Imagine you’re a chemist that is new to this laboratory and you’re trying to complete a specific reaction. What you’d ideally like to know is—across the history of the entire laboratory—who has done this reaction? What worked, and what didn’t work? Was one reagent better than another?
You could search that information in an ELN but it would be manual and time-consuming, and likely require a good amount of data merging.
On the other hand, you could ask Claude to compile a report on what has been done before and what worked best. With integration into domain-specific tools, Claude could even draft an email to the most successful researcher to confirm the parameters and solicit any additional guidance.
This hybrid approach of combining language models with domain-specific scientific tools is increasingly becoming the dominant architecture for scientific AI systems. In concert with agenic AI, these components give rise to the concept of the “AI co-scientist.”
An AI co-scientist is more than a complier and regurgitator of data: it is a scientific collaborator. For laboratory professionals, this shift could reshape how experiments are designed, data interpreted and research programs managed.
Democratizing expertise
Laboratory scientists routinely navigate a fragmented technology environment. Experimental work may be documented in an electronic laboratory notebook (ELN), sample information stored in a laboratory information management system (LIMS) and advanced analyses performed using separate modeling, statistical or bioinformatics platforms.
Because of this, obtaining answers often requires consulting multiple experts and software systems. Rob Brown, Head of the Scientific Office at Sapio Sciences, says AI co-scientists could eliminate many of these barriers.
In this vision, AI acts as a force multiplier rather than a replacement for scientific expertise. Instead of spending hours locating data, identifying the correct analysis workflow, or waiting for specialist support, researchers could ask questions in natural language and receive context-specific guidance generated from validated scientific workflows.
“The ideal co-scientist is one that gives every scientist the support [of an entire project team] without the bandwidth limitations of human scientists, like hours in a day,” said Brown. “It's taking every scientist, even if I'm a junior scientist in a company, and helping them operate at the level of the most experienced in the lab—the person who has been there 25 years and has all the science and history of the company in their head. The co-scientist can now be that ‘person’ for the junior scientist.”
Disparate systems
In many laboratories, integrated systems like LIMS and ELNs serve as the backbone of data operations. But, a recent survey conducted by Sapio Sciences that included 150 life sciences professionals, showed technology is outpacing current systems. For example, today's bench scientists spend too much time navigating between disconnected applications.
Experimental planning and documentation often occur in an ELN, while testing activities are managed through a LIMS. Scientists may then need to access multiple software packages to analyze results and design subsequent experiments. They may even need to call in specialists to perform critical analyses.
“The opportunity here is bringing AI agents inside the ELN," Brown said. “Rather than forcing users to learn multiple systems, AI agents could access relevant data sources, execute specialized software and return results through a familiar, single, unified workflow.
For example, project leaders often struggle to assemble a complete view of project status because information is scattered. Lab managers typically have to gather “a piece of an answer” from multiple sources before they can assess progress or make decisions. By enabling AI assistants to access and synthesize information across platforms, laboratories can provide decision-makers with a comprehensive, real-time view of project activity.
Additionally, Brown said he envisions a future in which scientists can ask natural-language questions within an ELN. For example, a researcher could ask, “What does the data of my experiment mean?” or request guidance on synthesizing a target molecule. The scientist would not need to understand which software tools are required behind the scenes. Instead, the agent is figuring out which data is needed from the ELN, while also retrieving information from other systems and running the necessary analytical applications.
This unified AI-driven workflow is ultimately centered on making laboratory knowledge accessible at the point of decision-making—in other words, democratizing knowledge and expertise.
Agentic AI
An AI-driven workflow, however, is only as good as its data. In the laboratory, trustworthy data doesn’t come from a large language AI—it comes from validated, specialized tools that have helped make the modern laboratory what it is today.
“Scientific AI must move beyond simple chatbots,” said Brown. “It has to be agentic.”
This distinction is important because scientific validity depends on established analytical methods. Sequence alignment, molecular docking, statistical testing and bioinformatics pipelines are already validated procedures. When an agentic AI brings all of this data together in one ecosystem, researchers and scientists can be assured of its genuine nature.
For example, Sapio Science’s “Elain” AI co-scientist includes data from Cadence Molecular Sciences, Cambridge Crystallographic Data Centre, Optibrium, Schrödinger, Simulations Plus, Elsevier & Scibite and MedBioInformatics.
“It has to be the algorithms the expert bioinformatician would use if they were running the experiment,” said Brown. “That can’t just be a large language model.”
The human co-scientist
Looking ahead, a progression from AI-assisted research to AI-led experimental planning is the next logical step.
Today, scientists remain responsible for designing experiments while AI provides support and recommendations. In the future, AI systems will likely plan large portions of in silico research programs independently, requesting human input only when major scientific decisions are required.
“To really move the needle, we have to flip it on its head,” said Brown.
Under this model, researchers would define scientific objectives while AI systems generate hypotheses, evaluate alternatives, execute computational workflows and determine when laboratory validation is needed.
However, Brown says this will not diminish the role of human scientists. Instead, it will concentrate their efforts on higher-value activities.
“You can increasingly focus the human on what they should be doing, which is that really high-level triage of hypotheses and decisions regarding the right path to take,” said Brown.
That perspective echoes a growing consensus among researchers that AI is most effective when augmenting scientific judgment rather than replacing it. Even as AI systems become more capable, human expertise seems to remain essential—for now.
New skills for a new era
The rise of AI co-scientists may also change how future laboratory professionals are educated and trained.
Gone, perhaps, are the days of “Advanced Organic & Inorganic Synthesis” classes. Instead, laboratory education may increasingly emphasize AI literacy, prompt design, critical evaluation of AI outputs, and the integration of computational reasoning with experimental science.
While traditional scientific fundamentals are unlikely to disappear completely, the ability to collaborate effectively with AI systems may become as important as learning a new analytical instrument or software platform. It may sound like science fiction now, but it’s already common place for the younger generations.
“It’s generational,” said Brown. “If I'm a fresh PhD graduate coming into a pharma research lab, I'm already using ChatGPT and Claude every day in my life. They are already ahead of this curve.”
Ultimately, for laboratory professionals, the emergence of AI co-scientists represents more than another software upgrade. It signals a shift toward a research environment in which expertise, analysis and institutional knowledge are available on demand in mere seconds.