From years to days: How AI agents are helping predict battery life in just days | Technology News
2 min readNew DelhiFeb 7, 2026 07:00 PM IST
The University of Michigan’s department of electrical and computer engineering claims it has developed a team of “agentic” AIs, which they say can replicate work by researchers in a lab. share data, test hypotheses and refine results.
The study, which was mentioned in Nature, was developed by a team led by assistant professor Ziyou Song and doctoral candidate Jiawei Zhang, with real-world data supplied by a US-based battery developer named Farasis Energy USA.
With information from just 50 cycles, researchers say they can predict how many charge-discharge cycles the battery can undergo before its health drops below 90%.
Researchers say these agentic AIs can help them save years of testing and massively reduce the amount of time required for battery prototyping and testing.
Compared to conventional testing methods, the team says that these AI-powered agents could help predict the life cycle of new battery designs with “just 5% of the energy and 2% of the time required by conventional testing.”
The team says it took inspiration from an approach referred to as discovery learning (learning by doing). In the study, the AI student learned from previous experiments like a human would, reviewed historical data from past designs and conducted small-scale experiments.
In the real world, this means that future systems will be able to more accurately predict the life cycle of a battery after a few days, which is comparatively less than traditional methods.
Story continues below this ad
If widely adopted, the system can potentially help battery makers advance the technology at a rapid pace. Since the team used a generalised approach, it claims similar approaches can also be applied in other areas like chemistry, material science and fields that require years of extensive feedback loops.
© IE Online Media Services Pvt Ltd
