AI agents can suggest the next experiment — getting them to actually use the available hardware efficiently is the harder problem.
Researchers have published a scheduling framework designed for autonomous laboratories, specifically tested on a metal-organic framework synthesis platform. The system works in two stages: first, constraint programming generates an optimal schedule that minimizes total run time while respecting each instrument's capacity and throughput limits. Second, a status-dependency layer tracks each task in real time, allowing the system to execute those schedules robustly even when real-world hardware behaves unpredictably.
The gap between "AI picks the next experiment" and "AI runs the lab efficiently" turns out to be wide. Most autonomous lab research focuses on the selection problem — which compound to try next — while the scheduling layer gets treated as someone else's problem. A platform that can fill instrument queues without human babysitting is a meaningful step toward labs that actually run overnight without a grad student watching.
Metal-organic frameworks are a niche but active area of materials science, and autonomous synthesis platforms for them are still rare enough that scheduling wins here don't immediately generalize. The real test will be whether constraint programming stays tractable as instrument counts and experiment queues scale up.