultural, ethical, and technical. The orchestrator’s modular, explainable architecture provides a solid foundation, but true mainstreaming will require continuous validation, ironclad security, and a commitment to augmenting rather than replacing human expertise.
Real-world, peer-reviewed evidence on clinical outcomes, safety incidents, and unintended consequences will be crucial. Health systems and clinicians should approach adoption with cautious optimism, extensive pilot testing, and robust governance structures.

SEO-focused Guidance for Healthcare IT Leaders​

• Evaluate agent orchestrator solutions for interoperability with existing EHRs and compliance with FHIR standards.
• Engage clinical champions in pilot implementations to flag human factors and usability issues early.
• Rigorously test for security compliance, especially around PHI handling and cloud data residency.
• Measure productivity, satisfaction, and clinical impact compared to baseline workflows.
• Remain vigilant for automation bias or over-reliance on AI-generated summaries.
• Insist on explainability and traceability for all AI-generated outputs in patient care contexts.

Frequently Asked Questions​

What is Microsoft’s AI agent orchestrator for healthcare?
It is a modular system that coordinates multiple AI agents to automate and augment complex clinical workflows, focusing first on cancer care. It leverages cloud-based AI models, is deeply integrated into the Microsoft productivity suite, and is designed for explainability and interoperability.
How does it improve oncology care?
By aggregating and contextualizing multimodal data—imaging, genomics, notes, trial guidelines—the orchestrator helps clinicians rapidly assemble patient histories, evaluate trial eligibility, and synthesize consensus recommendations during tumor boards.
Who is piloting these tools today?
Early development partners include Stanford Medicine, Johns Hopkins, Mass General Brigham, Providence Genomics, and the University of Wisconsin. Each is exploring unique use cases to determine the orchestrator’s real-world value.
Can these agents replace clinicians?
No. Microsoft frames agentic AI as augmentative—aimed at saving specialists time on data wrangling, surfacing insights for complex decisions, and reducing administrative clutter. Final clinical judgments remain squarely with the human expert.
What are the key risks?
Potential risks include over-reliance (automation bias), data privacy vulnerabilities, fragmented/incomplete source data, and long-term vendor lock-in. Regulatory approval and post-deployment governance are critical.

The Road Ahead: Balancing Promise and Prudence​

The debut of Microsoft’s AI agent orchestrator in cancer care is a watershed in the evolution of clinical AI—from isolated decision support tools to orchestrated, explainable, and truly collaborative digital teams. With modular customization, deep EHR integration, and a robust security posture, the orchestrator provides a glimpse into the future of multidisciplinary medicine augmented by artificial intelligence.
Yet its path to ubiquity will be shaped by careful, transparent validation, a relentless focus on privacy and safety, and above all—a partnership between innovative technology and the irreplaceable intuition of the clinical expert. The success of the orchestrator, like all transformative healthcare tools, will ultimately be measured not in technical milestones, but in the lives improved and saved on the front lines of cancer care.

---

Source: MobiHealthNews Microsoft unveils AI agent orchestrator for cancer care coordination