Tech executives have embraced AI agents for straightforward coding tasks, rating them 82.5 out of 100 for trustworthiness, but remain deeply skeptical about deploying the same technology for complex infrastructure work like database migration or disaster recovery. A survey conducted by MIT Technology Review Insights in partnership with Microsoft found a stark divide in how business leaders view AI agent capabilities, revealing both the promise and the persistent concerns that shape enterprise AI adoption today.

What Tasks Do Leaders Actually Trust AI Agents to Handle?

The trust gap between simple and complex tasks is striking. Beyond code writing, business leaders gave high confidence scores to automated report generation at 83.5 out of 100. But when the work became more intricate, confidence collapsed. Service mesh configuration scored just 37.5, disaster recovery testing earned 43, and database migration planning received 44.5. This pattern suggests that executives see AI agents as reliable for well-defined, repetitive work but worry about their judgment when stakes are higher and outcomes are harder to predict.

The hesitation isn't irrational. Nearly half of the survey respondents cited accountability for AI decisions as their top concern, and almost the same proportion worried about inaccurate results and hallucinations, where AI systems confidently generate plausible-sounding but false information. Three out of five respondents said they keep humans in the loop to address these risks.

How to Build Enterprise AI Strategy That Actually Works

• Establish Strategic Intent: Companies need top-down transformation and a clear voice driving AI adoption across the organization, not scattered pilot projects.
• Modernize Infrastructure First: Before deploying agents, invest in the underlying technology, connectivity, security, data access, and controls that make safe deployment possible.
• Focus on High-ROI Use Cases: Identify where AI can augment human work by removing lower-level tasks, freeing employees for higher-value work rather than simply automating what they already do.
• Enable Employees Through Training: Help staff understand how to work alongside AI tools and address fears that automation will eliminate their jobs through positive experiences like hackathons.

Cisco Chief Strategy Officer Ammar Maraqa explained the strategic thinking behind successful AI deployment. "There's got to be strategic intent to adopt these technologies," he stated. "The second thing is you have to, at a base level, start investing in the infrastructure required to deploy this technology. Just as we were getting comfortable with chatbots, you're seeing the agentic wave. That takes the problems we talked about and really magnifies them."

"Modernizing the infrastructure, it's both the actual technology, connectivity, investing in security; and also getting the base-level infrastructure, data access, controls, all of these things need to be in place before deploying chatbot-like technology," Maraqa explained.

Ammar Maraqa, Chief Strategy Officer at Cisco

Why the Gap Between Code Writing and Complex Tasks Matters

The trust disparity reflects a fundamental truth about AI agent maturity: these systems excel at tasks with clear inputs, well-defined outputs, and abundant training data. Code writing fits that profile perfectly. Complex infrastructure decisions involve judgment calls, edge cases, and consequences that ripple across entire systems. That's where AI agents struggle, and where human oversight becomes non-negotiable.

Maraqa emphasized that the real opportunity lies not in replacing human decision-making but in augmenting it. "Where I've seen it deployed, it's remarkable to be able to augment humans and take a bunch of lower-level work off the plate so people can have more opportunity for higher-level work," he noted. "Instead of thinking about AI taking the work you're doing today faster or more productive, you can deploy AI to do the work you never get to".

Maraqa

For organizations planning their AI strategy, the survey findings offer a practical roadmap. Start with high-confidence tasks like code generation and report automation to build organizational muscle memory with AI agents. Use those early wins to fund infrastructure modernization and employee training. Only then should companies attempt to deploy agents in higher-stakes domains, and even then, keep humans in the loop. The executives who trust AI most aren't those who believe it can work alone, but those who understand exactly where it adds value and where human judgment remains essential.