What if every experiment could inform the next? Lab-in-a-loop architectures are bringing scientific reasoning and automation together to accelerate biopharmaceutical R&D

Across the pharmaceutical industry, many research and development (R&D) labs still separate scientific thinking from automation thinking. A scientist typically reviews the data on hand, forms a hypothesis and decides what experiment should run next. From there, scientists need to a translate their experiment design into protocol that automated systems can act on, which often requires manually checking available instrument capabilities and method compatibility. It’s a fragmented workflow where each touchpoint slows down decision making.

The hidden translation layer comes at the expense of accelerated scientific discovery. Costs rise, manual handoffs introduce chances for error or rework and, perhaps most importantly, the loop between an experiment’s outcome and the next experiment rarely closes on its own. Results sit in spreadsheets or one or more laboratory information management system reports waiting for a scientist to analyse the data and decide what comes next, rather than feeding the output directly into the next round of experimentation. The good news is that these limitations can be overcome. In fact, as pharma organisations explore more advanced therapies and lean harder on artificial intelligence (AI) to help them accelerate drug development timelines, it’s becoming a strategic imperative to build a lab that operates in connected loops.

A lab-in-a-loop (LIAL) architecture links scientific insights directly to lab execution. Rather than a series of manual steps, hypothesis generation, experimental design, execution and analysis all sit inside one continuous cycle. The scientific reasoning and the automation planning happen in tandem with a human directing both and the result is an entirely different lab operating system. With a purposefully built LIAL architecture, every completed experiment automatically guides the choice and design of the next, which saves time, money and resources. Imagine a lab where the physical and digital infrastructure are designed to support this LIAL architecture. In this lab, scientists can work with a scientific agent to form a research question, such as a drug target under investigation and reason through what should be tested and how. The scientific agent then interacts with an automation agent to communicate the proposed experimental method and capability of the lab’s instrumentation and robotics, and determines whether a proposed approach is doable on any given day or what would need to change to make it work. The agent-to-agent communication will also identify instrument parameters that may require adjustments to ensure the scientific integrity of the experiment. This may even include recommending an alternative instrument that is better suited to the experiment’s specific requirements. These two agents are taking on the logistical and administrative burden of instruments and scientific execution before a sample enters the picture.

This exchange works in both directions, too. Scientists can start with the desired method and let the automation agent identify the tools needed to conduct the experiment, or start from what tools are available and ask the scientific agent if running the desired experiment is possible. Either way, the output is a feasible plan based on scientific intent and operational workability that’s ready for the scientist’s final approval.

However, it’s important to note that automation does not replace scientific judgement. In a LIAL setting, these two agents are doing the translation work that scientists have historically had to do themselves. Scientists are still in the loop confirming decisions at every meaningful step, but they don’t have to take on administrative work that can slow down discovery. Ultimately, when agents can convert scientific language into an executable method and then use the method to schedule a run, scientists are free to spend more time on high-value work that truly accelerates science.

Creating a LIAL architecture requires rethinking how the physical lab is built, even for labs who have already adopted some automation. In many instances, automated work was designed to operate in self-contained units. Tools for sample preparation were installed side by side with instruments that came next in the workflow. That approach made sense when automation was focused on speeding up one task at a time, but it also meant that every new method often required its own dedicated hardware footprint. This piecemeal approach made moving towards full automation a challenge, especially for smaller labs with fewer resources.

Autonomous mobile robots (AMRs) bring mobility to lab automation, enabling the autonomous transport of labware between benches, rooms or even floors in a multistorey environment. They can carry prepared samples to whichever instrument is available, even for precision-critical workflows that require vibration-free movement to maintain sample integrity. Instruments can even sit on open benches, giving scientists the flexibility to switch between manual use and automation as needed. From there, labs can add more power to automated systems with modular accessories that streamline analytical workflows from end to end.

The physical components in the lab only deliver value if software is in place to make the necessary connections among hardware. From scheduling systems to the underpinnings of AMRs to analysis, software enables labs to build an orchestration layer that allows instruments and workflows to operate as a connected ecosystem. It’s the key to making lab operations wholly more efficient. With the software layer, pharma R&D teams can optimise multi-variable experiments, improve data quality, increase throughput, conserve costly materials and, ultimately, see shorter development cycles. Additionally, when labs are ready to scale or organisations need to respond to changing market needs, a connected lab helps them adapt quickly. Software is essential for coordinating data, devices and processes.

Some labs have tried to develop this kind of orchestration internally, often as homegrown or individual solutions that meet the needs of their specific circumstances. However, software needs to be maintained and updated regularly, which is where these systems tend to break down. To sustain a LIAL architecture, labs need vendor-supported orchestration. Software that’s vendor agnostic helps labs work towards building a LIAL from their existing setup, even if they’ve automated certain workflows over the years.

Connecting insight, design, execution and results can create a true design-make-test-learn (DMTL) cycle within discovery environments. The value is two feedback loops running at once – one that continuously fine-tunes the method using results from recent runs (micro-loop) and another that takes the outcome of a completed experiment and feeds it back into the scientific agent for future experimental design (macro-loop). Every insight is grounded in scientific evidence and connected back to real lab workflows, so experiments aren’t held up by a scientist’s schedule. In a DMTL cycle, every insight is meaningful and actionable, and scientists can design evidence-based experiments more confidently. Pharma R&D teams can make clearer decisions about their pipeline and gain increased visibility across programmes.

Most labs today only have part of this cycle. Scientists design, make and test the experiments, but results may not automatically inform the next decision. Results sit in a report until someone has time to interpret them and start the next run. In a LIAL setting, the goal is to create a more seamless flow from hypothesis to experiment, analysis and the next decision throughout discovery. Importantly, the loop still includes scientists: they review results; apply scientific and strategic judgment; approve or refine the proposed next experiment, and provide the insight needed to keep the system aligned with the broader research objective. Automation and AI can accelerate the cycle, but the scientist remains an active participant in guiding it.

Automation in the pharma industry has historically been treated as a tactical investment that’s correlated with throughput, but that outlook is changing. Leaders, especially in the R&D space, increasingly see digital infrastructure and automation as strategic advantages, especially as AI-driven scientific discovery becomes more capable. An organisation can have the most sophisticated AI available, but still move slowly if the experiment design and execution can’t keep pace. When the lab is designed so that scientific reasoning and physical, automated execution work together, every experiment strengthens the next and next-generation tools can bring even more value. As more discovery becomes AI-assisted, labs that build a LIAL early will be the ones that can reap the benefits of intelligence in a true DMTL cycle, rather than watching data accumulate faster than the lab can act on it.

Dr Hansjoerg Haas is the senior director and general manager of laboratory automation business at Thermo Fisher Scientific. Haas has held several pivotal roles in business management, consistently focusing on automation technologies designed to accelerate scientific research. His current focus is on developing modular automation solutions that leverage digital transformation technologies for large-scale automation deployments. Haas served as president of the Society of Laboratory Automation and Screening in 2021. His contributions to the field and his commitment to innovation continue to drive advancements towards the lab of the future.