From Discovery to Remission: How Agentic AI Is Reshaping the Oncology Continuum

While researchers and care teams have been tackling the challenge of breast and other cancers for decades, today something is genuinely different in how organizations fight cancer. I've spent time with oncology researchers, health systems and life sciences companies, and overwhelmingly, the conversation has shifted from “We have a lot of data” and “How do we organize our data?” to “What can our systems do with it?”

Oncology generates more heterogeneous, high-stakes data than almost any other domain. Genomic sequences, pathology images, clinical notes, claims records and wearable signals have historically worked against clinicians rather than for them, fragmented across systems and slow to assemble when decisions matter most.

Agentic AI — systems that reason, plan and act across multistep workflows — is beginning to close that gap across the oncology continuum, moving healthcare and life sciences from siloed data toward AI experimentation and, further, measurable impact. This post follows the journey from the lab to survivorship, highlighting organizations moving this work forward.

Five stages, five long-standing bottlenecks

The oncology continuum has five connected stages, each with a long-standing data problem. Discovery and research teams rely on molecular and genomic data sets that remain siloed. Screening and early detection bring together imaging, biomarkers, electronic health record (EHR) data and family history only with difficulty. Diagnosis and treatment planning often require tumor boards to assemble patient profiles manually and incompletely. Treatment and care coordination depend on multimodal real-world data fragmented across sites and systems. Remission and survivorship generate evidence that can lag years behind clinical need.

These problems compound. The same data failure that slows discovery can delay the right patient from reaching the right trial. All five bottlenecks trace back to the same root cause: There is no unified data foundation that can reason across the continuum.

What ‘agentic’ actually means here and a foundation-first approach

Traditional AI answers a question. Agentic AI plans and executes across multiple steps without a person needing to trigger each one.

However, AI is only as effective as the data foundation that governs it. Cancer data isn't a spreadsheet. Genomic sequences, pathology images, unstructured notes, claims records and wearable signals are complex and multimodal in nature. They carry valuable insights needed to move care forward but only when organizations can ingest, unify and reason over them together. A genomic variant needs clinical context. An imaging finding gains meaning alongside molecular markers and treatment history.

Consider an agent that receives a sequencing result, compares it with similar patient profiles, identifies open trials, models feasibility and prepares a tumor board summary — all from a single data event. The convergence of multimodal data infrastructure, foundation models and governed AI platforms is making this operationally viable.

But agentic capabilities acting on regulated data and workflows require a different level of scrutiny. Governance and security must be built into your agentic strategy from the ground up — not bolted on afterward — to safeguard protected health information (PHI) and support compliance with industry standards such as HIPAA. The impact of an incorrect answer in the clinical context carries implications far beyond an incorrect dashboard and can affect patient outcomes. This means auditability, access controls, agent guardrails and clear rules for when to keep humans in the loop are mission critical and should be considered at the front end of the agentic journey.

Where agentic AI is creating real impact

Discovery and research

The first bottleneck — siloed molecular and genomic data — is where agentic AI is showing early promise. Oncology drug development has a persistent attrition rate above 95%, with an estimated $50 billion to $60 billion spent annually on failed oncology trials, according to a 2023 study in JAMA Network Open.

Agentic systems can traverse genomic, structural and literature data in one reasoning chain. They can surface viable candidates earlier, flag likely failure modes before the lab and continuously incorporate new findings. The practical result is a faster path from hypothesis to viable compound, giving research teams more time to spend on the strongest opportunities.

Screening and early detection

The second bottleneck — multimodal signals that never converge — creates a deeply human problem. After a biopsy, patients can spend two to six weeks waiting for pathology, with longer delays when molecular biomarker testing or genomic sequencing is required. During that time, they are left in uncertainty, unable to plan or make sense of what comes next, not knowing whether they are positive or negative for the disease.

AI is compressing parts of these workflows. AI-powered imaging triage can reduce turnaround from days to minutes by prioritizing time-critical findings; digital pathology can transform slide analysis — accelerating previously manual workflows with computational science. A large real-world U.S. screening study found that an AI-supported workflow identified 21.6% more breast cancers than standard mammography review.

The agentic opportunity lies in convergence: an imaging finding, biomarker result, EHR risk flag and family history pattern considered together, replacing the sequential manual assembly that creates the wait. The clinical impact is faster diagnosis. The human impact is fewer weeks in limbo.

Diagnosis and treatment planning

Tumor boards remain among oncology's most important and most manual workflows. Clinicians assemble genomic profiles, pathology, imaging and treatment histories from disparate systems, often under intense time pressure.

An agentic system triggered by a new diagnostic result can gather those sources, structure a patient brief, cross-reference trial eligibility and surface the material before the board convenes. It prepares the ground for clinical judgment and gives specialists a more complete starting point for discussion rather than starting from scratch.

The human impact here is real: less time waiting for answers and faster time to treatment.

Treatment and care coordination

The fourth bottleneck — fragmented multimodal real-world data — contributes to one of oncology's persistent failures: Patients who could benefit from clinical trials often never enroll.

Agentic workflows can identify eligible cohorts in near real time across complex trial criteria. They can also model trial feasibility, enrollment timelines, site capacity and protocol complexity before a trial launches, when sponsors still have time to act. That shifts trial operations from reactive enrollment to proactive matching.

Remission and survivorship

The final bottleneck — evidence generation that lags behind clinical need — affects regulators, payers, health systems and patients. Agentic workflows can continuously structure real-world evidence from treatment data, adverse-event feeds and long-term survival signals — shifting evidence generation from a retroactive, one-off exercise to a proactive, continuous process productized for scale. Those same workflows can monitor adherence signals such as prescription refill patterns, patient-reported symptoms and missed follow-up schedules, flagging patients at risk of discontinuing long-term therapies such as endocrine treatment, where early intervention can meaningfully affect recurrence rates.

That work can compress the path from treatment to insight from years toward months, depending on data availability, study design and governance. It also closes the loop: Real-world evidence and adherence patterns generated during survivorship can inform the next discovery program, trial design and treatment strategy — while helping reduce recurrence risk for patients still in care.

Organizations making AI real for healthcare

AstraZeneca

Using Snowflake Cortex AI, AstraZeneca reported that it had dramatically accelerated its data product creation, cutting development time from six months with 16 engineers to just four days with two. The efficiency gain led to more than 118 data products, unlocking thousands of hours in productivity and more than $10 million in savings. Critically, these advancements power AI-driven innovations such as AI-powered chest X-rays, which are intended to improve early lung disease detection and, in certain cases, have been associated with survival rate increases of up to 90%. With data processes optimized at scale, AstraZeneca is well positioned to drive even more innovation and deliver improved patient outcomes in the life sciences sector.

Zelis Healthcare

Zelis Healthcare shows the next step: moving from a unified foundation to production agents. Zelis migrated its entire data foundation to Snowflake and now ingests near real-time data from more than 40 sources through Snowflake Openflow. With that foundation in place, its team has deployed more than 20 production agents, building proof-of-concept agents in a single day and moving them into full production two days later — delivering near real-time answers and reducing demand on the core data team. Lakshman Pindikur, Senior Director of Data Engineering (Zelis Data Cloud) at Zelis Healthcare, describes the foundation behind that progress: "We can quickly enable our users through Snowflake CoWork, with Openflow as our go-to tool for bringing all data into Snowflake, and Horizon Catalog for enterprise-grade security and governance."

The broader pattern matters more than any single implementation. Research organizations are bringing together clinical, genomic and imaging data. Providers are connecting treatment histories with trial operations. Real-world evidence and health economics and outcomes research (HEOR) teams are combining claims, EHR, registry and patient-generated data to answer questions faster. Agentic AI gives these organizations a way to coordinate the work across systems while keeping experts in control of the decisions.

The work is already happening

The technology matters less in the abstract than what organizations are doing with it right now: turning agentic AI from an experiment into measurable growth and efficiency across the cancer journey. They do so by connecting fragmented evidence, supporting researchers, improving trial planning and giving clinicians a fuller view of each patient's journey.

To see how organizations are driving progress in this space, read about the ways the National Cancer Institute (NCI) modernized its enterprise data platform to gain greater flexibility, improve cost transparency and support future innovation.