Abstract
It is important for enterprises to be able to trust artificial intelligence (AI) agents to align with their domain knowledge and brand identity, and to adhere to design, business, and usability goals. This disclosure describes techniques that enable artificial intelligence (AI) application developers to steer AI agent actions by correcting mistakes via simple user interface (UI) interventions. Incorrect AI agent plans that are automatically generated using an LLM can be manually fixed. The corrected plan can be added as an example to an example store. New tasks for the LLM can leverage corrective examples stored in the example store. Customers are enabled to teach AI agents appropriate actions for the scenarios specific to their use cases by identifying agent mistakes and correcting them to illustrate desired actions.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Iyer, Anand R. and Milam, Kimberly, "Teaching AI Agents by Error Correction on Plans", Technical Disclosure Commons, (March 27, 2025)
https://www.tdcommons.org/dpubs_series/7941