355 Deloitte’s State of AI in the Enterprise 2026 report is useful because it captures a market that has moved beyond novelty but has not yet reached operational maturity. The report is based on a survey of 3,235 business and IT leaders across 24 countries and six industries, conducted in August and September 2025, along with interviews with senior executives and AI leaders. As with any survey, it reflects how enterprises assess their own progress, but the overall pattern is clear: AI access is expanding, investment is rising, and expectations are growing, while deployment, governance, and work redesign remain uneven. If there is one theme that runs through the report, it is that enterprise AI adoption is broadening faster than enterprise AI integration. Many organizations now have approved tools, live use cases, and senior-level support. Fewer have made the organizational changes required to turn those ingredients into consistent business value. The gap between experimentation and transformation remains substantial, and Deloitte’s findings suggest that this gap is now the main dividing line between companies that are improving existing workflows and companies that are beginning to redesign processes, offerings, and operating models around AI. Access has expanded faster than everyday usage One of the clearest findings in the report is the increase in workforce access to sanctioned AI tools. Deloitte says access has grown by roughly 50% in a year, moving from under 40% of workers to around 60%. A smaller group of organizations has gone much further, with 11% reporting near-universal access to approved AI tools across their workforce. On its face, that is a meaningful shift. AI is no longer confined to technical teams or isolated pilots. It is becoming part of the standard software environment inside large companies. The more revealing point is that broader access has not produced equally broad usage. Among workers who do have access, fewer than 60% use AI in their daily workflow, and that pattern has changed little from the prior year. This matters because it shows that deployment and adoption are not the same thing. Many enterprises have already handled the policy, procurement, and tooling side of AI. They have not yet made AI a normal part of how work is performed across functions and roles. That distinction helps explain why AI benefits often appear unevenly distributed inside the same organization. Some teams or individuals may be using AI well enough to improve speed, output quality, or decision support. But unless those gains are tied to actual workflows, data permissions, training, and management expectations, they remain local rather than enterprise-wide. Deloitte later describes this as the gap between access and activation, and that is probably the right frame. At this stage, the limiting factor for many companies is not whether AI tools exist. It is whether those tools have been embedded into ordinary operating behavior. Pilot activity still exceeds production deployment The next major theme in Deloitte’s State of AI in the Enterprise report is the distance between experimentation and production. AI pilots are widespread. Production deployment is much more selective. Only 25% of respondents say their organization has moved 40% or more of its AI experiments into production so far. More than half expect to reach that level within three to six months, which suggests confidence, but the current baseline still indicates that many organizations remain early in operationalization. The report is particularly credible when it explains why this gap persists. A pilot can be run by a small team, in a tightly scoped environment, with curated data and limited dependencies. Production is fundamentally different. It requires integration with existing systems, security review, compliance checks, ongoing monitoring, maintenance, and the ability to handle variability and edge cases. Those are not secondary considerations. In many cases, they determine whether a use case becomes part of normal operations or remains a successful demonstration that never scales. Deloitte’s interview material reinforces that point. Several leaders describe a pattern in which organizations keep approving new pilots because pilots are cheaper, more visible, and easier to support politically than full deployment. Meanwhile, the slower work of integration, governance, and rollout receives less focus. That is one reason pilot fatigue appears. Without a coherent strategy and a clear path to production, more experimentation does not necessarily produce more value. It can simply produce a growing inventory of disconnected trials. A practical reading of this section is that enterprise AI maturity is becoming less about how many ideas a company can test and more about whether it can keep useful systems running inside ordinary business operations. That is a harder standard, but it is also a more meaningful one. Most companies are seeing operational gains, not strategic reinvention Deloitte’s data also draws a useful distinction between the benefits companies are realizing today and the outcomes they hope to reach later. Efficiency and productivity improvements are already common. Better decision-making and lower costs are also showing up in meaningful numbers. Revenue growth, by contrast, is still much more a future ambition than a current result. Only 20% of organizations say they are increasing revenue through AI today, while 74% hope to do so over time. That distinction matters because it separates current operating results from longer-term strategic ambitions. At the moment, the most reliable enterprise AI returns are still concentrated in internal performance: faster work, lower cost, better support for employees, and stronger analytical capacity. Those outcomes are significant. They are also different from using AI to create new products, open new markets, or change how the company generates revenue. Deloitte’s survey suggests that most organizations remain closer to optimization than reinvention. The report organizes this into three broad groups. About 37% of surveyed organizations are using AI with little or no change to the underlying process. Around 30% are redesigning important processes around AI while leaving the core business model in place. Another 34% say they are using AI to more deeply transform products, processes, or business models. That breakdown is important because it shows that “using AI” is now too broad a label to be very informative. Two enterprises can both be investing in AI and still be at very different stages of organizational change. There are other signs of momentum. Deloitte reports that 25% of leaders now say AI is having a transformative effect on their company, up from 12% a year earlier. At the same time, 84% say their organizations are increasing AI investment and 78% report greater confidence in the technology. The direction is clear enough. AI is gaining credibility inside the enterprise. But the more consequential forms of value are still concentrated in a minority of organizations that are redesigning business processes or rethinking what the business offers in the first place. Work redesign is lagging behind automation expectations If the report has a section where the market still looks particularly underdeveloped, it is the one dealing with roles, jobs, and workforce design. Deloitte finds that 36% of companies expect at least 10% of their jobs to be fully automated within a year, and 82% expect that level of automation over a three-year horizon. Those are substantial expectations. Yet 84% of organizations have not redesigned jobs around AI capabilities. The contrast is difficult to miss: many companies expect meaningful automation before they have worked through what that means for role structure, career paths, supervision, or accountability. This is a more serious issue than a simple training gap. AI affects more than isolated tasks. It changes how decisions are made, what requires human judgment, how exceptions are handled, and how junior employees acquire the practical experience that used to come from routine work. Deloitte’s interview findings are helpful here. Leaders point out that entry-level roles in areas such as data entry, reconciliation, and first-line support are often among the first targets for automation, even though those jobs have historically served as the beginning of longer career tracks. If those paths narrow, organizations will need more deliberate ways to build expertise and internal mobility. The managerial implications also matter. As AI takes on more routine execution, supervisory roles may shift toward coordination of human and machine work rather than direct oversight of large teams performing repetitive tasks. Deloitte says 53% of companies have considered flatter or pod-based structures, but only 16% have moved in that direction to a great or maximum extent. Once again, the pattern is familiar: organizations are thinking about redesign more readily than they are executing it. The talent response so far has been narrower than the challenge. The report says insufficient worker skills are viewed as the biggest barrier to integrating AI into workflows, yet the most common response is to educate the broader workforce to improve AI fluency. That makes sense as a starting point, but fluency is only part of the answer. Teaching employees how to use AI does not, by itself, redesign roles, change incentives, redefine accountability, or address the way career progression works in a more automated environment. Deloitte’s findings suggest that most enterprises are still early in that part of the transition. Worker sentiment supports that reading. According to the survey, 13% of non-technical workers are highly enthusiastic and actively seeking to use AI, 55% are open to exploring it, 21% would rather not use it but will if required, and 4% actively distrust or avoid it. That is not a hostile employee base, but neither is it a workforce that can be assumed to adopt AI simply because tools have been rolled out. Enterprise AI adoption will depend on credibility, workflow fit, management design, and trust at least as much as on access. Sovereign AI is becoming a practical enterprise concern Another important thread in Deloitte’s State of AI in the Enterprise report is sovereign AI. This topic is often discussed at the level of national strategy, public infrastructure, or government regulation. Deloitte shows that it has become a practical issue for enterprises as well, especially those operating across jurisdictions or handling sensitive data. The survey finds that 83% of companies see data residency constraints and in-country compute considerations as at least moderately important to strategic planning. Sixty-six percent are at least moderately concerned about reliance on foreign-owned AI technologies and infrastructure. More concretely, 77% say an AI solution’s country of origin now influences vendor selection, and 58% say they primarily build their AI stacks with local vendors. Those figures make clear that sovereign AI is no longer a peripheral policy concept. It is affecting procurement, infrastructure design, deployment models, and market access. The question is not only whether a model performs well. It is also where it is hosted, which laws govern the data, how retraining and auditing work across borders, and whether the company can demonstrate enough control and transparency to satisfy regulators or customers in a particular market. For multinational businesses, this adds a practical layer of complexity. Some workloads will need to remain within national or regional boundaries. Some use cases may require local hosting or smaller, more specialized model deployments. Some markets will impose requirements that are different enough to force separate architectural choices rather than a single global default. Deloitte’s underlying point is straightforward: geography, legal jurisdiction, and infrastructure control now matter enough to shape enterprise AI decisions directly. Agentic AI is moving faster than governance If there is one section of the report where planned adoption appears especially aggressive, it is the section on agentic AI. Deloitte describes these systems as AI tools that can set goals, reason through multistep tasks, use tools and APIs, and coordinate work with people or other agents. On that measure, 23% of companies say they are already using agentic AI to at least a moderate extent. Within two years, 74% expect to be doing so. Eighty-five percent say they expect to customize agents to fit the needs of their business. Those expectations matter because agentic AI changes the governance problem. Traditional AI systems often support analysis, summarize information, or generate content for people to review. Agents can do more than that. They can initiate workflows, send communications, interact with software systems, or take actions that have downstream operational consequences. Once AI moves from assisting a user to acting inside a business process, the relevant questions expand. Model quality still matters, but so do permissioning, approval thresholds, audit trails, escalation design, and human accountability. Deloitte’s concern is that governance is not developing at the same pace as adoption. Only 21% of companies report having a mature governance model for autonomous agents. At the same time, the risks leaders worry about most are closely tied to governance: data privacy and security, legal and regulatory compliance, governance capabilities and oversight, and model quality and explainability. These are not theoretical concerns. They are the operating controls required to let more autonomous systems function inside the enterprise without creating avoidable legal, security, or process risk. One of the better points in the report is that governance should be understood as an enabler of scale rather than a brake on it. If governance is treated as an after-the-fact compliance layer, it will slow deployment. If it is designed into systems early, it becomes part of what makes deployment possible. That is especially true for agentic AI, where action logging, real-time monitoring, human escalation paths, and clearly defined boundaries for autonomy are part of operational design, not optional oversight. Physical AI is growing, but on a different timetable The report also gives significant attention to physical AI, which Deloitte defines as AI systems that perceive the real world, make decisions, and drive physical actions through machines or control systems. This includes robotics, automated machinery, drones, autonomous vehicles, and related forms of operational automation. Deloitte says 58% of surveyed companies are already using physical AI in some capacity, and that figure is expected to rise to 80% within two years. Adoption is strongest in manufacturing, logistics, and defense, with Asia Pacific reporting higher current usage and higher expected growth than the Americas or EMEA. Even so, the report draws an important distinction between software-based AI and physical AI. Agentic systems may spread quickly because they can be deployed largely through software. Physical AI involves hardware, sensors, facility changes, safety controls, integration with real-world operations, and longer deployment cycles. Deloitte identifies cost as the most commonly cited barrier and emphasizes total cost of ownership rather than model cost alone. In many cases, the AI model is only one piece of the business case. Infrastructure changes, maintenance, spare parts, and implementation downtime can be much larger factors. That difference in economics helps explain where physical AI adoption is happening first. It tends to work best in controlled environments such as factories and warehouses, where conditions are more predictable and the range of possible failure modes is narrower. Open environments are harder because the safety, liability, and regulatory issues become more complicated. Deloitte’s discussion of collaborative robotics, smart monitoring, digital twins, and autonomous logistics reflects that pattern. Physical AI is clearly expanding, but it is doing so through operating environments where control, reliability, and economics can be evaluated more concretely. Strategy is ahead of talent and operating readiness The final section of the report that deserves close attention is preparedness. Deloitte finds that 42% of companies believe their strategy is highly prepared for broad AI adoption and 30% say the same for risk and governance. Technical infrastructure is at 43%, data management at 40%, and talent much lower at 20%. The exact order matters less than the overall shape of the results. Relatively few leaders believe every part of the enterprise is equally ready, and the weaker scores cluster around the capabilities that are slowest to change in practice. Deloitte also notes that perceived preparedness has improved more clearly in strategy and governance than in infrastructure, data, and talent. That is a familiar pattern. It is generally easier for an organization to establish an AI agenda or governance posture than to modernize data architecture, redesign jobs, or make new systems work reliably at scale. The report’s interviews add another layer by noting that some companies invested in infrastructure for an earlier phase of AI, only to find that the rise of large language models and newer AI use cases changed what kinds of capabilities were actually needed. That is probably the most balanced way to read Deloitte’s State of AI in the Enterprise report as a whole. Enterprises are not unclear about the importance of AI. They are investing, broadening access, and increasing ambition. But they are also running into the slower-moving parts of organizational change: systems integration, data quality, governance maturity, workforce capability, job design, and regional compliance requirements. Those are the areas most likely to determine whether AI becomes embedded in core operations or remains a collection of promising but uneven initiatives. What business leaders should take from the report Taken together, Deloitte’s findings point to a fairly clear set of conclusions. First, enterprise AI adoption should no longer be measured mainly by how many tools are available or how many pilots are underway. Those are early indicators, not end states. Second, the value being realized today is still concentrated in operational improvement more than business-model reinvention. Third, the organizations most likely to pull ahead are the ones willing to treat AI as an operating model issue rather than a software add-on. That means process redesign, governance, data architecture, and workforce structure have to move alongside model deployment. Deloitte’s recommendations at the end of the report follow directly from that reading: close the gap between access and usage, redesign work around human and AI collaboration, build governance before scaling, address sovereign AI requirements directly, modernize the data and technology foundation, and use AI in service of strategic reinvention rather than only incremental efficiency. None of those recommendations are especially surprising on their own. What the report does is show, with current survey data, where enterprises are still falling short against them. A reasonable summary of the document is that enterprise AI has moved past the early experimentation phase, but it has not yet matured into a stable set of repeatable operating practices across most organizations. The leading edge is no longer defined by whether a company has tried AI. It is defined by whether the company can deploy it responsibly, fit it into core processes, adapt roles and management around it, and absorb the governance and regional constraints that come with wider adoption. Conclusion Deloitte’s State of AI in the Enterprise 2026 report is neither celebratory nor alarmist. It describes a market in transition. Access is expanding. Investment is increasing. Confidence is higher. Agentic AI and physical AI are moving from emerging topics into active planning. Sovereign AI is becoming a practical enterprise issue rather than an abstract policy discussion. At the same time, production deployment remains selective, work redesign is lagging, and governance maturity is not yet keeping pace with autonomy. The shortest honest reading of the report is this: AI now matters across the enterprise, but execution still matters more than enthusiasm. Over the next few years, the organizations that get the most value from AI are likely to be the ones that can turn access into regular usage, pilots into production systems, and isolated tools into something that actually changes how the business operates. Further Reading When AI hoards flash: The storage playbook that protects budget and performance in turbulent times
