Liquid handlers, closed manufacturing systems and AI-driven planning are changing how life science labs work. A look at the evidence, the benefits and the risks that are discussed less often.
Automation in the life sciences is not new. Plate readers, pipetting robots and automated sequencers have been part of industrial labs for decades. What has changed is the scope. For a long time, automation took over physical tasks. Today, algorithms increasingly take over decisions as well: which experiment to run next, how to distribute samples across plates, how to schedule limited equipment.
This raises two questions at once. Is automation necessary for biotech companies to stay competitive and produce reliable data? And what does it mean for the people who currently do this work? Both questions deserve an answer that goes beyond marketing promises and beyond headlines about robots replacing scientists.
Why the pressure to automate is growing
Reproducibility
The reproducibility problem in the life sciences is well documented. In a 2016 Nature survey of 1,576 researchers, more than 70 percent reported having failed to reproduce another group’s experiments, and more than half had failed to reproduce their own [1]. An economic analysis published in PLoS Biology in 2015 estimated that more than half of preclinical research spending may be associated with irreproducible research, corresponding to approximately US$28 billion annually in the United States [2].
These numbers are often used to argue for automation. That argument needs a caveat. The same analysis identified biological reagents and reference materials, such as contaminated or misidentified cell lines, together with study design as major contributors to irreproducibility [2]. Laboratory protocols were one of several contributing categories, but not the dominant one. Automation can reduce variability in how a protocol is executed. It cannot fix a flawed experimental design or a misidentified cell line. A robot that runs a poorly designed assay with perfect precision still produces poor data, only faster.
Throughput and complexity
Modern projects in protein production, cell line development and cell-based assays involve more variables than ever: multiple constructs, several expression hosts, different media and induction conditions, and analytics that depend on each other. A decision aid developed specifically for laboratory automation in small and medium-sized laboratories points out that the growing number of samples and increasingly complex workflows can make manual processing a significant bottleneck [3]. The number of possible combinations grows quickly, and manual handling becomes both a bottleneck and a potential source of error.
Regulated manufacturing
In GMP environments, the argument is even stronger. In cell therapy manufacturing, many processes were originally adapted from academic research and still rely on open handling steps, manual processing and equipment repurposed from other applications. A 2025 study describes how this can increase quality and safety risks and manufacturing costs, and evaluated a closed, automated system across 36 manufacturing runs during process development and GMP production [4]. Closed automated systems can reduce contamination risk and operator-dependent variability, both of which matter directly for product quality and regulatory compliance.
The benefits, backed by evidence
A widely cited review by Holland and Davies from the University of Edinburgh summarises the potential benefits of automation in life science research, including improvements in reproducibility, researcher efficiency, safety and the translation of research into clinical applications [5].
Efficiency is the most visible effect. A frequently cited example is the mobile robotic chemist developed at the University of Liverpool. Guided by a Bayesian search algorithm, it performed 688 experiments in a ten-variable experimental space over eight days and identified photocatalyst mixtures six times more active than the starting formulations [6]. The robot operated for 172 of the 192 hours, with most interruptions related to recharging [7]. No human team works at that rhythm, and none should.
A less discussed benefit concerns health. Manual pipetting is a repetitive, static task. A Swedish study of 128 laboratory technicians found that hand complaints were twice as common as in a large reference group of state employees. More than 300 hours of pipetting per year was associated with a substantially increased risk of hand and shoulder ailments [8]. A 2025 study of clinical laboratory technicians reported musculoskeletal symptoms in 77.3% of participants during the previous 12 months and identified repetitive laboratory activities, including pipetting, among relevant occupational exposures [9]. Earlier studies cited in that work have reported prevalence rates in the range of 40–60% among laboratory professionals [9]. Automating high-volume liquid handling is therefore also a question of occupational health.
Finally, automation shifts where human attention goes. When plate layouts, sample tracking and scheduling no longer require manual effort, scientists can spend more time on interpretation, troubleshooting and experimental design – the parts of the work that actually require scientific expertise.
The downsides that deserve more attention
Cost and hasty investments
Automation systems are expensive, and the return on investment is not guaranteed. The authors of the decision aid for small and medium-sized laboratories warn explicitly that hasty investments can lead to slow or no cost recovery and permanently occupy valuable laboratory space [3]. They also note that operating such systems requires both bioanalytical knowledge and technical expertise, which smaller teams rarely have in abundance.
Rigidity and obsolescence
Holland and Davies identify two limitations that are less obvious: obsolescence and a potential inhibitory effect on the freedom to innovate [5]. A highly integrated system is optimised for the protocols it was designed for. When the science moves on, the system may not. The review also describes how vendors can lock laboratories into specific consumables through equipment design, and argues that research labs need more flexible, modular and affordable solutions [5].
For contract research organisations, which may work on a different target every few weeks, flexibility can therefore be more valuable than maximum throughput.
Automation bias
Perhaps the most underestimated risk is human. Research on human interaction with automated decision aids shows two related effects: complacency, where people monitor automated processes less carefully, and automation bias, where they follow a system’s recommendation even when it is wrong. A review by Parasuraman and Manzey found that both effects can occur in inexperienced and expert users, and that automation bias cannot simply be prevented by training or instructions [10].
As AI-based tools take over planning and data analysis, this becomes a practical concern for every lab. A plausible-looking plan or a clean analysis output invites trust, whether or not it is correct.
Errors at scale
A related point follows from the logic of automation itself. A manual pipetting error may affect one sample or one plate. An error in a programmed method or a faulty parameter in a planning algorithm can be repeated consistently across every run until someone notices.
Automation can therefore reduce random variation while simultaneously increasing the impact of systematic errors. Validation, plausibility checks and human review remain essential.
Loss of hands-on skills
There is also a longer-term concern that is harder to quantify. If young scientists rarely perform certain techniques manually, they may lack some of the practical intuition needed to recognise when an automated result looks wrong.
There is currently little systematic evidence in the life sciences demonstrating the extent of this effect. It should therefore be treated as an open question rather than an established consequence of automation. It is, however, a reasonable argument for maintaining sufficient manual competence in scientific training.
Will automation cost jobs?
This is the question behind much of the scepticism. The honest answer is that nobody knows exactly, and predictions should be treated with caution.
What the literature does show is that the nature of laboratory work changes as automation spreads. The authors of the decision aid for small laboratories note that the tasks of laboratory personnel have evolved along with laboratory procedures [3]. A 2025 review of autonomous laboratories discusses the potential for such systems to displace some existing scientific tasks while also creating new roles and opportunities [11]. These include method programming, data management, system validation and oversight of automated processes.
There is also an economic argument that is rarely made in public discussions. For smaller biotech companies and service providers, the greater risk to jobs may come from not automating. Clients compare timelines, costs and data quality. Companies that remain slow and error-prone may lose projects, and lost projects can affect employment more directly than any individual machine.
This is an argument, not a proven fact. But it shifts the question from whether to automate to what to automate and how.
Where the development is heading
The most ambitious vision is the self-driving laboratory. According to the 2025 review by Tobias and Wahab, the most capable systems increasingly automate large parts of the scientific cycle, from experimental design and execution to data analysis and selection of subsequent experiments [11].
In 2023, researchers presented Coscientist, a system based on a large language model that autonomously designed, planned and performed experiments, including the optimisation of palladium-catalysed cross-coupling reactions [12].
These results are impressive, but most of these demonstrations come from chemistry and materials science, where experiments can be faster, cheaper and more predictable than many biological experiments. Bioprocessing has its own challenges. A recent review in Current Opinion in Biotechnology identifies biological complexity, regulatory and safety requirements and multiscale experimentation as important differences, and concludes that hybrid systems combining AI-driven decision-making with sustained human oversight are a practical near-term path rather than full autonomy [13]. The same review highlights the challenge of transferring successful approaches from screening environments to manufacturing scale [13].
Even the Liverpool researchers behind the mobile robotic chemist acknowledged that their technology does not answer the hardest question in research: what to work on in the first place [7].
Looking ahead, and clearly as an assessment rather than a certainty, several developments appear likely. AI will increasingly take over parts of the planning layer of experimental work, including scheduling, plate design and resource allocation. Modular, affordable systems could make automation accessible to smaller labs. Interoperability between instruments and data systems will become as important as the robots themselves. And the role of the scientist will shift further towards defining questions, interpreting results and critically reviewing what machines propose.
Conclusion
Automation in biotech is neither a cure-all nor a threat to be resisted. It solves specific problems very well: repetitive handling, variability in execution, throughput limits, ergonomic strain and scheduling complexity. It does not solve poor experimental design, questionable reagents or the need for scientific judgement. And it introduces its own risks, from costly misinvestments to automation bias and systematic errors at scale.
A useful rule of thumb is to ask three questions about any task:
Is it repetitive?
Is an error expensive or hard to detect?
Does it require scientific judgement?
Tasks that are repetitive and error-prone but do not require scientific judgement are strong candidates for automation. Tasks that require judgement should remain with people, supported by automation rather than replaced by it.
At trenzyme, this principle guides how we use automation, including an AI-based algorithm we developed in-house to plan high-throughput projects.
The algorithm proposes. Our scientists review, adjust and decide.
References
[1] Baker M. 1,500 scientists lift the lid on reproducibility. Nature. 2016;533:452–454. doi:10.1038/533452a
[2] Freedman LP, Cockburn IM, Simcoe TS. The economics of reproducibility in preclinical research. PLoS Biol. 2015;13(6):e1002165. doi:10.1371/journal.pbio.1002165
[3] Rupp N, Ries R, Wienbruch R, Zuchner T. Can I benefit from laboratory automation? A decision aid for the successful introduction of laboratory automation. Anal Bioanal Chem. 2024;416:5–19. doi:10.1007/s00216-023-05038-2
[4] Kok N, Dekkers Z, Loos A, et al. Closed-system manufacturing of therapeutic NK cells using automated cell enrichment and concentration processes enables scalable, robust and cost-effective solutions. Front Bioeng Biotechnol. 2025. doi:10.3389/fbioe.2025.1586912
[5] Holland I, Davies JA. Automation in the life science research laboratory. Front Bioeng Biotechnol. 2020;8:571777. doi:10.3389/fbioe.2020.571777
[6] Burger B, Maffettone PM, Gusev VV, et al. A mobile robotic chemist. Nature. 2020;583:237–241. doi:10.1038/s41586-020-2442-2
[7] Reporting on [6] by the University of Liverpool (via Phys.org, 2020) and Chemistry World (2020).
[8] Björkstén MG, Almby B, Jansson ES. Hand and shoulder ailments among laboratory technicians using modern plunger-operated pipettes. Applied Ergonomics. 1994.
[9] Prevalence and Risk Factors of Musculoskeletal Disorders Among Clinical Laboratory Technicians. PMC12193588. 2025.
[10] Parasuraman R, Manzey DH. Complacency and bias in human use of automation: an attentional integration. Human Factors. 2010;52(3):381–410. doi:10.1177/0018720810376055
[11] Tobias AV, Wahab A. Autonomous ‘self-driving’ laboratories: a review of technology and policy implications. R Soc Open Sci. 2025;12(7):250646. doi:10.1098/rsos.250646
[12] Boiko DA, MacKnight R, Kline B, Gomes G. Autonomous chemical research with large language models. Nature. 2023;624:570–578. doi:10.1038/s41586-023-06792-0
[13] Helleckes LM, Putz S, Gupta K, Franzreb M, Garcia Martin H. Perspectives for artificial intelligence in bioprocess automation. Curr Opin Biotechnol. 2026;97:103392. doi:10.1016/j.copbio.2025.103392
