By AI Trends Staff
Originally published November 12, 2020. Technically restored September 3, 2026, with later developments identified separately.
In late 2020, researchers were working to turn chemical synthesis procedures into machine-readable programs that laboratory robots could execute. The central idea behind the “chemputer” was not simply to automate pumps, heaters, and valves, but to separate a chemical recipe from any one machine—much as software can be compiled to run on compatible computers. A University of Glasgow team led by chemist Lee Cronin presented one such architecture in a 2020 Science paper. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/?utm_source=openai))
From procedure text to executable chemistry
The Glasgow system joined several components. SynthReader, a domain-specific natural-language-processing tool, tagged information in written procedures—including reagents, quantities, temperatures, durations, actions, and modifiers—and converted it into the Chemical Description Language, or XDL. A visual environment called ChemIDE allowed chemists to inspect and correct the resulting instructions without directly editing code. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/1/221626.pdf))
The edited XDL program could then be combined with a graph describing a robot’s available vessels and hardware modules. A chemical virtual machine reduced high-level steps such as adding, heating, separating, filtering, or evaporating into instructions for the target platform. The claimed hardware independence was conditional: a robot still needed compatible modules capable of performing every required operation. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/1/221626.pdf))
This workflow did not remove the chemist. The researchers explicitly warned that natural-language conversion could lose information and should not be trusted blindly. A trained user had to resolve ambiguities, verify the program, configure the hardware, load the correct materials, and ensure that the procedure could be executed safely. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/1/221626.pdf))
Cronin described the broader ambition as a “Chemical Spotify”—a repository in which synthesis procedures could be distributed as executable files rather than reconstructed from prose.
“What we’ve managed to do with the development of our ‘Chemical Spotify’ is something similar to ripping a compact disc into an MP3.” ([gla.ac.uk](https://www.gla.ac.uk/news/archiveofnews/2020/october/headline_755892_en.html?utm_source=openai))
What the Glasgow team demonstrated
The 2020 paper reported automated execution of 12 literature procedures. Demonstrations included lidocaine, produced in a reported 53% yield; the Dess–Martin periodinane oxidation reagent, produced in 52% overall yield; and a five-step synthesis of the fluorinating agent AlkylFluor, produced in 23% overall yield. The team also executed one procedure on a second robot with a different instruction set. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/1/221626.pdf))
These experiments established that selected published procedures could be translated, reviewed, compiled, and run on compatible automated equipment. They did not establish that the system could reproduce arbitrary chemistry without adaptation, match human chemists across the literature, or manufacture medicines autonomously at commercial scale.
A separate Glasgow GitHub repository contained XDL files for antiviral targets including remdesivir. That repository demonstrated the encoding of proposed synthesis instructions; it should not be confused with the 12 experimentally executed procedures reported in the Science paper, which did not list remdesivir. ([github.com](https://github.com/croningp/ChemputerAntiviralXDL?utm_source=openai))
The timing mattered. The US Food and Drug Administration approved Veklury, the brand name for remdesivir, on October 22, 2020, for specified hospitalized COVID-19 patients. The approved population and conditions were narrower than the drug’s earlier emergency-use authorization. ([fda.gov](https://www.fda.gov/news-events/press-announcements/fda-approves-first-treatment-covid-19?utm_source=openai))
SRI’s parallel approach
Researchers at SRI International were developing a separate automated synthetic-chemistry platform called SynFini—not “SynFyn,” as some early accounts rendered it. SynFini combined SynRoute for computational synthesis planning, SynJet for reaction screening and optimization, and AutoSyn for multistep flow synthesis. ([sri.com](https://www.sri.com/press/press-release/sri-international-and-exscientia-enter-collaboration-to-accelerate-drug-discovery-through-automation-and-artificial-intelligence/?utm_source=openai))
In a 2020 paper, Nathan Collins and colleagues reported using AutoSyn to synthesize ten known drugs at milligram-to-gram scale. The authors also calculated that, among FDA-approved small-molecule drugs for which their software could compute a route, 87% were predicted to be synthesizable on AutoSyn. That percentage was a computational prediction, not experimental validation across 87% of approved drugs. The researchers acknowledged that expert chemists still had to modify routes for execution, including changes related to solubility and reaction time. ([sciencedirect.com](https://www.sciencedirect.com/org/science/article/abs/pii/S1083616021023471?utm_source=openai))
SRI announced two efforts in 2020 to connect automated synthesis with AI-based molecular design. A collaboration with Exscientia proposed integrating the Centaur Chemist design system with SynFini for an oncology target. A separate collaboration with Iktos proposed combining generative molecular design with SynFini in a search for antiviral candidates, including compounds aimed at influenza and SARS-CoV-2. ([sri.com](https://www.sri.com/press/press-release/sri-international-and-exscientia-enter-collaboration-to-accelerate-drug-discovery-through-automation-and-artificial-intelligence/?utm_source=openai))
Those announcements documented planned collaborations and the intended division of labor: algorithms would propose or optimize molecules, while automated systems would plan and perform synthesis. They did not, by themselves, demonstrate that the partnerships had produced a clinical candidate, shortened a development program, or delivered a COVID-19 treatment.
What digitizing chemistry could—and could not—solve
The portable lesson from these projects was narrower and more durable than the promotional language surrounding them. Laboratory prose can be represented as executable instructions, potentially improving reproducibility and the transfer of procedures between researchers. But reproducibility depends on more than verbs such as “add,” “stir,” and “heat.” Chemical identities, quantities, material properties, equipment capabilities, process dependencies, analytical checks, error handling, and safety constraints must all be represented and validated together. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/1/221626.pdf?utm_source=openai))
Claims that digital chemistry would automatically produce safer drugs, reduce costs, democratize access to medicines, or automate pharmaceutical research from design through testing remained projections in 2020. The demonstrations showed that parts of the workflow were technically feasible; they did not establish those broader economic, clinical, or social outcomes.
Update through September 3, 2026
Later research extended the evidence for transferable chemical code. A Nature Synthesis paper published on January 11, 2024, reported the use of XDL procedures across four types of hardware in two laboratories. Three case studies involving seven reaction steps and three final compounds were repeated in two countries and on three independent robots. This was stronger evidence for cross-platform repeatability, although it still covered a limited set of reactions and machines. ([nature.com](https://www.nature.com/articles/s44160-023-00473-6?utm_source=openai))
The commercial identities also evolved after the original article. The present-day Chemify says it was founded in 2022, building on Cronin’s Glasgow research; the 2020 paper had already disclosed Cronin’s interest in an entity called Chemify Ltd. SRI announced the spinout of Synfini, Inc. on September 26, 2023, to commercialize its automated design, synthesis, and testing platform. These later developments should not be silently projected backward onto the organizations as they existed in November 2020. ([eprints.gla.ac.uk](https://eprints.gla.ac.uk/221626/1/221626.pdf))
The answer to the original question is therefore qualified. Chemical synthesis can be standardized as executable software for defined procedures and compatible equipment. Making that standard broadly reliable requires coordinated advances in protocol representation, hardware interoperability, analytical verification, safety engineering, and human review.
Primary sources
See the Glasgow team’s 2020 Science paper, its contemporary university account, the 2020 AutoSyn paper, SRI’s announcements concerning Exscientia and Iktos, and the 2024 cross-platform XDL study.
Responses