10 Key Takeaways From MIT Technology Review's Agent Confidence Report

A new joint report from MIT Technology Review and Microsoft has laid bare a startling gap in practitioner confidence across 101 AI agent tasks—with report generation emerging as a hands-down favorite and service mesh automation languishing at the bottom.

The MIT-Microsoft Agent Confidence Index ranks 101 distinct tasks that AI agents can perform, scoring them by how confident practitioners feel in deploying them. The goal? To identify where AI agents are already trusted and where trust remains elusive—critical intelligence for enterprises racing to adopt autonomous systems.

Report generation tops the index, reflecting a high level of practitioner confidence in using AI agents to produce structured documents, summaries, and dashboards. At the other end, service mesh work—handling complex microservice traffic, security policies, and observability—scored the lowest, indicating that practitioners remain wary of ceding control over network-layer operations to autonomous agents.

The report is the result of a collaboration between MIT Technology Review's custom research arm and Microsoft's AI team, drawing on a survey of more than 500 AI practitioners across industries including finance, healthcare, and software. Respondents rated tasks on a five-point confidence scale, covering everything from customer support triage to automated DevOps workflows.

Key findings reveal a clear pattern: confidence is highest for well-defined, repetitive, and low-risk tasks. Report generation, data extraction, and email classification all cluster near the top. By contrast, complex, multi-step operations that involve real-time decision-making—such as incident response, capacity scaling, and service mesh configuration—sit at the bottom. The gap highlights a maturity curve: AI agents excel at predictable output tasks but struggle to earn trust for operational heavy lifting.

Industry observers see this as a pivotal moment. 'The confidence index is essentially a map of where enterprise AI adoption will accelerate first,' says Ritu Jyoti, group vice president for AI research at IDC. 'Organizations should focus on high-confidence tasks to build momentum before tackling the low-confidence ones.' The report also flags that practitioners with hands-on agent experience are significantly more confident than those without—a signal that training and exposure can close the trust gap.

Looking ahead, the report suggests that confidence thresholds will shift as agent safety frameworks improve and explainable AI matures. Microsoft and MIT plan to update the index annually, offering a barometer for agent trust over time. Enterprises that invest in high-confidence tasks first—and systematically address the root causes of low confidence in complex domains—stand to gain a competitive edge. The service mesh problem won't solve itself, but the data is now clear: confidence is the silent gatekeeper of AI agent scale.