Agentic AI Brings Connectivity to Modern Lab Workflows

As techniques and technologies evolve, laboratories are generating increasingly complex insights at greater speed. However, many still rely on a mix of spreadsheets and/or disconnected systems to manage experiments, samples, and data. This can create inefficiencies and fragmented workflows, making it harder for labs to scale and collaborate effectively, despite them knowing that more integrated approaches are needed—that’s where agentic AI comes in.

AI is becoming an increasingly important part of modern laboratory environments, as it is across many areas of work and daily life. Alongside these opportunities come new considerations. AI and automation have the potential to improve efficiency and reduce manual effort, but they also raise important questions around data quality, compliance, and system reliability.

Cenevo is working to modernize lab infrastructure with software designed to connect different parts of the research process, helping laboratories manage workflows, samples, automation, and scientific data in a more unified way.

Its platform combines Labguru—which includes electronic lab notebooks (ELNs), laboratory information management systems (LIMS), inventory, registration, and automation tools—with Mosaic, an enterprise sample management system. Together, they aim to simplify operations and help labs become more connected and “AI-ready”.

At the Society for Laboratory Automation and Screening Europe conference, Technology Networks spoke with Devin Donnelly, vice president EMEA enterprise sales at Cenevo, and Kom Naidoo, senior vice president of global sales at Cenevo, who together bring a wealth of expertise on the challenges facing modern laboratories.

They discuss agentic AI and where it is already having a practical impact in laboratories—particularly in compound and biological sample management and in connecting lab workflows more efficiently, in combination with automation. The conversation also explores the challenges labs face as AI becomes more tightly embedded in scientific research.

Building the foundations of an AI-enabled Lab

According to Cenevo, the growing demand for AI-enabled research environments is pushing labs to rethink how their infrastructure and data systems are connected. Donnelly explained that Cenevo was formed in 2024, bringing together more than two decades of development across the Mosaic and Labguru platforms. The goal, he said, is to bring together technologies that support a more connected and digitally integrated laboratory ecosystem and knowing that AI needed to be integral to the future of the company.

“We were very much driven by what customers were asking for,” Donnelly said. “AI was no longer something on the horizon—it was already part of everyday life and increasingly part of scientific research.”

Cenevo’s approach combines sample management, automation, and orchestration capabilities with broader lab workflow tools, helping organizations connect experiments and manage inventory and scientific data systems more effectively. Naidoo said harmonization across laboratory infrastructure had become more important as organizations become more data-intensive and high-throughput.

“Unifying that approach enables customers to harness their data better and operate more efficiently,” he said. “It also created the foundation needed to layer in agentic AI solutions on top.”

Both interviewees emphasized that successful AI adoption depended on more than introducing new tools. Many labs are still working to connect fragmented ELN, LIMS, and sample management systems to improve workflows and reliable data exchange.

“For AI to really have power, the integration of the data ecosystem also had to happen,” Donnelly said. “It created the ability to have a workflow end-to-end.”

Moving beyond generative AI

While AI had already been present in lab software for quite a while, Donnelly explained that the industry is moving beyond basic generative capabilities towards more advanced “agentic AI” platforms.

Instead of simply helping researchers retrieve information or automate individual tasks, agentic AI can support connected workflows across the research process—from sample and inventory management through to experiment design, assay analysis, and reporting.

“Biotech leaders were telling us these cycles needed to move faster so they could iterate more quickly towards viable drug candidates,” Donnelly noted.

Agentic AI systems can help to reduce cycle times by supporting decision-making at multiple stages of lab workflows, while still allowing scientists to retain oversight at key intervention points.

The company said it was already embedding some of these capabilities into its platforms following proof-of-concept work with biotech and pharmaceutical customers.

Supporting faster drug discovery and more complex modalities

One of the major opportunities surrounding AI in the lab is its potential to accelerate drug development timelines. While pharmaceutical research has become increasingly sophisticated, the process of moving from early discovery to viable therapeutic candidates remains lengthy and resource-intensive.

Accelerate your PROTACs drug discovery and development

This eBook shows how an integrated, end-to-end analytical approach helps labs generate reliable, regulatory-ready PROTAC data at every stage of development.
View eBook

Advertisement

Many organizations are now exploring how connected AI-driven workflows could help to compress parts of that process by improving iteration speed and reducing operational bottlenecks.

Donnelly said this has become particularly important as research organizations explore increasingly diverse therapeutic modalities, from small molecules and antibody‒drug conjugates to oligonucleotides and cell and gene therapies.

“Organizations are being tasked with exploring mixed modalities, even in a non-released drug environment. Pharma, biotech, and academia are pursuing a wider range of new targets, and a more diverse set of new drug therapies is required as a result for those targets, all within the same environment and lab operation,” he said.

He explained that agentic AI systems, combined with integrated laboratory infrastructure, could help create more flexible, multi-domain research environments that more effectively support that complexity.

The conversation also touched on growing interest in drug repurposing, particularly as researchers revisit historical datasets and “abandoned” compounds that now may have therapeutic value.

“There has certainly been a heavy focus on how we repurpose existing drugs,” Donnelly said. “Having tools that can analyze what has gone before in a much quicker way allows that to become more realistic. You have so much data, and then there’s also the format it comes in, both structured and unstructured; it’s very challenging. But now there are more tools available through agentic AI to allow you to reassess and repurpose more productively.”

While there is clearly interest in this approach, both interviewees noted that data quality and structure remain key limitations. Many organizations still operate across fragmented systems with inconsistent or poorly standardized datasets, making it difficult for AI systems to extract meaningful insights reliably.

Why metadata matters

According to Naidoo, “metadata is the most important thing, with regard to getting the most out of AI tools.” It plays a central role in determining how effectively AI systems can support future research. “It’s the data within the data,” he said. “It’s understanding the scientific intent behind past experiments and future experiments that are about to be driven, and understanding how to repurpose past experiments for new opportunities.”

Tech Spotlight: High-Throughput Screening 2026

Explore reliable high-throughput screening workflows for vaccine and antibody discovery, covering assay design, validation and data analysis to improve reproducibility, data quality and candidate prioritization.

View Webinar

Advertisement

He explained that a lot of the decisions scientists make during experiments are based on experience and judgment “in the moment”, rather than being formally recorded. As a result, important context can get lost because it isn’t captured properly in the systems being used (e.g., why a particular assay condition was chosen or why a sample was excluded).

To fully support AI-enabled research, he believes metadata needs to become more standardized, harmonized, and supported by stronger ontologies across lab infrastructure. This aligns with the widely adopted FAIR data principles, which calls for machine-readable, ontology-backed metadata to make scientific data reusable across studies and platforms.

“That scientific intent is often not captured anywhere,” he said. “That decision-making could affect someone else’s analysis in the future.”

According to Donnelly and Naidoo, creating more structured and connected labs could also strengthen collaboration and reproducibility across research teams.

Balancing innovation with governance

Despite growing enthusiasm around AI in scientific research, concerns around governance, compliance, and data security remain key barriers for many organizations.

Donnelly explained that these concerns featured prominently in a recent industry survey conducted by Cenevo, particularly when it comes to laboratories introducing AI systems into highly regulated environments.

“One of the top things that came out around barriers and concerns was security and compliance,” he said. “It was still the number one barrier to the industry fully accepting agentic AI.”

For many organizations, concerns extended beyond technical implementation, though important, to broader questions surrounding intellectual property (IP) protection, regulatory compliance, and data ownership.

In Naidoo’s opinion, this hesitation is understandable, particularly when organizations consider how AI models interact with proprietary research data.

Cell and Gene Therapies: The Evolution of Advanced Biologics

This eBook takes a look at the challenges shaping advanced biologics, alongside practical strategies to improve consistency, scalability, and compliance across development and manufacturing.

View eBook

Advertisement

“The concern with large language models is simple: your data moves into infrastructure you don't own. That’s a big challenge right now,” noted Naidoo.

“As a scientist on the ground, that is daunting. You don’t want to release your IP into an agent or a system where you don’t fully know where it’s going, or how secure it is, or whether it could impact compliance from an FDA (US Food and Drug Administration) or EMA (European Medicines Agency) perspective.”

Both interviewees highlighted that AI cannot be adopted in labs without putting robust governance frameworks in place. According to Donnelly, this involves rigorous testing, compliance procedures, and security validation before new technologies are deployed into customer environments.

“Every policy and procedure we have related to releasing product has to adhere to strict ISO standards,” reiterated Donnelly. “It [agentic AI] has to go through extensive security and compliance checking and extensive testing before release.”

The conversation also highlighted the importance of training and upskilling as labs begin integrating more advanced AI systems into existing workflows. While many researchers see the potential benefits of these technologies, organizations face challenges when putting them into practice—specifically getting staff on board and ensuring they understand how to use them successfully.

“There are always going to be challenges and barriers in terms of adoption,” said Naidoo. “A key part of this is also empowerment and training, helping scientists understand how to work differently, and more importantly, the benefits it could drive.”

Building AI systems collaboratively

As AI capabilities continued to evolve, both Donnelly and Naidoo stressed that collaboration between technology providers and scientific organizations is essential.

Rather than developing solutions in isolation, Cenevo said it worked closely with customers to shape how AI agents and workflow tools are designed, tested, and implemented in live customer environments.

“Building our next-generation AI agents is not done in a siloed way—I know that's a very obvious statement, but the reality is, the industry is learning as we go,” Donnelly said. According to Donnelly, customer feedback and real-world use cases are vital to ensure AI systems deliver practical value rather than introducing avoidable complexity.

“Every time we launch an AI agent, it is on the back of high levels of user engagement,” he emphasized. “It has to be, or else we’re not going to get it right.”