Sustainable Green Computing and Carbon-Aware Artificial Intelligence
June 10, 2026openalex
AI agents that can autonomously decompose tasks, use tools, browse the web, write and execute code, and self-correct are a major frontier. Current systems struggle with long-horizon planning, error recovery, and knowing when to stop or ask for help. Reliable tool use requires robust function calling, error handling, and security against prompt injection. The intersection with safety — ensuring agents act within intended boundaries while remaining useful — adds a critical constraint.