Reproduce “AXIS: Efficient Human-Agent-Computer Interaction with API-First LLM-Based Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junting Lu, Zhiyang Zhang, Fangkai Yang, Jue Zhang, Lu Wang, Chao Du, Qingwei Lin, Saravan Rajmohan and 2 more
AXIS, an API-first LLM-based agent framework, autonomously explores applications like Microsoft Word to generate and validate skills that prioritize API calls over multi-step UI interactions. This approach reduces task completion time by 65-70% and user cognitive workload by 38-53% compared to manual and UI agent methods, while maintaining high accuracy.
Key insights
- Prioritizing API calls over UI interactions for LLM-based agents dramatically improves efficiency, reliability, and reduces LLM operational costs.
- Autonomous skill generation and validation from user interaction trajectories and application documentation allows agents to continuously expand their capabilities in an API-first manner.
- An API-first approach to agent-computer interaction significantly lowers human cognitive load and learning effort for complex tasks, positioning agents as effective intermediaries for Human-Agent-Computer Interaction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.