KAIST light-driven flow rig makes a drug scaffold at over 1 g a day from furans
A KAIST chemistry team ran a visible-light flow reactor for 8 hours and isolated 383 mg of an N-methylpyrrole, a common drug building block, which works out to more than 1 g a day, according to its paper in JACS Au.
KAIST said on Oct 18 that the setup links a sample-preparation robot, pumps, the light reactor and analysis into one unattended line.
A team in KAIST's chemistry department has turned furans, cheap five-membered ring compounds, into N-methylpyrroles, a ring found in many medicines, inside a thin tube lit by blue light. In an 8-hour continuous run the system delivered 383 mg of product, which the authors put at more than 1 g per day, according to the paper published online on Aug 7 in JACS Au, a journal of the American Chemical Society.
KAIST presented the work on Oct 18 as an unattended synthesis platform for drug discovery, according to a report by Korean trade outlet Robot Newspaper (irobotnews) on the university's release. The release itself was not yet posted on KAIST's news site when MEKAWIRE checked, so the description of the robotic hardware below rests on that report and is not confirmed on a KAIST page.
What the chemistry does in one step
Making an N-methylpyrrole normally means building the plain pyrrole first and then attaching a methyl group to its nitrogen, the paper says. Free pyrroles are unstable, and the methylation step usually needs toxic reagents such as methyl iodide plus strong bases. The KAIST route skips both. A photocatalyst swaps the oxygen atom in the furan ring for a nitrogen carrying the methyl group, a so-called skeletal or single-atom edit, using methylammonium chloride as the nitrogen source.
The team's first-generation catalyst failed with methylamine: it oxidised the furan but produced none of the target. Two redesigned photocatalysts, PC2 and PC3, fixed that. On two substrates that had been low-yielding, yields rose from 35% and 42% with the old catalyst to 81% and 75% with PC3, the paper reports.
Why the tube matters more than the flask
Moving the reaction from a batch flask to continuous flow ran it at room temperature and cut the reaction to a residence time under 3 hours, nearly an order of magnitude faster than the group's earlier batch process with other amines, with an 85% isolated yield on the model compound. The reactor is an open-source, 3D-printed design based on a design from the Noël group: a 1.8 m tube with a 0.75 mm bore and 0.795 mL volume, irradiated at 456 nm, with more than two flow lines running in parallel to raise throughput.
According to the irobotnews report, KAIST connected a raw-material preparation robot, automated pumps, the light reactor and analysis instruments, so a researcher sets conditions and the line mixes, reacts and measures without moving samples by hand. The report says the method was applied to more than 20 furan compounds, including complex natural products.
- Continuous run
- 8 h, 383 mg isolated
- Daily productivity
- more than 1 g (paper); about 1.15 g if the 8-hour rate holds (MEKAWIRE)
- Residence time
- under 3 h (model compound), 85% yield
- Low-yield substrates
- 35% to 81%, 42% to 75% (PC1 to PC3)
- Reactor
- 1.8 m tube, 0.75 mm bore, 0.795 mL, 456 nm
Source: Byeon et al., JACS Au (2026). MEKAWIRE daily figure = 383 mg / 8 h x 24 h.
As of Oct 18, 2026
MEKAWIRE analysis
The robot here is plumbing, not a humanoid: the value is that a light reaction becomes repeatable and scalable once it runs in a tube, which is the precondition for any self-driving chemistry lab.
AI drug design produces candidates faster than chemists can make them, and photochemistry is notoriously hard to scale because light penetrates only a thin layer of liquid. A narrow tube solves the light problem and makes conditions numeric, which is what an automated line needs. MEKAWIRE's arithmetic: 383 mg in 8 hours is about 48 mg an hour, or roughly 1.15 g over 24 hours if the rate holds, consistent with the paper's "more than 1 g".
The counter-argument is scale. A gram a day suits medicinal-chemistry libraries, not manufacturing, and the paper itself is about the reaction and catalysts, with the robotic integration described mainly in KAIST's publicity. Some complex substrates also had to be run in batch because of poor solubility in flow.
It is the fourth KAIST-related robotics story on MEKAWIRE this month, after the BeetleBot soft robot (Oct 13), the Neuromeka humanoid camera work (Oct 2) and the science ministry's humanoid R&D consultation (Oct 14).
What to watch
Whether KAIST publishes the release and hardware details on its own site; whether the InnoCORE AI-CRED drug programme that funded the work pairs the line with AI molecule design; and whether Korean pharma or CDMO groups license the photocatalysts.
- JACS Au, online Aug 7, 2026, DOI 10.1021/jacsau.6c01035
- 85% isolated on the model compound in flow; residence time under 3 h
- Ministry of Science and ICT, InnoCORE AI-CRED programme
- 3D-printed, 1.8 m x 0.75 mm tube, 0.795 mL, 456 nm light
- 383 mg in 8 h continuous flow, more than 1 g per day (paper)
- KAIST Department of Chemistry (Prof. Yoonsu Park)
- 35% and 42% to 81% and 75% on low-yield substrates (PC1 to PC3)
- Robot, pumps, reactor and analysis linked into one line (KAIST via irobotnews, Oct 18; not on KAIST site yet)
Why it matters
Self-driving labs need reactions that behave the same every time. Photochemistry in flow gives numeric, reproducible conditions, which is why groups pairing synthesis robots with AI design push it.
Signals to watch
KAIST release on news.kaist.ac.kr with hardware details; InnoCORE AI-CRED follow-up papers pairing AI design with this line; any licensing of PC2 or PC3.
Analyst view
A research result, not a product. The commercial question is whether the photocatalysts and the integrated line move into Korean pharma or contract manufacturers; nothing in the paper says so.