Microsoft-MIT Report: Tech Teams Trust AI Agents Most for Data Tasks
New survey of 300 tech experts reveals where AI agents are most trusted.
A new report from MIT Technology Review Insights, sponsored by Microsoft, surveyed 300 global technology experts to rank 101 tasks across AI, data, and cloud workflows based on confidence in agents acting autonomously. The research reveals surging trust in agentic AI for measurable, structured tasks such as generating reports, boilerplate code, data quality monitoring, visualization anomaly detection, and real-time data stream monitoring. Data workflows emerged as the breakthrough domain because structured environments provide a reliable foundation for agent decisions. However, confidence drops significantly for complex tasks requiring advanced reasoning and business context, which are still early-stage capabilities due to difficulties in wrangling enterprise data. The report underscores that human oversight is a key success factor, and that agent confidence will accelerate as operational boundaries, identity systems, and governance models mature to mirror trusted enterprise systems.
The report highlights that technology teams are overwhelmingly using agents to streamline processes, improve performance, and reduce repetitive tasks. For example, confidence is highest for generating boilerplate code and reports, but drops for multistep workflows needing sophisticated judgment. Microsoft's Jeremy Winter (CVP, Azure Platform) notes that as agents operate within existing governance models, they behave more like trusted systems. The research predicts an inflection point in 2026, per Gartner, as organizations align AI projects with strategic business objectives. Experts interviewed expect confidence to grow rapidly as developers and executives improve the speed and quality of business context integration into agent life cycles. The full report is available for download and includes detailed task rankings and career insights for tech leaders.
- Survey of 300 tech experts ranks confidence in agents across 101 tasks, with highest trust in data quality monitoring and anomaly detection.
- Data workflows are the breakthrough domain for agentic AI due to structured environments; complex tasks drop off due to lack of business context.
- Human oversight remains critical; expert expects confidence to accelerate as enterprise governance models and identity systems mature by 2026.
Why It Matters
Tech leaders can now pinpoint where AI agents are safe to deploy and where they need human oversight.