Agentic AI will help drive the next stage of sample and compound management. The days when sample management focused only on small molecules are over; today’s teams need to coordinate everything in the lab, from solvents and reagents to instruments, workflows, and multiple modalities.
AI is already active across life sciences, particularly in in silico analysis. Generative AI is increasingly being embedded within lab software solutions. Agentic AI extends that trend by helping connect the digital and physical lab, bringing software, data, and hardware together. It can guide the transition from broad computational outputs to a smaller set of candidates ready for wet-lab testing.
Many lab software solutions are highly specific to scientific workflows. What makes agentic AI distinctive is that its underlying approaches can be adapted from other industries and applied to life sciences. For example, lab data often exists in structured, unstructured, semi-structured, and vendor-proprietary formats. Agentic AI can help standardize and connect these sources to streamline analysis.
Working in analytical science?
Register for a FREE Separation Science account to subscribe to the Separation Science Newsletter.
Training is key to success. For AI to better manage experimental workflows and sample preparation, it needs to learn about sample usage patterns within the existing automation infrastructure.
Say Goodbye to If-Then Failure
Agentic AI takes sample and compound management beyond registration and inventory control. It can automate and connect data movement, devices, and analysis across the lab, helping reduce bottlenecks and streamline the Design-Make-Test-Analyze (DMTA) cycle.
While lab automation has been in place for many years, traditional automation can struggle when real-world conditions deviate from the predefined rules. In principle, agentic AI systems can detect unmet criteria, suggest alternative actions, or escalate the challenge to humans so that workflows can continue more flexibly.
Once DMTA has been built with agentic AI, agents can:
- Design: Select compounds and doses, and suggest experiments
- Make: Schedule equipment slots and generate protocols and plate maps
- Test: Record anomalies, oversee runs, and track raw readings
- Analyze: Determine curves and document hits, and advise on improvements for the next cycle
Connecting AI and Data
Application programming interfaces (APIs) have long been used to connect software to software and software to hardware. Agentic AI accelerates innovation in this area by simplifying how systems can work together. The more effectively data sources are connected, the faster information can move across all steps in the workflow.
Emerging agent-to-agent (A2A) communications protocols could help support structured exchanges between organizations, such as CROs and their pharma clients. Likewise, Model Context Protocol (MCP) is an emerging open standard for connecting AI models and agents to external tools and data sources. This will include platforms such as laboratory information management systems (LIMS) and electronic lab notebooks (ELNs).
FAIR (Findable, Accessible, Interoperable, and Reusable) principles are increasingly important in the agentic AI age because AI systems depend on data that can be discovered, accessed appropriately, combined across tools, and reused in consistent ways. That data, of course, needs to be managed with schemas and governance that bring together structured, unstructured, and semi-structured data.
Access management is critical for both generative and agentic AI. AI systems should only be given access to the platforms and permissions they genuinely need, with read and write privileges carefully controlled. They should not be allowed unrestricted control over databases, especially in confidential projects. In some cases, project data may also need to remain isolated, so it is not reused outside the intended environment.
Regulation, Governance, and Compliance
Human oversight remains essential in any lab process that uses agentic AI. People still set the priorities, strategies, and goals for research. Even if agents help design or execute experiments, researchers must review both the processes and results.
AI isn’t always going to be right; it may make errors if it is not constrained by domain knowledge and operational limits. For example, it may not automatically distinguish between rare, expensive compounds and inexpensive ones, or understand efficient plate usage without appropriate guardrails. Agentic AI needs to be trained and bounded with practical input and output limits, with its recommendations validated before anything is executed in real time.
With traditional automation, compliance and documentation depend on metadata, the record of which equipment was used, and who did what, when, and where. Metadata must now include the specific details about the humans who are reviewing and approving the AI activities within the audit trail.
Specialized software has been developed to help scientists with review processes, as AI can generate higher volumes of experimental results more quickly than scientists can inspect manually. Langfuse, for example, is an open-source large language model (LLM) engineering platform for tracing, monitoring, debugging, and evaluating AI application outputs. Tools like this can help teams measure and improve reliability.
AI Standards & Human Review
AI is evolving faster than the regulatory environment, but the International Organization for Standardization (ISO) has already been working toward relevant standards development.
Relevant ISO standards include:
- ISO/TC 276 (biotechnology): Covers biotechnology standardization, including areas such as data processing, annotation, analysis, validation, comparability, and integration that are relevant to AI-enabled life sciences workflows.
- ISO/IEC 5259 (data quality for all): Offers a framework that can be used to enhance and assess data quality and reliability for machine learning and analytics
Regulatory guidance is also beginning to take shape: the US Food and Drug Administration (FDA), together with international partners, has published high-level guiding principles for the use of AI in drug development.
Agentic AI and the Future of Discovery
Both generative AI and agentic AI offer significant promise, but the burden of implementation still falls on existing lab and technology teams. In many cases, that means doing substantial foundational work now to unlock more streamlined operations later—even while those teams continue to manage their day-to-day responsibilities.
That’s why the first step in any AI implementation is review and prioritization. Existing processes need to be mapped, then ranked according to where AI can deliver the greatest value with the least disruption. Those priorities can then be assessed across the organization so implementations are targeted, coordinated, and easier to replicate across labs.
As these capabilities mature, researchers should be able to reduce manual work and spend more time on higher-value decisions. Automation and robotic systems can take on more routine tasks, allowing scientists to focus on process design, interpretation, and discovery.
The more capable the automation becomes, the more likely some labs are to extend unattended operating hours. That doesn’t mean scientists are working 24/7. Instead, they’re making higher-level decisions about the research, focusing on process design and discovery.
Agentic AI is the future for sample and compound management; the type of future it will bring depends on the scientists managing the implementation: all of us.
