Atomic force microscopy (AFM) is a widely used tool for characterizing material surfaces. Here, we present a framework, namely, artificially intelligent lab assistant (AILA), which enables the automation of AFM experiments using large language model-based (LLMs) agents. To evaluate the performance of AILA, we present the first benchmarking dataset, AFMBench, which consists of 100 manually curated tasks corresponding to real-world AFM experiments. These include single-step, multi-step, and mathematical reasoning-based tasks that critically analyze the ability of AILA to perform AFM experiments. Finally, we present two automated experiments using AILA: first, the calibration of the AFM, and second, the imaging of a graphene step. The results presented here highlight the capability of LLMs to guide automated high throughput experiments, accelerating the materials characterizations.