The 2026 Agent Confidence Index: Where 300 builders see real momentum

A few months back, I found myself sitting with my daughter’s teachers during a parent-teacher conference. They were friendly but had their concerns. She takes her time working on her assignments, they mentioned, often lost in thought. How would she handle timed tests in the future? I reassured them I wasn’t concerned. What they viewed as a drawback, I see as an invaluable skill: the ability to tackle complex problems and think them through thoroughly. In our fast-paced world, traits like patience, determination, and focus aren’t weaknesses; they form the basis of solid decision-making, a crucial skill set for the future.

The more I engage with AI, the more I’m convinced that what’s really vital for her future isn’t how quickly she responds. It’s about having the discernment to know when an answer is reliable.

Over the years at Microsoft, I’ve witnessed this dynamic unfold. Initially, I created tools for developers, and later I focused on AI as it transitioned from theoretical concepts to practical applications at scale. Now, we’re developing Microsoft IQ, exploring how a collective intelligence within an organization can serve as a significant advantage. Throughout this journey, one truth stands out: a system must not only be powerful but also trustworthy.

Trust is what turns assistance into delegation. When we trust an agent to act as we wish, within our established limits, we can hand over tasks we’d rather not spend our lives doing. This includes repetitive duties that consume our focus, the mundane chores that fill our day without contributing anything meaningful, the risky jobs that shouldn’t involve humans, and tasks far too vast for any one person or team. Agents should take on these burdens, extend our capabilities, and free our time for what truly requires a human touch.

While my daughter might not understand this yet, by the time she grows up, much of the work that prioritizes speed and repetition will likely be automated. What will matter more is precisely what led her teachers to be concerned: the patience to stick with a challenging problem, think it through, and recognize when she has reached a trustworthy conclusion. What they feared might hinder her could actually be what’s most valued in the next era.

So no, I’m not anxious about the timed test. I hope she lives in a world where software handles the drudgery, allowing people to focus on our unique skills: to think, to judge, to create, and to support one another. That’s the future I envision agents helping to create. However, my hopes aren’t proof that it will happen. This envisioned future hinges on one crucial question: can we trust agents to perform the tasks we assign them? Trust is built one action at a time. Therefore, I sought out cases where trust has been established and where it has not.

We teamed up with MIT Technology Review Insights to conduct new research sourcing insights from the leaders in the field—those not just discussing but actively creating AI. We surveyed 300 technical specialists across AI, data, and cloud sectors, spanning 12 industries and four global regions, asking them to rate their confidence in 101 key tasks. The result is the 2026 Agent Confidence Index, a transparent look into where agents are providing genuine value, enabling our community to identify effective strategies and move forward with confidence.

Learn from Where Confidence is Highest

Among the 101 tasks we assessed, the average confidence score stands at 64 out of 100, with thirty tasks scoring over 70. The highest scores are linked to tasks that are both predictable and tedious: late-night work, interruptions, and repetitive low-value tasks. Automated report generation leads with a score of 83.5. The creation of boilerplate code for new features follows at 82.5, significantly cutting down on the hours developers spend rewriting similar patterns and allowing them to focus on more challenging tasks. Activities like monitoring and renewing certificate expirations scored 81.5, quelling the last-minute rush that diverts engineers from high-stakes issues. Real-time data monitoring ranked at 80.5, while generating release notes based on commit history reached 79.5—the manual end-of-sprint reviews are now obsolete. This showcases where teams are confidently delegating tasks to agents.

This trend is consistent across various disciplines. In developer and AI workflows, it includes upkeep of API clients and identification of code; in cloud operations, it covers ticket routing and cost management; in data work, it pertains to anomaly detection. Regardless of its position in the stack, technical teams now entrust agents with these responsibilities.

What’s crucial here isn’t just the data on these tasks; it’s what this indicates about the individuals delegating them. When technical experts have enough faith in something to assign it real work, that confidence extends outward. It influences their recommendations to leadership, the solutions they develop for clients, and the culture they foster in their teams.

Even the Toughest Agent Tasks Are Gaining Ground

What stands out to me is that the tasks ranking lower on the index still score highly overall. For instance, service mesh configuration and troubleshooting score 37.5, database migration scripting sits at 46.5, and memory leak detection reaches 48.5. These are at the cutting edge, the intricate, high-risk tasks demanding attention and innovation right now.

Think about what these tasks entail. Service mesh configuration interacts with various systems simultaneously. Database migrations impose significant risks, necessitating precision across data, applications, and infrastructure layers. Memory leak detection delves deep into a system’s performance when under pressure, accounting for varying conditions with each deployment. These challenges distinguish great engineers from the exceptional ones—and even here, experts see agents making valuable contributions. They’re not taking on all the work alone, but they assist in ways that were once deemed impossible. That assurance is steadily increasing, which is significant.

We’ve been consistently launching new features to support this growth. Database migration tools in GitHub Copilot now address not only scripting but the entire application and infrastructure migration scenarios. The Azure Site Reliability Engineering (SRE) Agent incorporates years of experience managing Azure at large scale alongside advanced profiling capabilities targeted at memory analysis and performance diagnostics.

Why Human Judgment Remains Essential

In our conversations with technical professionals about adopting agents, 59% indicated that “keeping humans involved” is their chief priority—ranking ahead of enhanced observability, better governance documentation, and everything else. This shows a tremendous level of maturity. Teams forging ahead with clear vision view oversight of agents as non-negotiable, no matter how the capabilities of these systems evolve.

The distinction is quite simple. Agents excel at clearly defined, high-volume, reversible tasks: they synthesise data, simplify established workflows, and identify anomalies at speeds and scales far beyond human capability. However, when decisions become critical, context-sensitive, or difficult to reverse, a human is always needed to give confirmation. This isn’t a limitation of the technology; it underscores the framework of a dependable system.

What’s shifting, and often overlooked, is the skill required to clearly delineate that boundary: the discipline of evaluating processes throughout and applying guardrails. Achieving success means measuring output against intent and making sure behaviour aligns with your business strategy. This is a new challenge for many engineering teams, and it’s quickly becoming essential in modern software development. The good news is that tools designed to generate work can also help establish the necessary constraints. Ask GitHub Copilot to draft evaluations, and it will. Teams at the forefront are already using this to their advantage, enabling them to move ahead of the competition.

Agents Are Opening Career Doors for Engineering

Within system reliability, operational oversight, evaluations, quality assurance, and data pipeline management, over 80% of respondents foresee significant career opportunities. We believe this represents one of the most pivotal moments in software development history, not because agents replace human roles, but because they handle the tedious tasks, allowing individuals to engage in what truly defines a career: making judgment calls, crafting architectural visions, and reasoning through complexity under pressure. This insight will shape the next generation of technical leadership.

We’re experiencing this transformation at Microsoft alongside our customers. Junior developers are now using agents to navigate codebases autonomously, leading to more insightful and sophisticated discussions during mentoring sessions. Experienced engineers are covering more ground as task redundancy has been eliminated, allowing them to focus on challenging, stimulating work. Both groups are evolving into more adept versions of themselves. For me, this is the outcome I’ve always envisioned technology could achieve.

An Integrated Approach to Intelligence and Trust

The design of advanced agent systems has highlighted a vital insight: agents flourish in well-coordinated environments, performing optimally when your entire system draws from a single source of truth. Tasks that carry high confidence are those we already understand, while the significant frontier lies in the more complicated, interconnected work, where observability, governance, security, and unified intelligence must function in harmony.

Microsoft IQ brings your enterprise context into a continuous intelligence layer. Within this framework, Work IQ develops a semantic understanding of your business’s operations across email, calendars, meetings, chats, files, people, and patterns of collaboration. This deep knowledge is precisely why technical teams choose us, and it motivates my focus on understanding how people actually work, ensuring their agents cater to their needs. My colleague Kim Manis, CVP of Product for Microsoft Fabric, has discussed what this means for data specialists and the significant role of Fabric IQ.

All this forms part of the Microsoft Agent Platform, which is becoming the operating system for enterprise AI at scale. From being developed within GitHub and contextualized by Microsoft IQ, to operations in Microsoft Foundry and governance through Microsoft Agent 365, Microsoft stands poised to assist customers in integrating data, models, agents, and human judgment into a system that continuously improves and remains secure.

The frontier of transformation is being led by innovators like you.

Next Steps:

• Download The 2026 Agent Confidence Index from our partners at MIT Technology Review Insights. It’s a free and detailed exploration of all 101 tasks, categorized by role and workflow, illuminating where confidence is strongest and identifying the expanding frontiers.