Enterprise investment in AI is booming, and Gartner calls 2026 an “inflection year” for organizations to align AI projects with strategic business objectives. The pressure is not theoretical. Executives want ROI they can point to, and technology leaders are increasingly looking to agentic AI as a way to produce measurable financial outcomes. In other words, this is not just about building chatbots, it is about deploying systems that can plan, coordinate, and execute workflows on the user’s behalf.

But there is a catch that tech leaders can feel in their bones: confidence in agents grows fast for tasks that are measurable, repeatable, and structured, then slips when the job requires business context. The MIT Technology Review report summarized here is based on a survey of 300 global technology experts and ranks 101 tasks across AI, data, and cloud workflows by respondents’ confidence in agents acting on their behalf. The headline problem is explicit in the findings: where agent readiness drops is largely due to a lack of business context being supplied to agentic systems. The more complex the task, the greater the need for reasoning and the more business context an agent requires to act safely, reliably, and securely.

To understand why this matters now, connect it to the cost pressure inside most enterprises. The source cites McKinsey saying IT infrastructure costs are projected to grow two to three times by 2030, even as budgets remain unchanged. That is a brutal incentive structure. If budgets do not rise but infrastructure and software complexity keep climbing, the only way to protect margins is to automate more of the operational burden. The report notes that over the last 18 months, tech teams are already putting agents to work, including engineers, developers, architects, and other practitioners who build, deploy, and continually improve infrastructure and applications.

Agentic AI, at its best, is supposed to do more than automate a single button click. The ultimate promise described in the report is to manage and coordinate entire workflows, with humans and agents working together. That “together” part is not a philosophical preference. Given the risks involved in automated decision-making, teams cannot delegate work without confidence that the agent can perform the task in a safe, reliable, and secure manner. Confidence is the permission slip. And according to the survey, teams are exceedingly confident about using agentic AI across a significant amount of AI, data, and cloud tasks.

So what are teams confident about? The report says confidence is surging for measurable tasks and growing in areas of complex judgment. Experts overwhelmingly believe agents help with everyday work: streamlining processes, improving performance, and reducing repetitive tasks. Confidence is highest for processes like generating reports and boilerplate code, which are often well-defined and easy to evaluate. There is also an opening where tasks involve multistep workflows and advanced reasoning to make decisions. Those are the use cases executives dream about, because they convert “AI help” into actual operational throughput.

Then comes the operational reality. Data workflows are the breakthrough domain, and confidence is highest where structure can provide a reliable foundation for decisions. The report calls out specific areas such as data quality monitoring, visualization anomaly detection, real-time data stream monitoring, and data profiling. This is also where domain experts closest to the point of data generation can supply context so agents can act and deliver trusted outcomes. Translation: when the world is structured enough, agents can reliably navigate it. When it is not, the system needs business context, and generating that context is still at an early stage of development, especially when enterprise data is difficult to wrangle and connect into the agent lifecycle at the speed and quality developers and executives need.

Human oversight is another non-negotiable lever. The report emphasizes that knowing tech teams are positioned to lead this transformation, experts expect agent confidence to accelerate as experience deepens and business environments mature. It also includes a quote from Jeremy Winter, corporate vice president and chief product officer at Microsoft Azure Platform: “As we design agents to operate within the same operational boundaries, identity systems, and governance models that teams already use, they start to behave more like the systems organizations already trust.” That framing matters for boards and CIOs because it links adoption to governance, not hype. In regulated and security-sensitive environments, trust is built through boundaries, identity, and governance models you already use, not through new paper promises.

For executives making allocation decisions, the second-order implication is simple: confidence is not evenly distributed across tasks. Your agent strategy cannot be a single rollout. It has to be a portfolio. Early wins will cluster around measurable outputs and structured data workflows. The bigger, harder returns will come when you can reliably supply business context, connect enterprise data into the agent lifecycle fast enough, and maintain human oversight as tasks get more complex. That is the reckoning hiding inside the optimism: agentic AI is moving from experiments to execution, but the companies that operationalize context and governance will capture the ROI, while the ones that treat context as an afterthought will hit diminishing returns.

This content was produced by Insights, the custom content arm of MIT Technology Review, and was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. The writing of surveys and collection of data for surveys were included in the process. AI tools that may have been used were limited to secondary production processes that passed thorough human review.