Nature presents us with incredible complexity, making computer scientists and engineers welcome partners in life science research. AI engineering and lab automation systems have been embraced for drug discovery in today’s global, competitive research environment - but the best is yet to be.
In this part two of a three-part series of parallel interviews with scientists and engineers taking part in Future Labs, Automation & Technology Summit West, Ilja Kuesters, Ph.D., Associate Director, Assay Automation and Qualification at Generate:Biomedicines, shares his thoughts on Strategic Lab Automation: Unlocking Efficiency, Sustainability, and ROI Across the Research Lifecycle. If you’re new to this space, Generate:Biomedicines is a leader in combining biology, machine learning, and biological engineering. Labconscious thanks Ilja Kuseters for sharing the following insights!
Research that tightly integrates design, build, and test accelerates discovery
Q1: What is your role, company, and expertise?
My focus is on turning research assays into robust, scalable, and fully traceable automated workflows that shorten the path from scientific idea to high-quality data.
In practice, that means I partner with biology, engineering, informatics, and data science teams to develop, automate and standardize assays, design automation modules, and manage digital onboarding of new assays. The goal is to reduce variability, characterize the assay noise across its dynamic range, and make every run easier to interpret and reproduce. My team also executes high-throughput screening owning the entire assay life cycle which avoids handoff friction and increases turnaround time.
Q2: What types of technologies do you focus on to advance research and discovery?
We are building an “agent augmented” lab that uses software assistants to help design and plan experiments, schedule equipment, and capture context such as materials, automation methods, and device settings as an auditable trail. This approach improves traceability for science today and creates better training data for future models that design and analyze experiments.
We implement plate-based automation using semi-automated workflows as well as several dedicated integrated robotic “work cells” that combine liquid handlers, readers, washers, and robotic arms for high throughput data generation at scale. The emphasis is on calibrated, high-quality methods and standard building blocks so teams can design complex workflows without reinventing them.
Where commercial tools fall short, we engineer custom components such as automated switch valves for dispensers, on-deck plate balances for automated calibration, and compact imaging modules. These additions close practical gaps, cut manual steps, and enable self-checking workflows that keep quality high.
Q3: Do you see a relationship between biological research, lab automation, and sustainability goals?
Yes, when methods are standardized and executed consistently, teams repeat fewer experiments and use fewer consumables and reagents. That saves time and money while reducing waste and environmental impact across the research lifecycle.
Q4: How do you leverage lab automation in life science research?
We start by using generative models to propose protein sequences that match a therapeutic goal, then we build those sequences as real proteins at scale in the laboratory. We then measure their structure and function with a battery of assays and feed those results back into our machine-learning models so they can design the next cycle of variants in an iterative improve-and-learn loop
Lab automation is the engine that makes this loop fast and trustworthy, because it turns DNA production, protein production, and assay execution into standardized, traceable workflows. By capturing the exact materials, steps, and instrument settings automatically, we get comparable data across time and instruments, which dramatically improves both scientific confidence and model learning.
The outcome is a continuously improving “lab-in-the-loop”: designs become samples, samples become measurements, and measurements become better designs. This closes the gap between computational ideas and biological reality and shortens the time from hypothesis to improved protein candidates.
To reach this pace reliably, we are also closing hardware gaps that slow or interrupt unattended runs, because many instruments were built for walk-up, human-driven use rather than autonomous orchestration. That is why we design and integrate custom components and controls in-house — so dispensers, washers, and liquid handling systems can operate with minimal intervention and plug cleanly into our digital ecosystem.
Q5: What makes pursuing this technology worthwhile, and how do you gauge value?
The value shows up as faster time to insight, higher reproducibility, and fewer errors that would cause rework or delays. We track metrics such as turnaround time, method precision and accuracy, and the number of successful runs between interventions.
Rich, standardized metadata also strengthens machine learning driven design cycles, because models learn more from well documented experiments. Our work on generative protein design (for example, the Chroma model) highlights how tight integration of design, build, and test accelerates discovery. (See Ingraham, J.B., Baranov, M., Costello, Z. et al. Illuminating protein space with a programmable generative model. Nature 623, 1070–1078 (2023).)
Q6: What are common barriers to integrating lab automation systems?
There are typical industry challenges.
Fragmented vendor systems. Many instrument makers offer limited or closed application programming interfaces, which makes it hard to connect devices, capture the full experimental context, and adapt workflows quickly. The result is a fragile patchwork that raises integration costs and slows scientific change.
Resourcing and handoff friction. Teams frequently have too few engineers relative to scientific demand, which forces queues and leads to rushed transfers between groups. Every handoff adds interpretation risk and delays, especially when method details and troubleshooting steps are not fully captured.
Here is how we address those challenges.
Our workforce strategy lowers the barrier. We run sustained technical training, so scientists learn to operate equipment, interpret device logs, and perform first- and second line troubleshooting with confidence. By giving people the skills and playbooks they need, we make automation feel like part of doing science rather than a separate specialty.
End-to-end ownership by automation scientists. Our automation scientists steward the full assay life cycle: develop the scientific method, translate it into reliable robotic steps, and execute screens to generate decision-ready data. This continuity reduces miscommunication, speeds iteration, and makes quality and traceability non-negotiable from the start.
Q7: What skillsets help biologists thrive in automation-enabled labs?
The most effective scientists are “technology biologists” who are deep scientific experts and comfortable with lab automation and equipment. They think in terms of workflows and modules and know how their choices affect reproducibility and downstream analysis.
Method discipline matters: testing for precision and accuracy, using design-of-experiments to tune key variables, and learning the basics of liquid handling scripting. These skills reduce variability at the source and make automation more trustworthy for others.
Troubleshooting ability and good data hygiene are essential, from reading device logs to participating in assay data reviews. These habits shorten recovery when issues occur and keep quality high as methods scale.
Q8: How are artificial intelligence algorithms helping in the lab?
In discovery research, generative models can design proteins with targeted structural and functional properties and then iterate based on experimental feedback. That tight design-build-test loop can open molecular spaces that were hard to explore before.
In the lab, our philosophy is to use artificial intelligence agents to assist scientists with planning, execution, error recovery, and review rather than to replace human judgment. This approach keeps humans in control while harvesting speed and consistency gains.
Q9: What is unique about the Future Labs, Automation and Technology (FLAT) summit?
The summit emphasizes peer-driven roundtables, panel discussions, and small group working sessions where practitioners share candid experiences. Those conversations tackle real problems — such as the pros and cons of artificial intelligence, the scarcity of application programming interfaces for automation hardware, and ways to bridge the language gap between engineers and scientists.
Learning is participatory rather than passive: people compare notes, pressure-test ideas, and leave with patterns they can apply the following week. The format also creates rich networking opportunities where lessons travel beyond the stage and into hallway conversations and working groups.
Q10: Could you envision a future with an artificial intelligence scientist or lab technician supporting projects?
In the next two to five years, we expect “AI lab technicians” to assist with workflow design, selection, scheduling, error recovery, and structured review steps. That is the near-term, human-in-the-loop “augmented lab” we are building now.
Over a ten to twenty-year horizon, greater autonomy becomes realistic if vendors provide open, reliable interfaces and if safety and governance keep pace. Until then, we will blend custom engineering with digital assistants to incrementally expand what labs can do reliably.
Interested in learning more perspectives?
Read our other Q&As on Lab Automation and Sustainability with Avinash Gill, Ph.D., Senior Principal Scientific Manager, gRED, Genentech and Yousef Baioumy, Automation Engineer, Adaptive Biotechnologies™ and Jesse Mayer, Ph.D., Senior Field Application Scientist, Automata.
Consider in person at the upcoming summit.
Use the discount code: PARTNER15