MIT Technology Review and Microsoft have ranked 101 agent tasks based on practitioner confidence to identify trends in connected intelligence workflows [1, 2].

The findings highlight a significant gap in how developers and operators trust AI agents to handle specific technical responsibilities. As companies integrate these agents into production environments, understanding which tasks are reliable helps determine where human oversight remains critical.

The report, released June 29, identifies report generation as the task with the highest level of practitioner confidence [1, 3]. Conversely, work involving service mesh, a critical infrastructure layer for managing service-to-service communication in microservices, was ranked as the least confident task [1, 2].

This ranking provides a roadmap for where the industry is succeeding and where technical hurdles persist. By measuring confidence across a wide array of activities, the study pinpointed the specific technical frontiers where AI agents are currently struggling to meet professional standards.

"A ranking of 101 agent tasks reveals where workflows are trending and where connected intelligence is critical," Dr. Edward Tian said in Technology Review [3].

The study involved a comprehensive assessment of 101 [1, 2] distinct agent tasks. The results suggest that while generative capabilities for documentation and reporting are maturing, the ability of agents to manage complex, low-level networking and infrastructure tasks like service mesh work is not yet trusted by practitioners [1, 2].

“Report generation is the most trusted AI agent task.”

The disparity in confidence levels suggests that AI agents are currently more effective at high-level cognitive synthesis—such as summarizing data into reports—than at executing precise, high-risk infrastructure changes. This indicates that the 'technical frontier' for AI agents is shifting from content creation toward complex systems engineering, where the cost of failure is higher and the required precision is greater.