The Deloitte State of AI Report: What 3,235 Leaders Tell Us About 2026
Deloitte's “State of AI in the Enterprise” report for 2026 is the most comprehensive survey of enterprise AI adoption available. The firm surveyed 3,235 business and technology leaders across 24 countries between August and September 2025. The respondents span industries including financial services, healthcare, manufacturing, technology, and government. The findings paint a picture of an enterprise landscape that is simultaneously accelerating AI adoption and failing to build the infrastructure that adoption requires.
The headline numbers are impressive. But the gaps between them tell the real story.
The Expansion: AI Access Has Broadened
The most straightforward finding in the report is that workforce access to AI tools expanded by roughly 50% in the past year. Previously, fewer than 40% of employees at surveyed organizations had access to AI tools. By mid-2025, approximately 60% did. This expansion reflects both top-down enterprise deployments (Microsoft 365 Copilot, Salesforce Einstein, ServiceNow AI agents) and bottom-up adoption (employees using ChatGPT, Claude, and other tools independently).
This is genuinely significant. For years, enterprise AI was confined to data science teams, ML engineers, and a small set of power users. The 50% expansion means AI tools are now reaching business analysts, customer service representatives, project managers, and operational staff. The surface area of AI adoption within organizations has crossed a threshold where it is no longer a specialized function. It is becoming a general-purpose capability.
The Ambition Gap: 34% Actually Reimagining Business
Here is where the numbers start to diverge. Despite widespread AI access, only 34% of organizations reported truly reimagining their business processes with AI. The rest are using AI to incrementally improve existing workflows: faster document summarization, automated first-draft generation, improved search, basic classification tasks.
These incremental improvements are valuable, but they do not represent the transformative potential that justifies the scale of investment. The 66% using AI for incremental improvements are getting incremental returns. The 34% that are reimagining business processes — rethinking how decisions are made, how workflows are structured, how resources are allocated — are capturing disproportionate value.
This aligns with McKinsey's finding that only 6% of companies capture disproportionate value from AI investments. The Deloitte data suggests the number may be somewhat larger (closer to 34% are at least attempting transformation), but the gap between adoption and transformation remains the defining challenge of enterprise AI.
The Agent Expectation: 85% Plan to Customize
The most forward-looking finding in the report is that 85% of respondents expect to customize AI agents for their organization's specific needs. This is not passive adoption of vendor-provided AI features. It is active customization: building agents that understand organizational context, follow organizational processes, and interact with organizational systems.
This number reflects the growing recognition that generic AI tools, while useful, do not capture the full value of AI for specific business contexts. A generic customer service chatbot helps. A customized agent that understands your product catalog, your return policies, your customer tiers, and your escalation procedures transforms the customer experience. The difference is customization.
But customization at this scale requires infrastructure that most organizations do not have. Building custom AI agents means managing model configurations, tool integrations, prompt libraries, evaluation benchmarks, and deployment pipelines. Doing this for one agent is a project. Doing this for the dozens of agents that 85% of organizations apparently intend to build is a platform problem.
The Automation Forecast: 36% Expect 10%+ Jobs Fully Automated Within a Year
The report's workforce findings are the most consequential for organizational planning. 36% of leaders expect more than 10% of jobs to be fully automated within one year. Looking further out, 82% expect the same threshold within three years.
“Fully automated” means roles where the entire set of tasks can be performed by AI without human involvement. This is a stronger claim than task automation (where individual tasks within a role are automated) or augmentation (where AI assists a human who retains decision authority). The leaders in this survey are predicting that complete roles will be replaced, not just individual tasks.
Whether these predictions are accurate remains to be seen. What is indisputable is that this is what enterprise leaders believe and what they are planning for. This belief shapes hiring decisions, organizational restructuring, technology investments, and workforce development strategies. Whether or not 10% of jobs are actually automated within a year, organizations are making decisions as if they will be.
This makes the governance gap even more urgent. If organizations are building AI systems intended to fully automate roles that humans currently perform, the governance requirements for those systems are substantially higher than for AI that assists humans. A copilot that helps an employee draft emails can make mistakes that the employee catches. A fully autonomous agent handling the same role has no human backstop. The governance, monitoring, and safety requirements for autonomous operation are categorically different from those for assisted operation.
The Governance Crisis: Only 21% Have Mature Frameworks
Here is the number that defines the current moment in enterprise AI: only 21% of organizations have mature governance frameworks for autonomous AI agents. Set this against the other findings: 85% plan to customize agents. 82% expect significant job automation within three years. 60% of the workforce has AI access. And only 21% have the governance infrastructure to manage any of it responsibly.
This is not a minor gap. This is a crisis in slow motion. Organizations are deploying AI agents into production, expanding access to the majority of their workforce, planning to fully automate significant portions of their operations — and 79% of them do not have mature frameworks for governing what those systems do.
The consequences are predictable. Without governance, AI agents act without audit trails. Decisions are made without documentation. Data is accessed without access controls. Actions are taken without approval workflows. And when something goes wrong — a customer receives incorrect information, a financial transaction is processed incorrectly, sensitive data is exposed — the organization cannot reconstruct what happened, why it happened, or how to prevent it from happening again.
The Job Redesign Deficit: 84% Have Not Restructured Around AI
A related finding compounds the governance concern: 84% of organizations have not redesigned jobs around AI capabilities. They are deploying AI tools into existing role structures and expecting them to be used productively without changing how roles are defined, how performance is measured, or how human-AI collaboration is structured.
This is like deploying email in the 1990s and expecting people to use it effectively while keeping all existing paper-based processes unchanged. The tool is new, but the work is designed for the old tool. The result is that AI becomes an add-on — something employees use occasionally when it is convenient, rather than a core capability that transforms how work is done.
The 34% that are truly reimagining business with AI are almost certainly the same organizations that are redesigning jobs around AI capabilities. Transformation requires changing the work, not just adding a new tool to the existing work.
The Governance Gap Is the Opportunity
Reading the Deloitte report as a whole, the central theme is a gap between ambition and infrastructure. Organizations have ambitious plans for AI agents (85% customization, 82% expect significant automation). They have expanded access dramatically (50% increase in workforce AI access). But they have not built the infrastructure those ambitions require (only 21% have governance, only 16% have redesigned jobs).
This gap is not going to close on its own. As AI adoption accelerates and agents become more autonomous, the governance requirement does not stay constant — it grows. An organization that did not need governance for a summarization chatbot absolutely needs it for an agent that processes financial transactions or handles customer personal data autonomously.
For organizations that recognize this, the governance gap is an opportunity. The enterprises that build governance infrastructure now — audit trails, human-in-the-loop controls, credential scoping, sandboxed execution, production monitoring — will be the ones that can safely deploy the custom agents they are planning, safely automate the roles they are targeting, and safely scale AI access across their workforce.
The 3,235 leaders in the Deloitte survey are telling us where enterprise AI is going. The 21% governance maturity figure tells us who will get there safely. Our platform builds the infrastructure that closes that gap. The Audit System provides immutable, cryptographically hashed logs that give the 79% without mature governance a production-ready audit trail from day one. The Workflow Builder with configurable node types and visual drag-and-drop design lets the 85% planning to customize agents do so in hours rather than months. And the SmartModelRouter with multi-provider model support ensures that as organizations scale from one agent to dozens, they are not locked to a single model vendor. This is governance, operations, and production deployment for the agentic AI future that 85% of enterprises are planning for and only 21% are prepared for.