The rise of precision delegation in enterprise AI

A recent test showed the most advanced AI agent completed 61.7% of 107 real-world commerce tasks, highlighting agentic AI’s ability to handle complex business tasks and reduce manual work. Businesses must align workflows with reliable, cost-effective systems, as efficiency is now as critical as model intelligence. Alibaba.com’s 27 years of commerce experience and real-world data provide context for AI systems, enabling task-specific model performance. The CommerceAgentBench benchmark revealed varying model strengths, such as claims settlement and listing compliance, emphasizing the need for workflow-based model selection. Accio, Alibaba’s AI-powered business agent, breaks complex requests into steps, routing them based on difficulty and requirements, mimicking a team of specialists. Cost comparisons showed Accio’s total token cost was $1.72 for 44 tasks, compared to $3.79 for Codex and $3.91 for Claude Code. As AI agents integrate with operational systems, companies must ensure reliability and define clear rules for agent authority. The next phase of enterprise AI will favor companies making decisions at the workflow level, allocating resources based on task needs and expanding agent authority as trust grows. The final deadline for the Most Innovative Companies Awards is Friday, October 2, at 11:59 p.m. PT. Apply today.