Pharmaceutical R&D is falling increasingly to artificial intelligence (AI)-guided autonomous laboratories, and the people behind them say the role of human researchers may soon be transformed but not replaced.
Fundamentally, autonomous labs comprise two things: automated machinery, such as robotic arms or bioreactors, and AI agents guiding this machinery. Developers of these systems say recent advances in AI have accelerated progress in the sophistication and adoption of these technologies for R&D throughout the biopharma industry.
Go deeper with GlobalData
Discover B2B Marketing That Performs
Combine business intelligence and editorial excellence to reach engaged professionals across 36 leading media platforms.
Improved AI brings these systems closer to full automation, such that they are able not only to independently carry out experiments but also plan them and learn from the data produced to make scientific decisions.
Autonomous labs or self-driving laboratories (SDLs) promise vastly increased experimental cycle times that would slash drug development costs and boost innovation for emerging modalities and underserved diseases. However, as in other industries, the encroachment of AI guiding automated R&D has led to concerns that many human researchers could become obsolete.
Those working on autonomous labs say the technology could instead redefine biopharma research as a collaboration between AI and researchers, and not replace scientists. They suggest the role of humans in R&D will become more conceptual, orchestrating the overarching direction of research as AI takes on greater responsibility for technical details.
What is an autonomous lab?
A 2020 Molecular Systems Biology paper established a spectrum of autonomy for biological laboratories, borrowed from a framework for vehicle autonomy created by the Society of Automotive Engineers. Starting at Level 0 – no autonomy – it reaches to Level 5, at which “humans only set goals and receive results”.
Authors of a 2024 Chemical Reviews paper took this further by dividing the five-level scale into two axes: hardware and software.
Robots, automated reactors, and AI analytics are the common basis of autonomous labs, according to Paloma Prieto, vice president of operations at the Vancouver, Canada-based Telescope Innovations. The chemical technology company works on both individual pieces of automated hardware, such as its online analysis tool DirectInject-LC, and fully integrated autonomous labs, in which Prieto says there is global interest.
In December 2025, Telescope installed South Korea’s first autonomous lab for the Korean Pharmaceutical and Biopharmaceutical Manufacturers Association (KPBMA). In January 2025, the company established a second lab for Pfizer.
AI integration is unlocking true automation
Robotic lab technologies have previously proven difficult to use and have gone underutilised, according to Nick Edwards, PhD, CEO of the Seattle, Washington-based AI developer Potato. Advances in AI changed this, he says; AI agents have made true autonomy possible and allowed researchers to interface more easily with automated lab machinery.
Potato is a seed-stage company whose AI model ‘Tater’ is “a multi-agent system we’ve built that uses tools and reasons across scientific information,” according to Edwards. “You can feed it raw data, it will analyse it, interpret it in the light of the literature; we can plan experiments with it,” he explains.
Edwards, a neuroscientist by training, says he first considered automating lab work during his PhD research when performing experiments on brain cells. Fast turnaround became key to his work, and years later when OpenAI’s generative AI large language model ChatGPT was released, Edwards looked to apply AI to accelerate biological research.
He validated Tater’s role when he gave the AI agent old data from his PhD thesis and it was able to rapidly generate figures that had taken him hours to create. Edwards then gave Tater unpublished data from a failed experiment and in minutes the AI agent identified possible issues with the experiment and offered solutions.
Both Edwards and Prieto talk about ‘closing the loop’ in biochemical experimentation, where AI agents overseeing automated machinery can plan and execute experiments, then learn from the data to make decisions for future experiments independently.
This kind of full automation run by self-sufficient AI is closer than many may think, according to Alexander Tobias, PhD, a researcher formerly with the MITRE Corporation, which operates R&D centres funded by the United States government. He says some developers may be approaching Level 5 automation, which he described with a co-author in the 2025 Royal Society Open Science paper as “a full-fledged (artificially) intelligent research scientist”.
Collaboration between humans and AI
The implications of autonomous lab adoption for the biopharmaceutical industry could be seismic. At its core, the speed of early R&D in drug development could be drastically increased, according to Hector Garcia Martin, PhD, staff scientist at the Oakland, California-based Lawrence Berkeley National Laboratory. He says beyond the physical pace of work, autonomous systems can also mitigate delays due to trial and error by predicting outcomes and directing experiments accordingly.
Greater speeds and lower costs could have knock-on effects. As the cost of data comes down, areas like rare disease research could see greater attention, says Prieto.
However, with autonomous labs approaching Level 5 automation, as Tobias claims, the role of human researchers in R&D comes under question. However, Tobias states AI is “highly complementary to humans.”
Edwards and Prieto see the role of human researchers fundamentally changing as a part of this. “We are not trying to replace chemists,” says Prieto. Scientists will dictate overarching “taste,” in Edwards’ words, setting out the broad questions and directions of research to AI, which will then define experimental specifics.
With that comes a democratisation of more intuitively accessible science, analogous to ‘vibe coding’ in post-AI software development space, according to Edwards, which Martin claims may increase human opportunity in R&D.
Still, there are some unavoidable concerns. Tobias notes that patents, including for drug discoveries, are currently awarded only to humans who make an inventive contribution. If AI-guided work becomes central in drug discovery, patent reform may be required.
Limitations on autonomous lab adoption
Edwards and Prieto say most of their business is conducted with larger pharma companies. There is growing attention from big pharma in building internal autonomous systems, says Prieto. But given the size and bureaucracy of large pharma companies, and their inherent difficulty in adapting quickly, smaller companies like Telescope can play a significant role, she adds.
While interest in autonomous labs is mostly commercial, there is also a degree of government adoption, notes Martin. In December 2025, the US Department of Energy contracted developer Ginkgo Bioworks to build an autonomous microbial phenotyping platform for $47m.
Key to the adoption of autonomous labs will be their cost. Expensive automated hardware involved may be a barrier to wider adoption, according to Tobias, and so he says it could be beneficial for labs to adopt greater AI infrastructure while continuing to rely largely on human technicians. Meanwhile, Prieto says the biggest barrier is the significant cost of training researchers to use these new systems.
She and Martin also note that the limitations of AI further curb the capabilities of autonomous labs. Good AI, no matter how sophisticated, relies on good data, and Tobias highlights that data blind spots remain on which AI is not getting trained. Principally, he says underreporting of failed experiments in the scientific literature means AI may fail to predict null outcomes in advance.
Tobias also says autonomous labs, while highly competent in routine tasks such as processing biological samples or performing chromatography, have limited use when confronting non-routine work. The nature of biopharma R&D is often shifting as new problems require innovative solutions. Tobias says the time to reprogramme labs to meet novel tasks takes too long to be practical, perhaps leaving an enduring role for human researchers.
