Partnering with Foundation Medicine and UVA: Deploying Agentic AI in Real-World Life Sciences Workflows, Powered by Manifold Agent OS

Accelerating Life-changing Medicines to Patients

At Manifold, our mission is to accelerate life-changing medicines to patients, from early discovery to the clinic.

We believe that agentic AI can overcome the hidden rationing of expert time that slows down every step along the way. It holds the promise of dramatically accelerating work across all types of multimodal biomedical data, from preparation through inquiry, insight, and ultimately impact.

However, while agentic AI is already proving to be incredibly useful in single-user desktop tasks, there are unique challenges in AI agents extended to complex workflows and deployed to production securely across an organization. In life sciences, those challenges include multimodal biomedical data of genomic scale, fragmentation across silos, cross-team analytic workflows across multiple datasets, and strict access control that has to work across organizational boundaries. Most agentic AI in life sciences still stalls out between a desktop prototype and a deployed workflow agent with behavior controls and verifiability that an organization can trust.

Realizing the promise of agentic AI in life sciences will require closing this gap.

Deploying Agentic AI in Real-World Research Workflows

Today, we are delighted to announce promising outcomes from deployment of agentic AI in real-world research workflows with Foundation Medicine and UVA. The initial high-value workflows with these customers are poised to unlock significant efficiency gains. Workflows that previously consumed weeks of expert time and coordination across multiple teams are now completed in minutes, with full controls and verifiability built in. Both run on Manifold Agent OS, the operating layer that lets agents, built with us or brought by you, work securely on an organization's life sciences data. The stories below show what that looks like in practice.

Foundation Medicine works with biopharma partners to advance the development of precision oncology therapies by leveraging their expertise and a broad portfolio of high-quality genomic profiling assays and real-world data solutions. To accelerate their partners’ research and development, Foundation Medicine has collaborated with Manifold to provide innovative access and analysis tools that enable fast, self-service provisioning and even faster insights generation from their multimodal, real-world database of de-identified genomics data, FoundationCore®. Foundation Medicine is now able to deliver that experience through FoundationInsights® powered by Manifold, their cloud analytics platform exclusively available to biopharma for the support of biomarker research and drug development.

The FoundationInsights experience runs on two agents. A cohort agent, currently being piloted, enables researchers to evaluate prevalence, determine the number of relevant records available, and construct subsets of target populations defined by precise genomic and clinical criteria using plain language. An analysis agent then provides rapid evaluation and insights generation through guided analyses on the defined cohort. Instead of spending time researching the nuances of the data model and developing custom code, analysts are now able to focus their efforts on efficiently exploring promising analytical directions and, through a highly transparent workflow, provide a layer of expert validation on top of the agent’s output.

The whole workflow runs within Foundation Medicine’s FoundationInsights powered by Manifold, with the data securely hosted within Manifold’s trusted research environment. A researcher can explore what is available, request access, receive access under Foundation Medicine’s governance, and promptly start their analyses, all in one environment with controls and audit logging applied to every agent action. This is the partner experience Agent OS is built to deliver: a branded product on top and a governed data and agent foundation underneath.

“FoundationInsights was designed to put real-world genomic and clinical data directly in the hands of our customers. With Manifold, we’re now able to advance and operationalize that vision as agents that customers can run themselves across organizational boundaries, while maintaining our high standards for data governance and compliance. Our collaboration with Manifold is helping turn complex hypotheses into rapid, self-serve insights delivered in minutes.”
— Brian Clancy, Foundation Medicine

The UVA Comprehensive Cancer Center sees roughly 5,000 new diagnoses and collects about 8,000 specimens each year, with clinical, biospecimen, and multi-omics data spread across four disconnected systems. Assembling a single research cohort could take up to 17 weeks and as many as eight staff members. The friction was high enough that researchers avoided studies that required complex data retrieval.

Today, UVA researchers query data on more than 130,000 patients through the Cancer Research DataMart, built on Manifold. A cohort agent lets investigators build and refine patient populations in plain language, with no SQL, no chart abstraction, and no data-team queue. The DataMart's analysis agent queries across datasets and generates visualizations like Sankey treatment-flow diagrams and Kaplan-Meier survival curves that would otherwise require a data analyst and custom scripts.

The same pattern holds across very different research questions. A cohort of advanced-stage prostate cancer patients with tissue blocks available went from six weeks and five staff to five minutes and zero staff. A harder query, breast cancer patients who later developed secondary leukemia with biospecimens available, went from 17 weeks and eight staff to the same five minutes and zero staff.

The DataMart also shows where Agent OS is heading. Alongside UVA's own registry and biospecimen data, the platform brings in public datasets (such as TCGA and GTEx) that researchers have long wanted to analyze next to their proprietary data, connects to sources like Open Targets Platform and ClinicalTrials.gov using public MCPs, and lets investigators ingest their own datasets directly. Agents reason across all of it under dataset-level access controls, audit trails, and a bring-your-own-cloud model, so the same governed foundation scales continuously as new patients, specimens, and studies arrive.

"I am very excited about this collaboration because it has the potential to change how clinical research is done at UVA and dramatically accelerate research progress."
‍— Dr. Elizabeth Mulcahy, UVA Comprehensive Cancer Center

Manifold Agent OS: Deployment Accelerator for Agentic AI in Life Sciences Enterprise Workflows

To meet the needs of Foundation Medicine, UVA, and future life sciences clients and partners, we had to reimagine the Manifold platform for how we believe humans and agents will work together in life sciences.

The new Agent OS is a deployment accelerator for agentic AI in life sciences enterprise workflows. It shortens the path from idea to production across the agent development lifecycle, helping the experts who want to create agents to test, deploy, and share them across their organization, their partners, and their customers. You bring the idea. Agent OS builds and deploys your agents on the same proven security and compliance foundations, now extended to every action. This release introduces several advances:

• Built and validated across multimodal real-world data: Agent OS has been proven on real life sciences workflows, cutting previous processes from weeks to minutes and spanning multiple data modalities from molecular to clinical.
• Enterprise security and compliance: Manifold's architecture was built for sensitive healthcare data from day one: your AI runs inside your compliance requirements with your data staying in your own AWS environment. Manifold is ISO 27001 and TX-RAMP certified, and maintains a SOC 2 Type II report covering the security and confidentiality pillars, the independent, audited credentials that international and enterprise buyers treat as table stakes. Manifold is also aligned with the NIST 800-171 and NIST 800-53 frameworks, and supports HIPAA, GDPR, and IRB review, including the BAAs, PHI handling, role-based access, and audit-trail controls those require. And because compliance is never one-and-done, we carry the work of keeping that foundation current as the technology and our platform evolve, so your teams inherit a compliant footing instead of building and re-validating it from scratch.
• Governance built for agent actions, and for managing multiple agents over time: Agent OS handles applying your access and privacy rules to every agent action across data sources and tools on a user's behalf, with audit logging of what the agent touched and produced.
• Semantics built for multimodal biomedical data agents: Agent OS grounds every agent in the meaning of biomedical data modalities like molecular, imaging, clinical, and claims so a plain-language question returns a sound answer rather than a naive query result.
• Data platform built to provide agents context and tool use: Agent OS does for agents what the modern data platform did for human data scientists in life sciences. It enables agents to access multiple datasets which provides the data foundation for a scalable solution. Agent OS also comes pre-configured for tool use with the many standard domain-specific tools (e.g. Python, R, Nextflow, WDL, BI/dashboarding, and more) so agents can get right to work without any setup. Agent OS also manages memory and context and solves for the reality of messy real-world data warehouses that model benchmarks miss.
• Studio built for designing complex agents: Agent OS gives you one place to design agents and manage all available domain-specific skills, tools, and data connectors your agents draw on. Scoping agents this way is intuitive and keeps their work reliable as the number of skills and agents grows.
• From expert’s prototype to production agent: Agent OS enables full path of deploying a production agent the whole organization can trust, with the testing, review, and monitoring required to keep it reliable and secure over time.
• Shared workspaces for partners that cross organizational boundaries: Agent OS enables partners to provision access to agents across organizational boundaries, enabling a faster path to partner inquiry and insight generation that’s not gated by servicing capacity.
• Built to fit how you already work with AI: Works through your organization’s preferred frontier model like Claude, ChatGPT, Cortex, and more. In addition, we enable power users to access the raw intelligence of any frontier model inside Compute Environments with the full security and compliance guarantees of the platform.

Evaluate the New Agent OS on the Workflows that Matter Most

The best way to understand what Agent OS can do is to evaluate it with your own data and the workflows you know best. Great candidates are promising desktop prototypes that haven't yet made it into production. We're opening a limited number of evaluation engagements per month for life sciences teams.

Our Applied AI team works with you every step of the way, from business and technical discovery to production deployment. The AI solution consultants and forward-deployed engineers bring deep life sciences fluency and deep technical expertise. They work directly alongside your team to design, launch, and iterate agents on the complex workflows that matter most, carrying each one through the full agent development lifecycle and acting as trusted advisors.

Sign up here to kick off an evaluation of the new Agent OS for your high-value workflows. We will be expanding slots as capacity becomes available.