$XTALPI(02228.HK)

I believe this is part of a new story:

LabHarness: Using AI to Harness Long-Cycle Autonomous Experiments

Jingtai Technology Jingtai Technology Jingtai Technology July 3, 2026 14:00 

Large models and Agent technology are accelerating the reshaping of scientific research paradigms. Examples such as Co-Scientist, which can generate scientific hypotheses, and the automated multi-agent system Robin, are deeply penetrating "dry lab" processes including literature deconstruction, hypothesis generation, code writing, and data analysis. This indicates that AI has already surpassed being a mere auxiliary tool and is now deeply involved in the research process as a digital "research partner".

However, from the perspective of industrial implementation, autonomous scientific discovery must achieve a closed loop involving wet lab experiments. Scenarios like pharmaceutical R&D, chemical synthesis, and new material creation are essentially complex systems engineering involving long-term interaction with the physical world. This means the next step for AI for Science not only requires AI to "be good at conceiving experiments" but also urgently needs to explore how to deeply integrate AI into real experimental systems, ensuring that physical devices can operate over long cycles with high stability, safety, and controllability.

Jingtai Technology, in collaboration with Fudan University, Zhejiang University, and Tongji University, among others, has proposed the LabHarness framework. In the field of general agents, "Harness" as an operational framework focuses on assisting Agents with tool invocation, state management, process tracing, and fault tolerance. In the more complex physical laboratory setting, facing diverse and heterogeneous factors such as equipment coordination, sample flow, environmental conditions, and safety boundaries, it is extremely difficult to directly translate AI-generated experimental plans into stable physical pipelines. Therefore, extending the "state control and stability guarantee" concept of Harness to the laboratory ecosystem, and building a "LabHarness" that connects digital intelligence with physical devices, is precisely the key technological path we are currently striving to overcome.

LabHarness is an operational assurance layer for real-world laboratories. It sits between scientific intelligence and real experimental systems, attempting to transform AI-generated experimental plans into experimental processes that real equipment can execute, systems can monitor, anomalies can handle, processes can trace, and key nodes can approve. If scientific intelligence is responsible for proposing experimental goals and strategies, and experimental equipment is responsible for performing specific operations, then what LabHarness aims to do is make the connection between these two more reliable.