Like most people, AI has been shaping how I work over the past two years, but more recently it has started to feel different. I am beginning to rethink entire workflows, not just individual tasks.
Even how I start my day has changed. I built what I think of as a “chief of staff” agent that brings together my calendar, key metrics, customer context, and recent communications into a single briefing—turning what used to take close to an hour into just a few minutes with a clear view of what matters most.
I am seeing the same shift across my teams. They are not just using AI to assist with tasks—they are redesigning how work gets done, with greater focus on higher-value work that requires judgment and creativity.
This is why AI agents are quickly becoming a core building block for how organizations operate. The question is no longer whether they can help, but where they can be trusted to act today—and how we unlock more complex scenarios.
We partnered with MIT Technology Review Insights on new research, the Agent Confidence Index, which explores exactly that. Based on a survey of 300 global technology experts, the report ranks 101 tasks across data, AI, and cloud workflows based on respondents’ confidence in agents acting on their behalf. It provides a clear view of how the agentic frontier is unfolding across organizations around the world.
Data workflows are a breakthrough domain for agent adoption
The results point to a clear pattern. Confidence is highest when tasks are well-defined, measurable, and grounded in structured data. It is one of the reasons data workflows have emerged as the leading domain for agent adoption, including areas such as data quality monitoring, visualization anomaly detection, real-time data stream monitoring, and data profiling.
Teams are actively delegating these tasks to agents, observing the outcomes, and refining how those agents operate over time. Confidence in these tasks has now moved beyond potential alone. It is built through experience, iteration, and measurable results.
These workflows share a set of characteristics that make them well-suited for agent execution. They are well-scoped, high-volume, and operate on structured data. They also produce outputs that are observable and, in many cases, reversible. This makes it easier for teams to validate results, iterate quickly, and build trust in how agents perform.
For data teams, this is the ideal starting point. Agents reduce the manual effort that has traditionally been required to maintain data pipelines, monitor systems, and generate insights. There has always been a significant amount of toil in data analysis, and agents are helping remove much of that friction. Teams can move faster, respond earlier, and focus on higher-value work.
We are seeing this play out in practical ways. Even simple workflows like preparing for the day, reviewing metrics, or getting ready for customer engagements can now be reimagined with agents that bring together calendar, communications, customer context, and business performance. What used to take meaningful time and effort can now be consolidated into a few minutes with a more complete and consistent view.
These early wins do more than drive efficiency. They create a feedback loop. When teams can see how agents perform, measure outputs, and refine prompts, inputs, and context, confidence grows quickly. That is what enables organizations to expand from well-understood data scenarios into broader, more complex use cases.
As complexity increases, confidence shifts
As organizations push beyond these initial scenarios into more complex asks, the picture changes. Confidence drops as tasks become more complex, more interconnected, and more dependent on business judgment. That gap tells us something important.
The next phase of agentic AI is not just about improving models. It is about building the foundation around the models—the right context to train the agents, along with sophisticated observability to monitor their actions—enabling organizations to increasingly trust agents to handle new areas of the business.
Lower-confidence tasks in the survey include areas like cross-cloud data synchronization, database migration planning, feature engineering automation, and legacy system modernization. These are not just technical problems. They are coordination problems that span systems, teams, and business priorities.
What makes these tasks harder is not just their complexity. These often require coordinating long, multi-step workflows, such as database migration and planning. Many of these scenarios do not have a single correct answer, even for a human expert. They require understanding business priorities, interpreting incomplete information, and making decisions based on nuance.
Even a seemingly simple request can depend on assumptions that are obvious to a human but invisible to an agent. If you ask for the top customers by revenue, you still need to know which definition of customer to use, how revenue is calculated, and which time frame applies. Without that shared understanding, the result can vary significantly.
This is why the same pattern shows up repeatedly in practice. When an agent produces an incomplete or incorrect answer, it is often not because the model lacks capability. Instead, it is because the agent does not yet have the business context required to answer the question correctly.
At scale, this creates a real challenge. If each agent has to relearn how the business operates, where data resides, and what rules apply, organizations cannot effectively scale their use of AI. The bottleneck is no longer model capability. It is the lack of shared, consistent data context.
Building trusted agents on shared data context
To move beyond isolated use cases, agents need a common understanding of the business: the definitions of customers, orders, products, revenue, and the relationships between them. Historically, people have filled in this context through experience and interpretation. But as agents take on more responsibility, this gap becomes more pronounced. Without a consistent, shared view of how the business operates, agents struggle to reason accurately, coordinate across workflows, or take reliable action.
That’s the challenge we are solving with Microsoft IQ, which brings together different layers of enterprise context. Web IQ adds real-time global context from the web. Work IQ provides context around people, meetings, documents, and collaboration. Foundry IQ connects enterprise knowledge and processes. Fabric IQ brings together business and operational data into a unified semantic foundation.
This approach mirrors how organizations onboard people. When you bring in a new team member, you provide context about how the organization works, what metrics matter, how decisions are made, and what processes to follow. Agents require that same grounding through Microsoft IQ. Without it, they operate in isolation.
For data teams, Fabric IQ plays a critical role in delivering that shared understanding. It provides a consistent semantic layer across data, allowing agents to work with business entities, relationships, rules, and metrics rather than disconnected tables and schemas.
This shift is foundational. Instead of trying to infer meaning from raw data, agents can operate on a shared model that reflects how the business actually runs. That enables more accurate insights and more reliable decision support.
Just as importantly, it enables coordination. When agents operate on the same context, they can work across workflows, share understanding, and contribute to larger processes. Confidence grows when data, tools, and agents all operate on that same foundation, rather than being stitched together across disconnected systems.
Real-time data brings agents closer to action
Context also needs to be current.
Many of the highest-value data scenarios involve responding to live signals. Monitoring pipelines, detecting anomalies, supporting operations, and enabling timely decisions all depend on having up-to-date information. As organizations invest in real-time data platforms, they enable agents to operate with real-time context. Instead of analyzing what happened in the past, agents can surface insights and support decisions as events unfold.
While Fabric IQ provides the shared definitions, relationships, and context that ground reasoning, Real-Time Intelligence (RTI) is Microsoft’s answer to operational intelligence, bringing in live signals, events, and real-time state across the business. Together, these layers form a closed loop—where agents not only understand what is happening, but act on it as it unfolds, with decisions grounded in consistent context and executed against current conditions.
Data agents and operations agents in Fabric then reason over shared live context, make decisions based on policy, and take action in the moment.
These capabilities enable agents to begin shifting from analysis to action. They are not just helping teams understand what happened but helping them respond to what is happening now.
Trust is built through visibility and control
As agents take on more responsibility, trust becomes critical.
Most organizations are taking a pragmatic approach to building this trust. A majority, 59%, plan to keep humans actively involved in decision-making, particularly for scenarios with higher stakes.
At the same time, 53% are increasing observability by closely monitoring agent activity and tracing how decisions are made. In practice, this means agents are used to analyze data, surface insights, and recommend next steps, while humans remain responsible for final decisions when outcomes carry meaningful impact or are difficult to reverse.
At the same time, governance and observability need to be built in from the start. This includes clear policies, defined guardrails, and the ability to monitor how agents behave in practice. Observability is not just about tracking activity. It is about understanding how outputs align with intent and continuously evaluating and improving how agents perform.
This is where capabilities like Microsoft Agent 365 become increasingly important. As organizations scale their use of agents, they need a centralized way to monitor agent activity, trace decisions back to their inputs, and understand how agents are operating across workflows. These tools provide the visibility needed to safely expand agent autonomy, while ensuring that governance and accountability remain intact.
These capabilities are quickly becoming a discipline in their own right. As organizations build and operate agent systems, they are developing new practices around evaluation, validation, and control to ensure consistent and reliable outcomes at scale.
The path forward
The path to agentic AI is becoming clearer. Organizations should start where confidence is already high, focusing on structured data workflows where value is immediate and outcomes are measurable.
From there, the focus shifts to building the foundation that enables more complex scenarios. This means investing in unified data platforms, defining shared business context, and implementing governance and observability as first-class capabilities.
As agents take on more of the repetitive and high-volume work, the role of humans shifts. More of the remaining work becomes centered on judgment, strategy, and decision-making. That makes having the right data context even more critical.
The long-term opportunity extends far beyond individual automations. It is about creating systems where agents can work together, grounded in a shared understanding of the business, to support decisions and execute across workflows.
I touched on the implications for data professionals, but for those interested in what this research means for builders and traditional developers, my colleague Amanda Silver has written a companion perspective—check her blog for a deeper dive.
Next steps
- Download the full Agent Confidence Index from our partners at MIT Technology Review Insights. It’s a free, ungated deep dive into all 101 tasks, broken out by role and workflow, with the full patterns and reasoning behind where confidence is strongest and where the frontier is expanding fastest.
- Learn more about how Microsoft IQ, Fabric IQ, and Microsoft Agent 365 can help you build a unified, trusted foundation for your agents. Catch up on everything we announced at Microsoft Build for the latest on agentic capabilities across the stack.
- If you’re at the AI Engineer World’s Fair this week, stop by the Microsoft booth. We’re having open, direct conversations with technical leaders about where agents are proving real value today—focused on what’s working and what teams are still learning.
See the Agent Confidence Index
