Agent Plasticity is proposed to measure how efficiently AI agents improve on held-out environments

Researchers introduce Agent Plasticity as a measure of how efficiently agents improve on held-out environments. They study agents that turn experience into reusable artifacts and say the top performer isn't necessarily the most efficient learner. A separate post praises the metric and argues for benchmarks that isolate abilities beyond task performance, including learning from experience, scaling collaboration, managing context and self-training.