CellType: The Agentic Future of Drug Discovery

In an era when artificial intelligence is rapidly reshaping industries, biotechnology stands at the edge of a profound transformation. CellType, a New York–based startup founded in 2025 and part of Y Combinator’s Winter 2026 batch, is positioning itself at the forefront of that shift. With a team of just two founders, the company is attempting something extraordinarily ambitious: reinventing how drugs are discovered and developed by building what it calls an “agentic drug company.”

Rather than functioning as a traditional biotech firm with sprawling laboratories and large research teams, CellType operates on a radically different premise. The company is designing AI agents capable of running the entire drug discovery pipeline — from hypothesis generation to experimental planning — on top of biological foundation models that simulate human biology. In essence, CellType aims to replace slow, manual scientific workflows with autonomous computational systems guided by human experts.

The startup’s central belief is simple yet disruptive: if AI can already write software, design products, and generate complex strategies, then drug discovery — a process rooted in pattern recognition, data analysis, and hypothesis testing — is ripe for automation as well. By building a company where AI performs most of the operational work and scientists steer the direction, CellType hopes to compress timelines that currently span decades into mere weeks.

Why Is Drug Discovery So Slow and Inefficient Today?

To understand CellType’s mission, it is necessary to examine the inefficiencies embedded in the traditional pharmaceutical model. Drug discovery has long been notorious for its high costs, long timelines, and staggering failure rates. On average, developing a single drug can take more than ten years and cost upwards of two billion dollars. Even after that investment, approximately 90 percent of drug candidates fail during clinical trials.

A major reason for this failure lies in the models used during preclinical research. Scientists typically rely on mice, cell cultures, or other laboratory proxies to predict how a drug will behave in humans. However, these systems often fail to replicate the complexity of human biology. A therapy that appears promising in animal studies may prove ineffective — or even dangerous — when tested in patients.

Beyond flawed models, the process itself is fragmented and slow. Large pharmaceutical organizations operate through layers of teams, departments, and external contractors. Researchers manually sift through scientific literature, analyze experimental data, formulate hypotheses, design experiments, and coordinate with laboratories. Each stage introduces delays and opportunities for miscommunication.

CellType argues that the entire framework optimizes for the wrong reality. Instead of testing drugs directly against accurate representations of human biology, companies iterate on imperfect proxies, spending years refining treatments that may never succeed in real patients.

How Does CellType’s Technology Simulate Human Biology?

At the heart of CellType’s approach is a breakthrough concept: modeling biology in the same way language models understand text. The company’s core technology, called Cell2Sentence, converts biological data — particularly single-cell gene expression profiles — into a structured format that foundation models can learn from.

The underlying insight is that biological systems possess structure similar to language. Gene expression patterns can be interpreted as sentences, cellular states as meaning, and drug responses as transformations. By translating complex biological interactions into a machine-readable “language,” AI systems can identify patterns across cells, tissues, and disease states at a scale impossible for human researchers.

Unlike earlier AI-driven biology efforts focused narrowly on proteins — such as protein-folding models — CellType aims to simulate entire biological systems. This includes interactions between cells, tissues, microenvironments, and therapeutic compounds. Instead of studying isolated components, the company attempts to model the whole organism as an integrated machine.

The result is what CellType describes as a “virtual human”: a computational representation of human biology that allows scientists to test drugs digitally before exposing real patients to risk. Researchers can ask questions such as how a treatment will affect specific tissues, whether it may cause toxicity, or which patients are most likely to benefit — all before launching physical trials.

What Role Do AI Agents Play in the Discovery Pipeline?

While biological foundation models provide the knowledge base, autonomous AI agents supply the operational engine. CellType is building systems that can execute end-to-end drug discovery workflows with minimal human intervention.

In traditional research environments, hundreds of scientists collaborate to review literature, analyze data, propose targets, design experiments, and manage laboratory relationships. CellType’s agents are designed to perform those tasks automatically, iterating continuously and learning from outcomes.

Human experts remain essential, but their role shifts from manual execution to strategic oversight. Scientists guide research directions, evaluate results, and ensure ethical and scientific rigor, while AI handles the heavy lifting of analysis and experimentation planning.

This “agentic” structure mirrors trends in software development, where automated systems increasingly generate code, test functionality, and deploy updates. CellType believes that applying similar automation to biology could dramatically accelerate innovation.

How Did CellType Validate Its Approach With a Cancer Discovery?

A concept this ambitious requires tangible evidence, and CellType has already reported a significant early result. Using its models and agents, the company screened more than 4,000 drugs and identified a previously unknown signal related to cancer treatment.

Specifically, the system predicted that a certain drug could transform “cold” tumors — which are invisible to the immune system — into “hot” tumors that immune cells can attack. This phenomenon is a major focus of modern oncology, as it can determine whether immunotherapies succeed.

Importantly, CellType did not stop at computational predictions. The team conducted laboratory validation using both traditional cell lines and complex tumor microenvironments designed to mirror human conditions. The findings held up, suggesting that the model captured biologically meaningful insights rather than statistical artifacts.

Such a discovery typically requires years of coordinated research across multiple teams. CellType claims its approach achieved comparable results in a fraction of the time, offering a glimpse of what accelerated drug development might look like.

Who Are the Founders Behind CellType?

CellType’s credibility rests heavily on the backgrounds of its two founders, who combine deep expertise in biology and machine learning.

Co-founder and CEO David van Dijk is a Yale professor with more than 11,000 academic citations and publications in leading journals such as Cell and Nature, as well as major AI conferences. His research has focused on applying computational methods to biological systems, positioning him as a pioneer in the field of biological foundation models.

Co-founder Ivan Vrkic brings complementary experience in large-scale machine learning and scientific computing. His work spans institutions such as Yale and EPFL, as well as engineering projects including software systems for controlling CERN’s Large Hadron Collider. He also contributed to large foundation model training efforts at a biotech startup.

Together, the founders developed CellType’s core technology during their academic research before transforming it into a commercial venture. Their decision to leave established institutions and build a startup reflects confidence in the technology’s transformative potential.

Why Are Major Pharmaceutical Companies Already Interested?

Despite its small size, CellType has attracted attention from some of the world’s largest pharmaceutical companies. According to the startup, collaborations with top-tier pharma firms have emerged entirely through inbound interest rather than aggressive sales efforts.

This interest likely stems from the industry’s urgent need for more efficient discovery methods. As development costs continue to rise and patent cliffs threaten revenue streams, pharmaceutical companies are under pressure to replenish pipelines quickly. A platform that promises faster, more accurate predictions of human outcomes could significantly reduce risk.

By partnering with established players rather than attempting to become a full-scale pharmaceutical manufacturer immediately, CellType can validate its technology while accessing data, resources, and expertise that would otherwise take years to acquire.

Could Agentic Drug Companies Transform the Future of Medicine?

CellType’s vision extends beyond improving existing workflows. The startup imagines a future in which drug discovery becomes primarily computational, with physical experiments serving as confirmation rather than exploration.

If successful, such an approach could dramatically expand the range of treatable diseases. Conditions currently deemed too complex or unprofitable might become viable targets once development costs fall and timelines shrink. Personalized medicine could also benefit, as virtual models enable testing therapies on specific patient populations before clinical deployment.

However, significant challenges remain. Biological systems are extraordinarily complex, and even the most advanced models cannot capture every variable. Regulatory frameworks, ethical considerations, and data limitations will also influence how quickly agentic drug companies can reshape the industry.

What Does CellType’s Emergence Say About the Convergence of AI and Biology?

CellType represents a broader trend: the convergence of artificial intelligence and life sciences into a new technological frontier. Just as AI transformed software engineering, finance, and logistics, it is now poised to redefine how humanity understands and manipulates biology.

By framing drug discovery as an information problem rather than purely an experimental one, startups like CellType challenge long-standing assumptions about how medical breakthroughs occur. Instead of relying solely on laboratories and clinical trials, future research may increasingly unfold inside computational environments capable of simulating living systems.

For patients, the implications could be profound. Faster discovery cycles mean quicker access to treatments, reduced costs, and potentially higher success rates. For scientists, it signals a shift toward interdisciplinary expertise, blending biology, computer science, and data engineering.

CellType’s journey is only beginning, but its ambitions illustrate the scale of transformation underway. If the company succeeds in building a truly agentic drug discovery platform, it may not only accelerate pharmaceutical innovation but also redefine the very nature of scientific research in the twenty-first century.