Abstract
Computer-control agents are artificial intelligence (AI) tools that automatically perform user-defined tasks over the user interface (UI) of a computer. UI-control-based agents can be slow because they involve the agent interacting with the web page via elementary actions such as clicks and text-fills, waiting for web-page loads, and processing the web-page state and calling the AI model to predict the next elementary action(s), all of which introduce delays. This disclosure describes techniques that improve the efficiency of task performance via computer-control agents by training the agents to bundle multiple web interactions into a single, compact web-service interaction via uniform resource locator (URL) submissions or HTTP requests. Compact requests can be used independently or together with UI control actions to achieve the lowest number of per-task interactions with web services. Bottlenecks are removed and task completion times are reduced, and a substantial speed-up in agent-website interaction results.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Tran, Duc-Hieu and Hartmann, Florian, "Training Artificial Intelligence Agents to Compress Web Page Actions into URL Rewrites", Technical Disclosure Commons, (October 07, 2025)
https://www.tdcommons.org/dpubs_series/8687