Here are two numbers from mid-2026 that should not coexist. Fifty-seven percent of enterprises have AI integrated into their operations, up significantly from last year. Twenty-three percent of business leaders think their workforce is ready for it, down six percentage points from 2025.
Those figures come from a Kyndryl global study of 1,100 senior business and technology leaders across eight countries, reported by MarketScale in July (MarketScale, July 2026). Nearly four out of five respondents agreed that “the pace of AI development will outstrip their organization’s workforce, governance, and operating models.”
The gap between those two trends is the real enterprise AI story of 2026. Companies are deploying AI faster than they are preparing the humans who have to live with it.
This isn’t a small edge case affecting a handful of laggard organizations. The Kyndryl data covers eight countries and spans manufacturing, financial services, healthcare, and retail. The pattern holds across sectors. The investment is there. The tools are there. The workforce confidence isn’t.
What makes this particularly striking is the direction of travel. Adoption is going up. Readiness is going down. These lines should move together. In a healthy transformation, as more people use AI tools, they get more competent with them, and confidence rises. That the opposite is happening — across 1,100 leaders and eight countries — suggests something about the way organizations are deploying AI is actively eroding workforce confidence rather than building it.
More AI everywhere, less confidence it’ll work
The Kyndryl data captures a dynamic that anyone working inside a large enterprise has probably felt. The C-suite announces an AI initiative. The vendor demos look great. The pilot runs. Then the tool lands on people’s desks and the friction starts.
The problem doesn’t show up in adoption metrics. It shows up in confidence metrics. When workforce readiness confidence drops even as deployment expands, the signal is clear: the people closest to the work are seeing something the deployment dashboards don’t capture.
The Kyndryl study didn’t just ask about workforce readiness. It found that the concern extends to governance and operating models too. Nearly 80% of leaders surveyed said the pace of AI development will outrun their organization’s ability to govern it properly. That’s not a workforce training problem. It’s a structural problem. If your governance framework was designed for software that doesn’t make autonomous decisions, it breaks when AI starts operating with real agency inside your workflows.
This is why cramming AI into existing processes without redesigning the processes tends to produce confusion rather than efficiency. The tool arrives before the rules for using it do.
Robert Kramer, writing in Forbes, frames this as an operating model problem rather than a technology problem. “The next enterprise transformation is not another technology implementation. It is a redesign of how the enterprise operates,” he argues (Forbes, July 2026). “No platform compensates for an organization unwilling to change how work gets done.”
Put another way: you can drop a Copilot license onto every seat in the company. You cannot drop an operating model onto every seat. The operating model has to be designed, communicated, and lived. That takes time. The AI deployment doesn’t wait.
Boards are cheering the wrong thing
Part of the readiness gap traces back to how success is defined and rewarded. A separate Q2 2026 survey, cited by Newsweek, found that one-third of business leaders identify “limited understanding of AI usage costs” as a major deployment challenge (Newsweek, July 2026). Forty-two percent report only partial visibility into what they’re actually spending on AI. Without clear cost tracking, ROI is a guess — and most guesses about AI ROI turn out to be optimistic.
The organizations that do track costs properly are five times more likely to hit their ROI targets than those that don’t. That’s not a small correlation. If you can’t see what AI is costing you, you’re mostly guessing about whether it’s working.
The Newsweek analysis points to a misalignment at the board level. AI announcements get rewarded in the market. CEOs get credit for being “AI-forward.” But the organizations that turn AI into actual financial performance are the ones that treat AI investments the same way they’d treat an acquisition or a major capital expenditure — with clearly defined financial objectives, baseline metrics, implementation milestones, and post-deployment evaluations.
Most companies aren’t doing that yet. They’re celebrating the launch and looking the other way when the usage data comes in. The Newsweek piece puts it bluntly: “Boards should require every AI initiative to answer the same questions they would ask of any major capital investment: What measurable outcome will this deliver? How will success be tracked? When will we know whether it deserves additional funding?” The fact that these questions are notable enough to call out in a major publication suggests they’re not yet standard practice.
What the readiness gap actually looks like on the ground
Workforce readiness isn’t about whether people can use ChatGPT. It’s about whether they trust the output enough to act on it, whether they know when to override it, and whether the workflows around the AI tool make the work better instead of just different.
Kramer’s operating model framework breaks this down into concrete pieces. AI needs clear decision rights. Someone has to define what the AI is authorized to do, what requires human review, and what gets escalated. Governance isn’t a compliance checkbox. It’s the architecture of who decides what when the AI’s output conflicts with someone’s intuition.
The systems piece matters too. Forbes notes that ERP, supply chain, data, AI, and security are “not separate transformation destinations. They are parts of one operating environment.” An AI decision that looks smart in isolation can create problems downstream if it doesn’t account for supply chain constraints, security requirements, or data quality issues.
The less visible layer is the human one. When an AI tool changes how a procurement manager approves purchase orders or how a customer service rep handles escalations, the people doing those jobs need to adapt their mental models. That adaptation is the readiness gap. It doesn’t close by itself just because the tool got deployed.
Consider a customer service team that gets an AI summarization tool. On paper, it’s deployed. In practice, the rep now has to decide whether the AI’s summary of a thirty-minute call is accurate enough to use as the basis for the next action. If the rep doesn’t trust the summary, they re-listen to the call — and the “time-saving” tool adds a step. If they trust it blindly, they forward bad information downstream. The readiness gap lives in that moment of judgment, and most training programs haven’t figured out how to teach it.
The same tension plays out in procurement, in legal review, in code review. The tool isn’t the problem. The lack of shared understanding about what good looks like is the problem.
Three things that actually close the gap
The data from Kyndryl, Forbes, and the Newsweek survey points to a handful of practices that distinguish companies where AI readiness is improving from companies where it’s not.
Stop measuring deployment. Start measuring adoption depth. A tool being “deployed” to 10,000 seats means very little if 8,500 of those people opened it once and never came back. Track actual usage patterns, not license distribution. Deployment is a procurement metric. Adoption depth is a transformation metric. Confuse the two and your dashboard will look great while your workforce quietly works around the tool.
Make AI spending visible before making it bigger. The 42% of companies with only partial visibility into AI costs are flying blind. Before authorizing the next round of AI investment, require a clear accounting of what the current round is costing and what it’s producing.
Redesign the workflow, not just the tool. Kramer’s point about the operating model is the hardest to execute and the most important. If you drop AI into a broken process, you get a faster broken process. The companies doing this well are mapping the entire decision chain — from data to AI to ERP to human override — before they deploy. Kramer writes that “ERP, supply chain, data, AI and security are not separate transformation destinations. They are parts of one operating environment.” An AI tool that recommends a supply chain adjustment without understanding the security implications of that adjustment is not an efficiency. It’s a risk generator wearing an efficiency costume.
Start small, measure everything, then scale what works. The organizations with strong cost visibility are five times more likely to hit ROI. That’s not a coincidence. Visibility creates accountability. Accountability creates better decisions. If you can’t tell the board what your AI initiative cost last quarter and what it produced, you shouldn’t ask the board for more money to expand it.
Define what “ready” looks like before you ask if people are ready. The Kyndryl survey found that workforce confidence is declining. But confidence follows competence. If you haven’t defined what competent use of an AI tool looks like — what output quality is acceptable, what override rate is healthy, what error rate is tolerable — you can’t tell people what “ready” means, and they can’t tell you whether they’ve arrived.
Build feedback loops between the people using the tools and the people buying them. One reason confidence drops as deployment expands is that the people closest to the work see problems that never make it back to the decision-makers. A procurement manager who figures out that the AI keeps misclassifying capital expenses as operating expenses has valuable information. Unless there’s a channel for that information to reach the team that chose the tool, the problem compounds in silence while the dashboard says “fully deployed.”
The enterprise AI story in mid-2026 is not about whether the technology works. The technology mostly works. The story is about whether organizations work — whether their boards demand outcomes instead of announcements, whether their operating models are designed for collaboration between humans and AI, and whether their workforce has been given the clarity and training to make that collaboration productive instead of exhausting.
Related reading: The Great Enterprise AI Gap: Why 99% Adoption Hasn’t Translated to Transformation in 2026, Agentic Enterprise: The Next Phase of Digital Transformation in 2026