AI knowing

As we move into the second half of the decade, AI – particularly science-aware AI – seems destined to transform biopharma R&D, super-powering scientists, shortening discovery timelines and accelerating scientific breakthroughs.

We’re not going to see AI technologies replace the work of skilled scientists and researchers, but we will witness more and more ways that technology can step up to become a viable scientific assistant, capable of enhancing decision-making, supporting experimentation, reducing the reliance on bench work and moving significant parts of the drug discovery process to in silico.

Over the past year, several developments have made advanced AI capabilities available for everyday use in biopharma R&D, integrating AI directly into the platforms scientists already use, enhancing workflows without adding complexity.

The integration of NVIDIA’s BioNeMo platform into lab informatics platforms brings access to pre-trained biomolecular language models and generative AI directly into the lab. Researchers can generate and evaluate candidate molecules, simulate docking and predict protein structures in a unified workflow.

Tools and models like AlphaFold2, MoIMIM and DiffDock are available as microservices, making advanced modelling tasks accessible without requiring standalone applications or custom infrastructure.

In parallel, partnerships with providers like AWS are expanding access to foundational AI models through services such as Amazon Bedrock, allowing the use of proprietary data to generate experiments, query data sets and engage with AI-powered assistants, without compromising on security or compliance.

Technological advancements like these are already having a real impact on how scientists work. In early-stage research, the use of science-specific AI tools is helping to reduce the time scientists need to dedicate to bench work by automating more and more of the discovery process.

Natural language AI chat is expected to become the everyday interface of choice for scientists.

In pharmaceutical and biotech environments, the integration of advanced natural language processing, machine learning and large language models into lab informatics platforms has the potential to significantly improve the efficiency and outcomes of drug discovery programmes.

These technologies enable researchers to enhance and document experimental procedures, quickly summarise and contextualise complex data, comprehensively review scientific and clinical literature, and accelerate the development of new therapeutic candidates.

To achieve meaningful impact in life sciences, several critical challenges must be addressed. AI systems must be capable of accessing and interpreting large, complex data sets and offer the specificity required to process the wide range of data types generated by modern instrumentation and experiments.

Equally important is a rethinking of how scientists engage with experiments, assays, instruments and data across discovery, manufacturing and clinical domains.

Traditional lab informatics interfaces often rely on hierarchical menu navigation, rigid form-based data entry, multiple disconnected systems or even direct SQL scripting. This places significant burdens on scientists and informatics teams alike.  Replacing these interfaces with AI-powered, natural language-based systems, whether text-based or voice-enabled, can remove these barriers.

These platforms have the potential to elevate lab software into active, collaborative assistants that support ideation and decision-making in a way that mirrors interactions with a colleague.

By bridging the gap between human creativity and machine efficiency, natural language AI interfaces offer a practical path towards a more intuitive, productive and collaborative research environment.

As artificial intelligence becomes more deeply embedded in life sciences, the limitations of general-purpose AI are becoming clearer. Scientific research presents a distinct set of requirements, from the handling of highly specialised terminology to the interpretation of multimodal experimental data.

These challenges cannot be effectively addressed by models trained solely on broad, non-scientific content.

Science-aware AI refers to systems that are purpose-built to meet the specific needs of researchers, particularly in biopharma R&D and clinical diagnostics. These systems are trained on structured biological, chemical and clinical data, enabling them to support a wide range of use cases, including experiment design, workflow automation and real-time data analysis.

The goal is to reduce manual overhead and improve decision-making without requiring users to adapt to complex interfaces or learn technical commands.

To operate effectively, these systems must access proprietary, domain-specific data, which introduces challenges around data privacy and intellectual property protection, especially in regulated industries.

Recent developments such as Amazon Bedrock help address these concerns by enabling biopharma organisations to access leading AI foundation models from AI21 Labs, Anthropic, Stability AI and others within secure and private cloud environments.

This secure framework allows organisations to build custom, science-aware applications that leverage proprietary data sets. This is an essential step in making AI a true partner in scientific discovery.

As science-aware AI becomes more deeply embedded in lab workflows, attention is turning to the next phase: agentic AI.

These systems are designed not just to respond to instructions but to take initiative, autonomously executing tasks, making decisions and adapting based on results, all in pursuit of a defined research goal.

In a typical research setting, scientists know what they are looking for but must navigate a series of disconnected steps, searching across data sets, coordinating instruments, designing protocols and reviewing outputs.

Agentic AI aims to streamline this process by acting as an active collaborator, capable of linking these steps together and executing them without manual handoffs.

Initial use cases will likely focus on early-stage drug discovery. AI agents trained on scientific data can identify potential targets, propose candidate molecules, run virtual screens and refine experimental plans.

These tools will not replace researchers but will extend their capabilities by reducing the overhead associated with repetitive or routine tasks.

For example, a researcher might request a new protein assay, and the AI agent would design the protocol, schedule equipment and generate visualisations without further input.

As these systems grow in sophistication, they will serve as collaborative co-researchers who enable faster iteration and more informed conclusions.

In effect, agentic AI will supercharge scientific research. It will enable researchers to simultaneously orchestrate a number of different research processes to perform complex tasks.

Agentic AI is not intended to replace human intelligence. Its role is to augment it, allowing researchers to prioritise strategic thinking and deeper analysis.

The integration of AI into biopharmaceutical research is no longer a hypothetical future. It is an active, ongoing transformation.

From conversational interfaces that make lab systems easier to navigate to science-aware AI that understands experimental nuance to agentic systems capable of autonomous execution, the tools available to researchers are evolving rapidly.

Each innovation addresses a practical challenge. Natural language interfaces reduce friction in daily lab work. Science-aware systems manage and interpret complex, multimodal data. Agentic AI introduces a new model for collaboration between humans and machines, freeing researchers to focus on high-value work while maintaining scientific oversight.

Together, these technologies represent more than incremental improvements. They reflect a broader shift in how research is conducted, towards platforms that are more intuitive, more adaptive and more aligned with how scientists think and work.

The goal is not to replace researchers but to help them achieve more, with greater insight, flexibility and time to focus on the science itself.

As AI capabilities continue to mature, they will become embedded not as standalone tools but as foundational elements of the modern lab.

For R&D teams across biopharma and diagnostics, embracing these technologies is quickly becoming a matter of staying competitive.

Ensuring that data, decisions and workflows are supported by systems designed for modern science is fast becoming the new standard.

We’re all wondering the same thing – how can AI truly transform scientific workflows? How far can it take us? Double the speed of discovery, tenfold breakthroughs or something beyond our imagination?

Kevin Cramer is Founder, CEO & CTO at Sapio Sciences