New research from Microsoft and Georgia Tech has introduced ActionEngine, an innovative AI agent that utilizes a memory map to navigate websites efficiently with just one prompt, moving away from traditional slow, step-by-step web browsing methods. This approach boasts a 95% success rate on complex tasks and reduces operational costs by 11.8 times by pre-mapping websites and generating a single automation script, rather than making multiple expensive requests for each action. Current AI web agents often suffer from high latency and error-prone workflows due to their reliance on iterative large language model queries and screenshots. The ActionEngine’s method of separating exploration from execution aims to enhance the reliability and efficiency of AI interactions on dynamic websites.
ActionEngine achieves 95% task success rate with programmatic AI navigation
